diff --git a/.claude/settings.local.json b/.claude/settings.local.json index cd13860..07d62b5 100644 --- a/.claude/settings.local.json +++ b/.claude/settings.local.json @@ -6,7 +6,13 @@ "Bash(python:*)", "Bash(source:*)", "Bash(ls:*)", - "Bash(git add:*)" + "Bash(git add:*)", + "Bash(git commit:*)", + "Bash(tree:*)", + "Bash(git push:*)", + "Bash(conda activate:*)", + "WebSearch", + "Bash(pip install:*)" ] } } diff --git a/analyze_low_accuracy_prev_day.py b/analyze_low_accuracy_prev_day.py new file mode 100644 index 0000000..a96d9da --- /dev/null +++ b/analyze_low_accuracy_prev_day.py @@ -0,0 +1,459 @@ +"""分析2019年和2026年1月准确率<40%交易日的前一天技术特征""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from data.fetcher import BinanceDataFetcher +from datetime import datetime, timedelta + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +def calculate_technical_features(df): + """计算丰富的技术特征""" + df = df.copy() + + # 基础价格特征 + df['price_change'] = df['close'] - df['open'] + df['price_change_pct'] = (df['close'] - df['open']) / df['open'] * 100 + df['high_low_range'] = df['high'] - df['low'] + df['high_low_range_pct'] = (df['high'] - df['low']) / df['low'] * 100 + df['body_ratio'] = abs(df['close'] - df['open']) / (df['high'] - df['low'] + 1e-10) + + # 移动平均线 + df['sma_5'] = ta.sma(df['close'], length=5) + df['sma_10'] = ta.sma(df['close'], length=10) + df['sma_20'] = ta.sma(df['close'], length=20) + df['sma_50'] = ta.sma(df['close'], length=50) + + # 价格相对MA位置 + df['price_vs_sma5'] = (df['close'] - df['sma_5']) / df['sma_5'] * 100 + df['price_vs_sma20'] = (df['close'] - df['sma_20']) / df['sma_20'] * 100 + + # RSI + df['rsi_14'] = ta.rsi(df['close'], length=14) + df['rsi_7'] = ta.rsi(df['close'], length=7) + + # MACD + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + + # 布林带 + bbands = ta.bbands(df['close'], length=20) + if bbands is not None and not bbands.empty: + df['bb_upper'] = bbands.iloc[:, 0] + df['bb_mid'] = bbands.iloc[:, 1] + df['bb_lower'] = bbands.iloc[:, 2] + df['bb_width'] = (df['bb_upper'] - df['bb_lower']) / df['bb_mid'] * 100 + df['bb_position'] = (df['close'] - df['bb_lower']) / (df['bb_upper'] - df['bb_lower'] + 1e-10) * 100 + + # ATR波动率 + df['atr_14'] = ta.atr(df['high'], df['low'], df['close'], length=14) + df['atr_pct'] = df['atr_14'] / df['close'] * 100 + + # 成交量特征 + df['volume_sma_20'] = ta.sma(df['volume'], length=20) + df['volume_ratio'] = df['volume'] / (df['volume_sma_20'] + 1e-10) + + # 价格动量 + df['momentum_1'] = df['close'].pct_change(1) * 100 + df['momentum_3'] = df['close'].pct_change(3) * 100 + df['momentum_7'] = df['close'].pct_change(7) * 100 + df['momentum_24'] = df['close'].pct_change(24) * 100 + + # 波动率 + df['volatility_5'] = df['close'].pct_change().rolling(5).std() * 100 + df['volatility_10'] = df['close'].pct_change().rolling(10).std() * 100 + df['volatility_20'] = df['close'].pct_change().rolling(20).std() * 100 + + # 连续涨跌天数 + df['is_up'] = (df['close'] > df['open']).astype(int) + df['consecutive_days'] = 0 + + return df + + +def analyze_period(model, preprocessor, device, period_name, start_date, end_date): + """分析单个时期的数据，返回低准确率日期""" + + print(f"\n{'='*80}") + print(f"分析 {period_name}") + print(f"{'='*80}") + + # 获取数据（需要额外获取前一天的数据） + fetcher = BinanceDataFetcher() + + # 获取包含前一天的数据 + start_dt = pd.to_datetime(start_date) - pd.Timedelta(days=2) + df = fetcher.fetch_ohlcv('BTC', '1h', start_date=start_dt.strftime('%Y-%m-%d'), + end_date=end_date, use_cache=False) + + print(f"✓ 数据: {len(df)} 行") + + # 预处理 + X, y, df_processed = preprocessor.preprocess(df, fit=False) + + print(f"✓ 有效样本: {len(X)}") + + # DLinear预测 + X_tensor = torch.FloatTensor(X).to(device) + with torch.no_grad(): + logits = model(X_tensor).cpu().numpy() + probs = 1 / (1 + np.exp(-logits)) + preds = (probs > 0.5).astype(int).flatten() + + true_labels = y + + # 构建每日结果 + timestamps = df['timestamp'].iloc[168:168+len(X)].values + timestamps_utc = pd.to_datetime(timestamps) + timestamps_et = timestamps_utc.tz_localize('UTC').tz_convert('America/New_York') + dates = timestamps_et.date + + results_df = pd.DataFrame({ + 'date': dates, + 'timestamp_et': timestamps_et, + 'correct': preds == true_labels, + }) + + # 按天统计准确率 + daily_stats = results_df.groupby('date').agg({ + 'correct': ['count', 'sum'] + }) + daily_stats.columns = ['total', 'correct'] + daily_stats['accuracy'] = daily_stats['correct'] / daily_stats['total'] * 100 + + # 找出准确率<40%的日期 + low_acc_dates = daily_stats[daily_stats['accuracy'] < 40].index.tolist() + + print(f"✓ 准确率<40%的日期: {len(low_acc_dates)}天") + for date in low_acc_dates: + acc = daily_stats.loc[date, 'accuracy'] + print(f" {date}: {acc:.1f}%") + + # 计算技术特征 + df_with_features = calculate_technical_features(df) + + # 为每天聚合技术特征（使用当天的24小时数据） + # 先确保timestamp是datetime类型 + if not pd.api.types.is_datetime64_any_dtype(df_with_features['timestamp']): + df_with_features['timestamp'] = pd.to_datetime(df_with_features['timestamp']) + + df_with_features['date'] = df_with_features['timestamp'].dt.tz_localize('UTC').tz_convert('America/New_York').dt.date + + # 返回低准确率日期和完整数据 + return low_acc_dates, df_with_features, daily_stats + + +def main(): + print("="*80) + print("分析准确率<40%交易日的前一天技术特征共同点") + print("="*80) + + # ========== 加载模型 ========== + print("\n[1/3] 加载DLinear模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + from models.dlinear.model import DLinear + checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device, weights_only=False + ) + model = DLinear(checkpoint['config']) + model.load_state_dict(checkpoint['model_state_dict']) + model = model.to(device) + model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + preprocessor = pickle.load(f) + + print("✓ DLinear模型已加载") + + # ========== 分析两个时期 ========== + print("\n[2/3] 分析2019年和2026年1月...") + + # 分析2019年1月 + low_acc_2019, df_2019, stats_2019 = analyze_period( + model, preprocessor, device, + "2019年1月", "2019-01-01", "2019-01-31" + ) + + # 分析2026年1月 + low_acc_2026, df_2026, stats_2026 = analyze_period( + model, preprocessor, device, + "2026年1月", "2026-01-01", "2026-01-31" + ) + + # ========== 提取前一天特征 ========== + print("\n[3/3] 分析前一天技术特征...") + + # 合并所有低准确率日期 + all_low_acc_dates = low_acc_2019 + low_acc_2026 + + print(f"\n总计{len(all_low_acc_dates)}个低准确率交易日") + + # 提取前一天的特征 + prev_day_features = [] + + for date in all_low_acc_dates: + # 获取前一天日期 + prev_date = date - timedelta(days=1) + + # 确定使用哪个数据集 + if date in low_acc_2019: + df = df_2019 + period = "2019年" + else: + df = df_2026 + period = "2026年" + + # 获取前一天的数据 + prev_day_data = df[df['date'] == prev_date] + + if len(prev_day_data) == 0: + continue + + # 聚合前一天的技术特征（使用24小时的平均值、最大值、最小值等） + features = { + 'date': date, + 'prev_date': prev_date, + 'period': period, + 'accuracy': stats_2019.loc[date, 'accuracy'] if date in low_acc_2019 else stats_2026.loc[date, 'accuracy'], + # 价格特征 + 'prev_close': prev_day_data['close'].iloc[-1], + 'prev_open': prev_day_data['open'].iloc[0], + 'prev_price_change_pct': prev_day_data['price_change_pct'].mean(), + 'prev_high_low_range_pct': prev_day_data['high_low_range_pct'].mean(), + 'prev_body_ratio': prev_day_data['body_ratio'].mean(), + # RSI + 'prev_rsi_14': prev_day_data['rsi_14'].iloc[-1], + 'prev_rsi_7': prev_day_data['rsi_7'].iloc[-1], + # MACD + 'prev_macd': prev_day_data['macd'].iloc[-1] if 'macd' in prev_day_data.columns else None, + 'prev_macd_hist': prev_day_data['macd_hist'].iloc[-1] if 'macd_hist' in prev_day_data.columns else None, + # 布林带 + 'prev_bb_width': prev_day_data['bb_width'].iloc[-1] if 'bb_width' in prev_day_data.columns else None, + 'prev_bb_position': prev_day_data['bb_position'].iloc[-1] if 'bb_position' in prev_day_data.columns else None, + # 波动率 + 'prev_atr_pct': prev_day_data['atr_pct'].iloc[-1] if 'atr_pct' in prev_day_data.columns else None, + 'prev_volatility_5': prev_day_data['volatility_5'].iloc[-1] if 'volatility_5' in prev_day_data.columns else None, + 'prev_volatility_20': prev_day_data['volatility_20'].iloc[-1] if 'volatility_20' in prev_day_data.columns else None, + # 成交量 + 'prev_volume_ratio': prev_day_data['volume_ratio'].iloc[-1] if 'volume_ratio' in prev_day_data.columns else None, + # 动量 + 'prev_momentum_1': prev_day_data['momentum_1'].iloc[-1] if 'momentum_1' in prev_day_data.columns else None, + 'prev_momentum_3': prev_day_data['momentum_3'].iloc[-1] if 'momentum_3' in prev_day_data.columns else None, + 'prev_momentum_7': prev_day_data['momentum_7'].iloc[-1] if 'momentum_7' in prev_day_data.columns else None, + 'prev_momentum_24': prev_day_data['momentum_24'].iloc[-1] if 'momentum_24' in prev_day_data.columns else None, + # 价格相对MA位置 + 'prev_price_vs_sma5': prev_day_data['price_vs_sma5'].iloc[-1] if 'price_vs_sma5' in prev_day_data.columns else None, + 'prev_price_vs_sma20': prev_day_data['price_vs_sma20'].iloc[-1] if 'price_vs_sma20' in prev_day_data.columns else None, + } + + prev_day_features.append(features) + + # 转换为DataFrame + features_df = pd.DataFrame(prev_day_features) + + # ========== 分析共同点 ========== + print("\n" + "="*80) + print("📊 前一天技术特征统计分析") + print("="*80) + + print(f"\n分析样本数: {len(features_df)}个低准确率交易日") + print(f" 2019年: {(features_df['period']=='2019年').sum()}天") + print(f" 2026年: {(features_df['period']=='2026年').sum()}天") + + # 统计各特征的平均值、中位数、标准差 + numeric_cols = [col for col in features_df.columns if col.startswith('prev_') and features_df[col].dtype in [np.float64, np.int64]] + + print(f"\n前一天技术特征统计:") + print(f"{'特征':<30} {'平均值':>10} {'中位数':>10} {'标准差':>10} {'共同点':<30}") + print("-" * 95) + + for col in numeric_cols: + if features_df[col].notna().sum() > 0: + mean_val = features_df[col].mean() + median_val = features_df[col].median() + std_val = features_df[col].std() + + # 判断共同点 + common_trait = "" + if 'rsi' in col: + if mean_val > 70: + common_trait = "超买区域" + elif mean_val < 30: + common_trait = "超卖区域" + elif 45 <= mean_val <= 55: + common_trait = "中性区域" + elif 'volatility' in col: + if mean_val > 3: + common_trait = "高波动" + elif mean_val < 1: + common_trait = "低波动" + elif 'momentum' in col: + if mean_val > 2: + common_trait = "强势上涨" + elif mean_val < -2: + common_trait = "强势下跌" + elif abs(mean_val) < 0.5: + common_trait = "横盘震荡" + elif 'bb_position' in col: + if mean_val > 80: + common_trait = "接近布林上轨" + elif mean_val < 20: + common_trait = "接近布林下轨" + elif 'volume_ratio' in col: + if mean_val > 1.5: + common_trait = "成交量放大" + elif mean_val < 0.8: + common_trait = "成交量萎缩" + + print(f"{col:<30} {mean_val:>10.2f} {median_val:>10.2f} {std_val:>10.2f} {common_trait:<30}") + + # ========== 对比分析 ========== + print("\n" + "="*80) + print("🔍 2019年 vs 2026年对比分析") + print("="*80) + + features_2019 = features_df[features_df['period'] == '2019年'] + features_2026 = features_df[features_df['period'] == '2026年'] + + if len(features_2019) > 0 and len(features_2026) > 0: + print(f"\n{'特征':<30} {'2019年平均':>12} {'2026年平均':>12} {'差异':>10}") + print("-" * 70) + + for col in numeric_cols[:15]: # 显示前15个特征 + if features_2019[col].notna().sum() > 0 and features_2026[col].notna().sum() > 0: + mean_2019 = features_2019[col].mean() + mean_2026 = features_2026[col].mean() + diff = mean_2026 - mean_2019 + + print(f"{col:<30} {mean_2019:>12.2f} {mean_2026:>12.2f} {diff:>10.2f}") + + # ========== 关键共同点总结 ========== + print("\n" + "="*80) + print("💡 关键共同点总结") + print("="*80) + + # RSI分析 + rsi_mean = features_df['prev_rsi_14'].mean() + print(f"\n1. RSI指标:") + print(f" 平均RSI: {rsi_mean:.1f}") + if rsi_mean > 70: + print(f" 🔴 共同点: 前一天处于超买状态") + elif rsi_mean < 30: + print(f" 🟢 共同点: 前一天处于超卖状态") + else: + print(f" ➖ 共同点: 前一天RSI在中性区域") + + # 波动率分析 + vol_mean = features_df['prev_volatility_20'].mean() + print(f"\n2. 波动率:") + print(f" 平均波动率: {vol_mean:.2f}%") + if vol_mean > 3: + print(f" ⚠️ 共同点: 前一天波动率较高") + elif vol_mean < 1: + print(f" ✅ 共同点: 前一天波动率较低") + + # 动量分析 + mom_24_mean = features_df['prev_momentum_24'].mean() + print(f"\n3. 24小时动量:") + print(f" 平均动量: {mom_24_mean:.2f}%") + if abs(mom_24_mean) > 5: + print(f" 🚀 共同点: 前一天有强趋势（{'+' if mom_24_mean > 0 else ''}{mom_24_mean:.1f}%）") + else: + print(f" 🔄 共同点: 前一天横盘震荡") + + # 布林带位置 + bb_pos_mean = features_df['prev_bb_position'].mean() + print(f"\n4. 布林带位置:") + print(f" 平均位置: {bb_pos_mean:.1f}%") + if bb_pos_mean > 80: + print(f" 🔴 共同点: 前一天价格接近布林上轨") + elif bb_pos_mean < 20: + print(f" 🟢 共同点: 前一天价格接近布林下轨") + else: + print(f" ➖ 共同点: 前一天价格在布林带中部") + + # 成交量 + vol_ratio_mean = features_df['prev_volume_ratio'].mean() + print(f"\n5. 成交量:") + print(f" 平均成交量比: {vol_ratio_mean:.2f}x") + if vol_ratio_mean > 1.5: + print(f" 📈 共同点: 前一天成交量显著放大") + elif vol_ratio_mean < 0.8: + print(f" 📉 共同点: 前一天成交量萎缩") + + print("\n" + "="*80) + print("✓ 技术特征分析完成!") + print("="*80) + + # 保存详细数据 + output_file = "./results/low_accuracy_prev_day_features.csv" + os.makedirs("./results", exist_ok=True) + features_df.to_csv(output_file, index=False) + print(f"\n✓ 详细数据已保存至: {output_file}") + + +if __name__ == "__main__": + main() diff --git a/catboost_info/catboost_training.json b/catboost_info/catboost_training.json new file mode 100644 index 0000000..3d858ff --- /dev/null +++ b/catboost_info/catboost_training.json @@ -0,0 +1,144 @@ +{ +"meta":{"test_sets":["test"],"test_metrics":[{"best_value":"Max","name":"Accuracy"},{"best_value":"Min","name":"Logloss"}],"learn_metrics":[{"best_value":"Max","name":"Accuracy"},{"best_value":"Min","name":"Logloss"}],"launch_mode":"Train","parameters":"","iteration_count":1000,"learn_sets":["learn"],"name":"experiment"}, +"iterations":[ +{"learn":[0.5470774019,0.6925856518],"iteration":0,"passed_time":0.06120060416,"remaining_time":61.13940356,"test":[0.5215327622,0.6929069842]}, +{"learn":[0.5485192475,0.6920296194],"iteration":1,"passed_time":0.06652548291,"remaining_time":33.19621597,"test":[0.5201435517,0.6926980686]}, +{"learn":[0.5508078912,0.6915437145],"iteration":2,"passed_time":0.07133867461,"remaining_time":23.70821953,"test":[0.5235008104,0.6925995058]}, +{"learn":[0.5511969607,0.6911042525],"iteration":3,"passed_time":0.07626899597,"remaining_time":18.99098,"test":[0.5243111831,0.6923741119]}, +{"learn":[0.5528218977,0.6906429178],"iteration":4,"passed_time":0.08172279652,"remaining_time":16.26283651,"test":[0.5240796481,0.6922359346]}, +{"learn":[0.553096535,0.6902328044],"iteration":5,"passed_time":0.08686295957,"remaining_time":14.39029697,"test":[0.5217642973,0.6921532273]}, +{"learn":[0.5537144688,0.6899070645],"iteration":6,"passed_time":0.0917179446,"remaining_time":13.01084557,"test":[0.5241954156,0.6920533177]}, +{"learn":[0.552730352,0.6895966245],"iteration":7,"passed_time":0.09718091218,"remaining_time":12.05043311,"test":[0.520838157,0.6919667553]}, +{"learn":[0.5539204467,0.6892625204],"iteration":8,"passed_time":0.1018938082,"remaining_time":11.21964044,"test":[0.520838157,0.6919667553]}, +{"learn":[0.554561267,0.6889386437],"iteration":9,"passed_time":0.1079246734,"remaining_time":10.68454267,"test":[0.520838157,0.6918984047]}, +{"learn":[0.554561267,0.6886345552],"iteration":10,"passed_time":0.1134390597,"remaining_time":10.19920273,"test":[0.5213012271,0.6918473786]}, +{"learn":[0.5548587907,0.6883728473],"iteration":11,"passed_time":0.1182243753,"remaining_time":9.7338069,"test":[0.5217642973,0.691760433]}, +{"learn":[0.5552020872,0.6881185803],"iteration":12,"passed_time":0.1234375829,"remaining_time":9.371761104,"test":[0.5231535078,0.6917269838]}, +{"learn":[0.5564150684,0.6879043975],"iteration":13,"passed_time":0.1290524316,"remaining_time":9.088978394,"test":[0.5210696921,0.6916992448]}, +{"learn":[0.5559802261,0.6877314929],"iteration":14,"passed_time":0.1346781556,"remaining_time":8.843865552,"test":[0.522343135,0.6916911046]}, +{"learn":[0.5564150684,0.6875118582],"iteration":15,"passed_time":0.1392779638,"remaining_time":8.565594775,"test":[0.5213012271,0.6916992139]}, +{"learn":[0.5569414565,0.6872755986],"iteration":16,"passed_time":0.1448462273,"remaining_time":8.375520082,"test":[0.5204908544,0.6917199845]}, +{"learn":[0.5561862041,0.6870435386],"iteration":17,"passed_time":0.1499537223,"remaining_time":8.18080863,"test":[0.5226904376,0.6917129587]}, +{"learn":[0.5561633176,0.6868442431],"iteration":18,"passed_time":0.1555536537,"remaining_time":8.031480752,"test":[0.5225746701,0.6916992772]}, +{"learn":[0.5575593903,0.6866694237],"iteration":19,"passed_time":0.1608244053,"remaining_time":7.88039586,"test":[0.5207223894,0.6917050845]}, +{"learn":[0.5575593903,0.686503203],"iteration":20,"passed_time":0.1667731839,"remaining_time":7.774807002,"test":[0.5247742533,0.6916027729]}, +{"learn":[0.557124548,0.6863452751],"iteration":21,"passed_time":0.171689088,"remaining_time":7.632360368,"test":[0.5239638805,0.6916046624]}, +{"learn":[0.5569185701,0.6862172686],"iteration":22,"passed_time":0.1765527817,"remaining_time":7.49965512,"test":[0.5241954156,0.6916203654]}, +{"learn":[0.5573762988,0.6860730724],"iteration":23,"passed_time":0.1818841191,"remaining_time":7.396620842,"test":[0.5243111831,0.6916491893]}, +{"learn":[0.5571703209,0.6859302623],"iteration":24,"passed_time":0.186612224,"remaining_time":7.277876738,"test":[0.5244269507,0.6916666458]}, +{"learn":[0.5570101158,0.6858087212],"iteration":25,"passed_time":0.1914440415,"remaining_time":7.171788323,"test":[0.5235008104,0.691666967]}, +{"learn":[0.5571474344,0.6857036471],"iteration":26,"passed_time":0.1967539613,"remaining_time":7.090429791,"test":[0.5229219727,0.6916827648]}, +{"learn":[0.5571932073,0.6855803136],"iteration":27,"passed_time":0.2023491842,"remaining_time":7.024407393,"test":[0.5245427182,0.6917018334]}, +{"learn":[0.5572160937,0.6854590307],"iteration":28,"passed_time":0.2073690924,"remaining_time":6.943289267,"test":[0.5222273674,0.6916998315]}, +{"learn":[0.5568041379,0.6853245403],"iteration":29,"passed_time":0.2177755069,"remaining_time":7.041408056,"test":[0.5224589025,0.6917500551]}, +{"learn":[0.5570558887,0.6852010211],"iteration":30,"passed_time":0.2259016222,"remaining_time":7.061247481,"test":[0.5222273674,0.691867972]}, +{"learn":[0.5570330022,0.6850648161],"iteration":31,"passed_time":0.2344519627,"remaining_time":7.092171873,"test":[0.5215327622,0.6918745283]}, +{"learn":[0.5578340276,0.6849486439],"iteration":32,"passed_time":0.2411928145,"remaining_time":7.067680354,"test":[0.5214169947,0.691873578]}, +{"learn":[0.5576738225,0.6848167763],"iteration":33,"passed_time":0.2475354838,"remaining_time":7.032919923,"test":[0.5215327622,0.6918621914]}, +{"learn":[0.5577424818,0.6847064338],"iteration":34,"passed_time":0.255081701,"remaining_time":7.0329669,"test":[0.5222273674,0.691870106]}, +{"learn":[0.5582917563,0.6846073419],"iteration":35,"passed_time":0.2616022524,"remaining_time":7.005126981,"test":[0.5219958324,0.6919348629]}, +{"learn":[0.5587037122,0.6845016816],"iteration":36,"passed_time":0.2691936381,"remaining_time":7.006310094,"test":[0.5245427182,0.6920133967]}, +{"learn":[0.5587494851,0.6844106082],"iteration":37,"passed_time":0.2758961134,"remaining_time":6.984527923,"test":[0.5246584858,0.6920145372]}, +{"learn":[0.5591156681,0.6843432648],"iteration":38,"passed_time":0.2833089086,"remaining_time":6.981022081,"test":[0.5247742533,0.6920231409]}, +{"learn":[0.5589325766,0.6842678359],"iteration":39,"passed_time":0.2897567071,"remaining_time":6.95416097,"test":[0.5248900208,0.6920429941]}, +{"learn":[0.5590012359,0.6841515192],"iteration":40,"passed_time":0.3004245903,"remaining_time":7.027004441,"test":[0.5248900208,0.6920318228]}, +{"learn":[0.5590698952,0.6840610381],"iteration":41,"passed_time":0.3083540311,"remaining_time":7.033408614,"test":[0.5252373235,0.6920202842]}, +{"learn":[0.5593445324,0.6839565281],"iteration":42,"passed_time":0.3168845792,"remaining_time":7.052524239,"test":[0.525353091,0.6920430337]}, +{"learn":[0.5592529867,0.6838539206],"iteration":43,"passed_time":0.3241559521,"remaining_time":7.043024778,"test":[0.5247742533,0.6920265347]}, +{"learn":[0.559321646,0.6837417759],"iteration":44,"passed_time":0.3312213169,"remaining_time":7.029252391,"test":[0.525353091,0.6920697067]}, +{"learn":[0.559687829,0.6836669298],"iteration":45,"passed_time":0.3364545253,"remaining_time":6.977774285,"test":[0.5263949988,0.692116771]}, +{"learn":[0.5603744221,0.6835634424],"iteration":46,"passed_time":0.3416096056,"remaining_time":6.926679875,"test":[0.5259319287,0.692098178]}, +{"learn":[0.5599853527,0.6834698484],"iteration":47,"passed_time":0.3493788734,"remaining_time":6.929347655,"test":[0.5260476962,0.692152499]}, +{"learn":[0.5601913306,0.6833860083],"iteration":48,"passed_time":0.3567442501,"remaining_time":6.923750649,"test":[0.5262792313,0.6922137468]}, +{"learn":[0.5602599899,0.6833086288],"iteration":49,"passed_time":0.3643377608,"remaining_time":6.922417456,"test":[0.5270896041,0.6922167687]}, +{"learn":[0.5606490594,0.683239475],"iteration":50,"passed_time":0.3696800569,"remaining_time":6.87894851,"test":[0.5272053716,0.6922875182]}, +{"learn":[0.5609694695,0.6831625258],"iteration":51,"passed_time":0.3754608288,"remaining_time":6.844939726,"test":[0.526858069,0.6923662256]}, +{"learn":[0.5611983339,0.683056699],"iteration":52,"passed_time":0.3809127543,"remaining_time":6.806120346,"test":[0.526858069,0.6923707756]}, +{"learn":[0.5617018355,0.6829611591],"iteration":53,"passed_time":0.3866230651,"remaining_time":6.773063325,"test":[0.5270896041,0.6923776257]}, +{"learn":[0.5619993592,0.6828690012],"iteration":54,"passed_time":0.3917164763,"remaining_time":6.730401274,"test":[0.5259319287,0.692400255]}, +{"learn":[0.5620451321,0.6827776665],"iteration":55,"passed_time":0.3965791283,"remaining_time":6.685191019,"test":[0.5266265339,0.692269877]}, +{"learn":[0.5624342015,0.6826927644],"iteration":56,"passed_time":0.4017419589,"remaining_time":6.646362583,"test":[0.5260476962,0.6922932744]}, +{"learn":[0.5626401794,0.6825957686],"iteration":57,"passed_time":0.4064048529,"remaining_time":6.60057537,"test":[0.5266265339,0.6923264029]}, +{"learn":[0.5626630659,0.6825449467],"iteration":58,"passed_time":0.4116929385,"remaining_time":6.566153478,"test":[0.5263949988,0.692314059]}, +{"learn":[0.5627088387,0.6824559998],"iteration":59,"passed_time":0.4168931457,"remaining_time":6.531325949,"test":[0.5263949988,0.6923015601]}, +{"learn":[0.562983476,0.6823684725],"iteration":60,"passed_time":0.4216150837,"remaining_time":6.4901076,"test":[0.5263949988,0.6923592968]}, +{"learn":[0.5632352268,0.6822923813],"iteration":61,"passed_time":0.4271669298,"remaining_time":6.462622261,"test":[0.5267423015,0.6923476579]}, +{"learn":[0.5630521353,0.6822333406],"iteration":62,"passed_time":0.4320567079,"remaining_time":6.425986275,"test":[0.5265107664,0.692348824]}, +{"learn":[0.5634412047,0.6821367311],"iteration":63,"passed_time":0.4365471784,"remaining_time":6.384502485,"test":[0.5259319287,0.6923600086]}, +{"learn":[0.5636242962,0.6820867227],"iteration":64,"passed_time":0.441430998,"remaining_time":6.349815124,"test":[0.5254688585,0.692372751]}, +{"learn":[0.563715842,0.6820256999],"iteration":65,"passed_time":0.4463418185,"remaining_time":6.316413008,"test":[0.5260476962,0.6924455742]}, +{"learn":[0.5638531606,0.6819329129],"iteration":66,"passed_time":0.4521836345,"remaining_time":6.296825836,"test":[0.5255846261,0.6924748157]}, +{"learn":[0.5638989335,0.6818706983],"iteration":67,"passed_time":0.4572557949,"remaining_time":6.26709413,"test":[0.5273211392,0.6924789867]}, +{"learn":[0.5639447064,0.6818045601],"iteration":68,"passed_time":0.4627077204,"remaining_time":6.24320127,"test":[0.5267423015,0.6924876711]}, +{"learn":[0.564082025,0.68171304],"iteration":69,"passed_time":0.4675020363,"remaining_time":6.211098482,"test":[0.5266265339,0.692489792]}, +{"learn":[0.5640362521,0.6816251881],"iteration":70,"passed_time":0.4724773595,"remaining_time":6.182133337,"test":[0.5266265339,0.6924869911]}, +{"learn":[0.5645168673,0.6815203712],"iteration":71,"passed_time":0.4776511072,"remaining_time":6.156392048,"test":[0.5267423015,0.6925206162]}, +{"learn":[0.564448208,0.6814559429],"iteration":72,"passed_time":0.4825719282,"remaining_time":6.12800243,"test":[0.5262792313,0.6925313913]}, +{"learn":[0.5646541859,0.6813334168],"iteration":73,"passed_time":0.4878153037,"remaining_time":6.104283395,"test":[0.5258161611,0.6925367884]}, +{"learn":[0.5649059367,0.6812138418],"iteration":74,"passed_time":0.493698288,"remaining_time":6.088945552,"test":[0.5265107664,0.6925626459]}, +{"learn":[0.5649059367,0.6811428702],"iteration":75,"passed_time":0.4991032116,"remaining_time":6.068044309,"test":[0.5273211392,0.6925711713]}, +{"learn":[0.565180574,0.6810679335],"iteration":76,"passed_time":0.5041589963,"remaining_time":6.043360436,"test":[0.526858069,0.6926776812]}, +{"learn":[0.5653407791,0.6810189357],"iteration":77,"passed_time":0.509317035,"remaining_time":6.020388543,"test":[0.5274369067,0.6926884897]}, +{"learn":[0.5655696434,0.6809239905],"iteration":78,"passed_time":0.5151628512,"remaining_time":6.005885898,"test":[0.5266265339,0.6926819413]}, +{"learn":[0.5651119147,0.6808452293],"iteration":79,"passed_time":0.5199159989,"remaining_time":5.979033987,"test":[0.5272053716,0.692682082]}, +{"learn":[0.5653407791,0.6807843648],"iteration":80,"passed_time":0.5252913379,"remaining_time":5.959786908,"test":[0.5270896041,0.6927326486]}, +{"learn":[0.565180574,0.6807084812],"iteration":81,"passed_time":0.5302294096,"remaining_time":5.935982903,"test":[0.5266265339,0.69272742]}, +{"learn":[0.5654780977,0.6806237044],"iteration":82,"passed_time":0.5348115921,"remaining_time":5.908701566,"test":[0.5272053716,0.6927260891]}, +{"learn":[0.5656840756,0.680541791],"iteration":83,"passed_time":0.5399401714,"remaining_time":5.887919012,"test":[0.5267423015,0.6927850957]}, +{"learn":[0.5653636655,0.6804402561],"iteration":84,"passed_time":0.5459297433,"remaining_time":5.876773119,"test":[0.5266265339,0.6928797523]}, +{"learn":[0.565546757,0.6803841011],"iteration":85,"passed_time":0.5557997614,"remaining_time":5.906988161,"test":[0.526858069,0.6928753106]}, +{"learn":[0.5657298485,0.6803302076],"iteration":86,"passed_time":0.5625750729,"remaining_time":5.903805075,"test":[0.525353091,0.6929332764]}, +{"learn":[0.5663248959,0.6802294187],"iteration":87,"passed_time":0.5682492989,"remaining_time":5.889129097,"test":[0.5265107664,0.6929225447]}, +{"learn":[0.5660273722,0.6801427052],"iteration":88,"passed_time":0.5754560859,"remaining_time":5.890342632,"test":[0.5262792313,0.6929127979]}, +{"learn":[0.5664393281,0.680092343],"iteration":89,"passed_time":0.5809644303,"remaining_time":5.874195906,"test":[0.5278999768,0.6929168776]}, +{"learn":[0.5666224196,0.6800188515],"iteration":90,"passed_time":0.5858589586,"remaining_time":5.852151575,"test":[0.5261634638,0.6930123071]}, +{"learn":[0.567377672,0.6799633931],"iteration":91,"passed_time":0.5921711266,"remaining_time":5.844471554,"test":[0.5260476962,0.6930271402]}, +{"learn":[0.5676523092,0.6799142593],"iteration":92,"passed_time":0.597831102,"remaining_time":5.830460318,"test":[0.525353091,0.693074147]}, +{"learn":[0.5677667414,0.6798218244],"iteration":93,"passed_time":0.602484454,"remaining_time":5.806924632,"test":[0.5257003936,0.693098238]}, +{"learn":[0.5679727194,0.6797624264],"iteration":94,"passed_time":0.6079672558,"remaining_time":5.791688068,"test":[0.5252373235,0.6931168231]}, +{"learn":[0.5682702431,0.6796825487],"iteration":95,"passed_time":0.6136230643,"remaining_time":5.778283856,"test":[0.5252373235,0.6930889734]}, +{"learn":[0.5683160159,0.6796119175],"iteration":96,"passed_time":0.6183110011,"remaining_time":5.756029216,"test":[0.5265107664,0.6931814079]}, +{"learn":[0.5681786973,0.6795251979],"iteration":97,"passed_time":0.6239766434,"remaining_time":5.743131962,"test":[0.5265107664,0.6935736538]}, +{"learn":[0.5684533346,0.6794647206],"iteration":98,"passed_time":0.6295706578,"remaining_time":5.729728916,"test":[0.5257003936,0.6935709125]}, +{"learn":[0.5690026091,0.6793963783],"iteration":99,"passed_time":0.634798366,"remaining_time":5.713185294,"test":[0.5262792313,0.6935839531]}, +{"learn":[0.5689797226,0.6793126045],"iteration":100,"passed_time":0.6405399697,"remaining_time":5.701439929,"test":[0.5250057884,0.6936165151]}, +{"learn":[0.569208587,0.6792105367],"iteration":101,"passed_time":0.6456661739,"remaining_time":5.684394354,"test":[0.5250057884,0.6936424871]}, +{"learn":[0.5693687921,0.6791494493],"iteration":102,"passed_time":0.6504003208,"remaining_time":5.6641659,"test":[0.5246584858,0.6936432847]}, +{"learn":[0.5695518836,0.6790732211],"iteration":103,"passed_time":0.6554747729,"remaining_time":5.647167274,"test":[0.5245427182,0.6936493519]}, +{"learn":[0.5698265208,0.6789905315],"iteration":104,"passed_time":0.6618330678,"remaining_time":5.641339006,"test":[0.5248900208,0.6936865636]}, +{"learn":[0.5698951801,0.67895488],"iteration":105,"passed_time":0.667103236,"remaining_time":5.626323519,"test":[0.5252373235,0.693675143]}, +{"learn":[0.5700324987,0.6788948137],"iteration":106,"passed_time":0.6723126518,"remaining_time":5.61098316,"test":[0.5255846261,0.6936874053]}, +{"learn":[0.5704902275,0.6788457101],"iteration":107,"passed_time":0.6775288179,"remaining_time":5.595886163,"test":[0.5254688585,0.693689784]}, +{"learn":[0.5704902275,0.6787737664],"iteration":108,"passed_time":0.6826580222,"remaining_time":5.580259613,"test":[0.5251215559,0.6936707742]}, +{"learn":[0.5708106376,0.6786897689],"iteration":109,"passed_time":0.6879676087,"remaining_time":5.566283379,"test":[0.5241954156,0.6937073345]}, +{"learn":[0.5709021834,0.6786357569],"iteration":110,"passed_time":0.6932970293,"remaining_time":5.552622153,"test":[0.5243111831,0.6937614267]}, +{"learn":[0.5711768206,0.6785404316],"iteration":111,"passed_time":0.6991878889,"remaining_time":5.543561119,"test":[0.5219958324,0.6937989599]}, +{"learn":[0.5714972307,0.6784635222],"iteration":112,"passed_time":0.7044325144,"remaining_time":5.529483542,"test":[0.5218800648,0.6938027615]}, +{"learn":[0.5716345494,0.6783830081],"iteration":113,"passed_time":0.7094082543,"remaining_time":5.513471169,"test":[0.5232692753,0.6939119665]}, +{"learn":[0.5716803222,0.6783261377],"iteration":114,"passed_time":0.714549459,"remaining_time":5.498924098,"test":[0.5235008104,0.6939374584]}, +{"learn":[0.5717947544,0.6782551829],"iteration":115,"passed_time":0.7192722305,"remaining_time":5.481350446,"test":[0.5229219727,0.6939266092]}, +{"learn":[0.5717260951,0.6782039946],"iteration":116,"passed_time":0.7258457006,"remaining_time":5.477963706,"test":[0.5222273674,0.6939398216]}, +{"learn":[0.5719549595,0.6781381441],"iteration":117,"passed_time":0.7316315144,"remaining_time":5.468635557,"test":[0.5221115999,0.6939581303]}, +{"learn":[0.5720007324,0.6780959678],"iteration":118,"passed_time":0.7375646257,"remaining_time":5.460457439,"test":[0.522343135,0.6940189884]}, +{"learn":[0.5721609374,0.6780580972],"iteration":119,"passed_time":0.7444090649,"remaining_time":5.45899981,"test":[0.5224589025,0.6940203223]}, +{"learn":[0.5722982561,0.6779909469],"iteration":120,"passed_time":0.7511739177,"remaining_time":5.456874989,"test":[0.5228062051,0.6940373975]}, +{"learn":[0.5722753696,0.6779155506],"iteration":121,"passed_time":0.7581201944,"remaining_time":5.455979759,"test":[0.5216485298,0.6940836851]}, +{"learn":[0.5726186662,0.67787157],"iteration":122,"passed_time":0.7638367137,"remaining_time":5.446217869,"test":[0.5217642973,0.6940971538]}, +{"learn":[0.5729161899,0.6778021605],"iteration":123,"passed_time":0.7741768755,"remaining_time":5.469185024,"test":[0.5222273674,0.6941038908]}, +{"learn":[0.5732594864,0.6777441308],"iteration":124,"passed_time":0.7811099433,"remaining_time":5.467769603,"test":[0.5222273674,0.6941318899]}, +{"learn":[0.5737172152,0.6776640278],"iteration":125,"passed_time":0.7923285151,"remaining_time":5.495993034,"test":[0.5245427182,0.6941318581]}, +{"learn":[0.5741062846,0.6775768826],"iteration":126,"passed_time":0.7972769206,"remaining_time":5.480494108,"test":[0.5240796481,0.6939774214]}, +{"learn":[0.5738545338,0.6775322572],"iteration":127,"passed_time":0.8045955871,"remaining_time":5.481307437,"test":[0.5246584858,0.6939790446]}, +{"learn":[0.5738545338,0.6774381457],"iteration":128,"passed_time":0.8096396213,"remaining_time":5.466636513,"test":[0.5244269507,0.6939875313]}, +{"learn":[0.5739460796,0.6773933791],"iteration":129,"passed_time":0.8141404672,"remaining_time":5.448478511,"test":[0.5239638805,0.6940065251]}, +{"learn":[0.5738545338,0.6773073305],"iteration":130,"passed_time":0.8197628995,"remaining_time":5.437969158,"test":[0.5237323455,0.6940062666]}, +{"learn":[0.5746555591,0.6772430348],"iteration":131,"passed_time":0.8257601383,"remaining_time":5.429998485,"test":[0.5231535078,0.6940655128]}, +{"learn":[0.5745640134,0.6771630132],"iteration":132,"passed_time":0.8309698042,"remaining_time":5.41692346,"test":[0.5226904376,0.6940488902]}, +{"learn":[0.5747242184,0.6770868641],"iteration":133,"passed_time":0.8383795993,"remaining_time":5.418184574,"test":[0.5221115999,0.6940625077]}, +{"learn":[0.5749988557,0.6770144627],"iteration":134,"passed_time":0.8433877987,"remaining_time":5.403929229,"test":[0.5225746701,0.6940868198]}, +{"learn":[0.5747699913,0.6769488704],"iteration":135,"passed_time":0.848858975,"remaining_time":5.392751135,"test":[0.5229219727,0.6940640084]}, +{"learn":[0.5751590607,0.6768712881],"iteration":136,"passed_time":0.8548430883,"remaining_time":5.384887483,"test":[0.5232692753,0.6941097371]}, +{"learn":[0.5755939031,0.6767999464],"iteration":137,"passed_time":0.860643611,"remaining_time":5.375904295,"test":[0.5244269507,0.6940823091]}, +{"learn":[0.5755710166,0.6767315844],"iteration":138,"passed_time":0.8670014892,"remaining_time":5.370419297,"test":[0.5237323455,0.6941720366]}, +{"learn":[0.5758914267,0.6766580612],"iteration":139,"passed_time":0.8721688617,"remaining_time":5.357608722,"test":[0.5235008104,0.6942806828]} +]} \ No newline at end of file diff --git a/catboost_info/learn/events.out.tfevents b/catboost_info/learn/events.out.tfevents new file mode 100644 index 0000000..779407d Binary files /dev/null and b/catboost_info/learn/events.out.tfevents differ diff --git a/catboost_info/learn_error.tsv b/catboost_info/learn_error.tsv new file mode 100644 index 0000000..c293302 --- /dev/null +++ b/catboost_info/learn_error.tsv @@ -0,0 +1,141 @@ +iter Accuracy Logloss +0 0.5470774019 0.6925856518 +1 0.5485192475 0.6920296194 +2 0.5508078912 0.6915437145 +3 0.5511969607 0.6911042525 +4 0.5528218977 0.6906429178 +5 0.553096535 0.6902328044 +6 0.5537144688 0.6899070645 +7 0.552730352 0.6895966245 +8 0.5539204467 0.6892625204 +9 0.554561267 0.6889386437 +10 0.554561267 0.6886345552 +11 0.5548587907 0.6883728473 +12 0.5552020872 0.6881185803 +13 0.5564150684 0.6879043975 +14 0.5559802261 0.6877314929 +15 0.5564150684 0.6875118582 +16 0.5569414565 0.6872755986 +17 0.5561862041 0.6870435386 +18 0.5561633176 0.6868442431 +19 0.5575593903 0.6866694237 +20 0.5575593903 0.686503203 +21 0.557124548 0.6863452751 +22 0.5569185701 0.6862172686 +23 0.5573762988 0.6860730724 +24 0.5571703209 0.6859302623 +25 0.5570101158 0.6858087212 +26 0.5571474344 0.6857036471 +27 0.5571932073 0.6855803136 +28 0.5572160937 0.6854590307 +29 0.5568041379 0.6853245403 +30 0.5570558887 0.6852010211 +31 0.5570330022 0.6850648161 +32 0.5578340276 0.6849486439 +33 0.5576738225 0.6848167763 +34 0.5577424818 0.6847064338 +35 0.5582917563 0.6846073419 +36 0.5587037122 0.6845016816 +37 0.5587494851 0.6844106082 +38 0.5591156681 0.6843432648 +39 0.5589325766 0.6842678359 +40 0.5590012359 0.6841515192 +41 0.5590698952 0.6840610381 +42 0.5593445324 0.6839565281 +43 0.5592529867 0.6838539206 +44 0.559321646 0.6837417759 +45 0.559687829 0.6836669298 +46 0.5603744221 0.6835634424 +47 0.5599853527 0.6834698484 +48 0.5601913306 0.6833860083 +49 0.5602599899 0.6833086288 +50 0.5606490594 0.683239475 +51 0.5609694695 0.6831625258 +52 0.5611983339 0.683056699 +53 0.5617018355 0.6829611591 +54 0.5619993592 0.6828690012 +55 0.5620451321 0.6827776665 +56 0.5624342015 0.6826927644 +57 0.5626401794 0.6825957686 +58 0.5626630659 0.6825449467 +59 0.5627088387 0.6824559998 +60 0.562983476 0.6823684725 +61 0.5632352268 0.6822923813 +62 0.5630521353 0.6822333406 +63 0.5634412047 0.6821367311 +64 0.5636242962 0.6820867227 +65 0.563715842 0.6820256999 +66 0.5638531606 0.6819329129 +67 0.5638989335 0.6818706983 +68 0.5639447064 0.6818045601 +69 0.564082025 0.68171304 +70 0.5640362521 0.6816251881 +71 0.5645168673 0.6815203712 +72 0.564448208 0.6814559429 +73 0.5646541859 0.6813334168 +74 0.5649059367 0.6812138418 +75 0.5649059367 0.6811428702 +76 0.565180574 0.6810679335 +77 0.5653407791 0.6810189357 +78 0.5655696434 0.6809239905 +79 0.5651119147 0.6808452293 +80 0.5653407791 0.6807843648 +81 0.565180574 0.6807084812 +82 0.5654780977 0.6806237044 +83 0.5656840756 0.680541791 +84 0.5653636655 0.6804402561 +85 0.565546757 0.6803841011 +86 0.5657298485 0.6803302076 +87 0.5663248959 0.6802294187 +88 0.5660273722 0.6801427052 +89 0.5664393281 0.680092343 +90 0.5666224196 0.6800188515 +91 0.567377672 0.6799633931 +92 0.5676523092 0.6799142593 +93 0.5677667414 0.6798218244 +94 0.5679727194 0.6797624264 +95 0.5682702431 0.6796825487 +96 0.5683160159 0.6796119175 +97 0.5681786973 0.6795251979 +98 0.5684533346 0.6794647206 +99 0.5690026091 0.6793963783 +100 0.5689797226 0.6793126045 +101 0.569208587 0.6792105367 +102 0.5693687921 0.6791494493 +103 0.5695518836 0.6790732211 +104 0.5698265208 0.6789905315 +105 0.5698951801 0.67895488 +106 0.5700324987 0.6788948137 +107 0.5704902275 0.6788457101 +108 0.5704902275 0.6787737664 +109 0.5708106376 0.6786897689 +110 0.5709021834 0.6786357569 +111 0.5711768206 0.6785404316 +112 0.5714972307 0.6784635222 +113 0.5716345494 0.6783830081 +114 0.5716803222 0.6783261377 +115 0.5717947544 0.6782551829 +116 0.5717260951 0.6782039946 +117 0.5719549595 0.6781381441 +118 0.5720007324 0.6780959678 +119 0.5721609374 0.6780580972 +120 0.5722982561 0.6779909469 +121 0.5722753696 0.6779155506 +122 0.5726186662 0.67787157 +123 0.5729161899 0.6778021605 +124 0.5732594864 0.6777441308 +125 0.5737172152 0.6776640278 +126 0.5741062846 0.6775768826 +127 0.5738545338 0.6775322572 +128 0.5738545338 0.6774381457 +129 0.5739460796 0.6773933791 +130 0.5738545338 0.6773073305 +131 0.5746555591 0.6772430348 +132 0.5745640134 0.6771630132 +133 0.5747242184 0.6770868641 +134 0.5749988557 0.6770144627 +135 0.5747699913 0.6769488704 +136 0.5751590607 0.6768712881 +137 0.5755939031 0.6767999464 +138 0.5755710166 0.6767315844 +139 0.5758914267 0.6766580612 diff --git a/catboost_info/test/events.out.tfevents b/catboost_info/test/events.out.tfevents new file mode 100644 index 0000000..84d2e9c Binary files /dev/null and b/catboost_info/test/events.out.tfevents differ diff --git a/catboost_info/test_error.tsv b/catboost_info/test_error.tsv new file mode 100644 index 0000000..b6ce44a --- /dev/null +++ b/catboost_info/test_error.tsv @@ -0,0 +1,141 @@ +iter Accuracy Logloss +0 0.5215327622 0.6929069842 +1 0.5201435517 0.6926980686 +2 0.5235008104 0.6925995058 +3 0.5243111831 0.6923741119 +4 0.5240796481 0.6922359346 +5 0.5217642973 0.6921532273 +6 0.5241954156 0.6920533177 +7 0.520838157 0.6919667553 +8 0.520838157 0.6919667553 +9 0.520838157 0.6918984047 +10 0.5213012271 0.6918473786 +11 0.5217642973 0.691760433 +12 0.5231535078 0.6917269838 +13 0.5210696921 0.6916992448 +14 0.522343135 0.6916911046 +15 0.5213012271 0.6916992139 +16 0.5204908544 0.6917199845 +17 0.5226904376 0.6917129587 +18 0.5225746701 0.6916992772 +19 0.5207223894 0.6917050845 +20 0.5247742533 0.6916027729 +21 0.5239638805 0.6916046624 +22 0.5241954156 0.6916203654 +23 0.5243111831 0.6916491893 +24 0.5244269507 0.6916666458 +25 0.5235008104 0.691666967 +26 0.5229219727 0.6916827648 +27 0.5245427182 0.6917018334 +28 0.5222273674 0.6916998315 +29 0.5224589025 0.6917500551 +30 0.5222273674 0.691867972 +31 0.5215327622 0.6918745283 +32 0.5214169947 0.691873578 +33 0.5215327622 0.6918621914 +34 0.5222273674 0.691870106 +35 0.5219958324 0.6919348629 +36 0.5245427182 0.6920133967 +37 0.5246584858 0.6920145372 +38 0.5247742533 0.6920231409 +39 0.5248900208 0.6920429941 +40 0.5248900208 0.6920318228 +41 0.5252373235 0.6920202842 +42 0.525353091 0.6920430337 +43 0.5247742533 0.6920265347 +44 0.525353091 0.6920697067 +45 0.5263949988 0.692116771 +46 0.5259319287 0.692098178 +47 0.5260476962 0.692152499 +48 0.5262792313 0.6922137468 +49 0.5270896041 0.6922167687 +50 0.5272053716 0.6922875182 +51 0.526858069 0.6923662256 +52 0.526858069 0.6923707756 +53 0.5270896041 0.6923776257 +54 0.5259319287 0.692400255 +55 0.5266265339 0.692269877 +56 0.5260476962 0.6922932744 +57 0.5266265339 0.6923264029 +58 0.5263949988 0.692314059 +59 0.5263949988 0.6923015601 +60 0.5263949988 0.6923592968 +61 0.5267423015 0.6923476579 +62 0.5265107664 0.692348824 +63 0.5259319287 0.6923600086 +64 0.5254688585 0.692372751 +65 0.5260476962 0.6924455742 +66 0.5255846261 0.6924748157 +67 0.5273211392 0.6924789867 +68 0.5267423015 0.6924876711 +69 0.5266265339 0.692489792 +70 0.5266265339 0.6924869911 +71 0.5267423015 0.6925206162 +72 0.5262792313 0.6925313913 +73 0.5258161611 0.6925367884 +74 0.5265107664 0.6925626459 +75 0.5273211392 0.6925711713 +76 0.526858069 0.6926776812 +77 0.5274369067 0.6926884897 +78 0.5266265339 0.6926819413 +79 0.5272053716 0.692682082 +80 0.5270896041 0.6927326486 +81 0.5266265339 0.69272742 +82 0.5272053716 0.6927260891 +83 0.5267423015 0.6927850957 +84 0.5266265339 0.6928797523 +85 0.526858069 0.6928753106 +86 0.525353091 0.6929332764 +87 0.5265107664 0.6929225447 +88 0.5262792313 0.6929127979 +89 0.5278999768 0.6929168776 +90 0.5261634638 0.6930123071 +91 0.5260476962 0.6930271402 +92 0.525353091 0.693074147 +93 0.5257003936 0.693098238 +94 0.5252373235 0.6931168231 +95 0.5252373235 0.6930889734 +96 0.5265107664 0.6931814079 +97 0.5265107664 0.6935736538 +98 0.5257003936 0.6935709125 +99 0.5262792313 0.6935839531 +100 0.5250057884 0.6936165151 +101 0.5250057884 0.6936424871 +102 0.5246584858 0.6936432847 +103 0.5245427182 0.6936493519 +104 0.5248900208 0.6936865636 +105 0.5252373235 0.693675143 +106 0.5255846261 0.6936874053 +107 0.5254688585 0.693689784 +108 0.5251215559 0.6936707742 +109 0.5241954156 0.6937073345 +110 0.5243111831 0.6937614267 +111 0.5219958324 0.6937989599 +112 0.5218800648 0.6938027615 +113 0.5232692753 0.6939119665 +114 0.5235008104 0.6939374584 +115 0.5229219727 0.6939266092 +116 0.5222273674 0.6939398216 +117 0.5221115999 0.6939581303 +118 0.522343135 0.6940189884 +119 0.5224589025 0.6940203223 +120 0.5228062051 0.6940373975 +121 0.5216485298 0.6940836851 +122 0.5217642973 0.6940971538 +123 0.5222273674 0.6941038908 +124 0.5222273674 0.6941318899 +125 0.5245427182 0.6941318581 +126 0.5240796481 0.6939774214 +127 0.5246584858 0.6939790446 +128 0.5244269507 0.6939875313 +129 0.5239638805 0.6940065251 +130 0.5237323455 0.6940062666 +131 0.5231535078 0.6940655128 +132 0.5226904376 0.6940488902 +133 0.5221115999 0.6940625077 +134 0.5225746701 0.6940868198 +135 0.5229219727 0.6940640084 +136 0.5232692753 0.6941097371 +137 0.5244269507 0.6940823091 +138 0.5237323455 0.6941720366 +139 0.5235008104 0.6942806828 diff --git a/catboost_info/time_left.tsv b/catboost_info/time_left.tsv new file mode 100644 index 0000000..762a38b --- /dev/null +++ b/catboost_info/time_left.tsv @@ -0,0 +1,141 @@ +iter Passed Remaining +0 61 61139 +1 66 33196 +2 71 23708 +3 76 18990 +4 81 16262 +5 86 14390 +6 91 13010 +7 97 12050 +8 101 11219 +9 107 10684 +10 113 10199 +11 118 9733 +12 123 9371 +13 129 9088 +14 134 8843 +15 139 8565 +16 144 8375 +17 149 8180 +18 155 8031 +19 160 7880 +20 166 7774 +21 171 7632 +22 176 7499 +23 181 7396 +24 186 7277 +25 191 7171 +26 196 7090 +27 202 7024 +28 207 6943 +29 217 7041 +30 225 7061 +31 234 7092 +32 241 7067 +33 247 7032 +34 255 7032 +35 261 7005 +36 269 7006 +37 275 6984 +38 283 6981 +39 289 6954 +40 300 7027 +41 308 7033 +42 316 7052 +43 324 7043 +44 331 7029 +45 336 6977 +46 341 6926 +47 349 6929 +48 356 6923 +49 364 6922 +50 369 6878 +51 375 6844 +52 380 6806 +53 386 6773 +54 391 6730 +55 396 6685 +56 401 6646 +57 406 6600 +58 411 6566 +59 416 6531 +60 421 6490 +61 427 6462 +62 432 6425 +63 436 6384 +64 441 6349 +65 446 6316 +66 452 6296 +67 457 6267 +68 462 6243 +69 467 6211 +70 472 6182 +71 477 6156 +72 482 6128 +73 487 6104 +74 493 6088 +75 499 6068 +76 504 6043 +77 509 6020 +78 515 6005 +79 519 5979 +80 525 5959 +81 530 5935 +82 534 5908 +83 539 5887 +84 545 5876 +85 555 5906 +86 562 5903 +87 568 5889 +88 575 5890 +89 580 5874 +90 585 5852 +91 592 5844 +92 597 5830 +93 602 5806 +94 607 5791 +95 613 5778 +96 618 5756 +97 623 5743 +98 629 5729 +99 634 5713 +100 640 5701 +101 645 5684 +102 650 5664 +103 655 5647 +104 661 5641 +105 667 5626 +106 672 5610 +107 677 5595 +108 682 5580 +109 687 5566 +110 693 5552 +111 699 5543 +112 704 5529 +113 709 5513 +114 714 5498 +115 719 5481 +116 725 5477 +117 731 5468 +118 737 5460 +119 744 5458 +120 751 5456 +121 758 5455 +122 763 5446 +123 774 5469 +124 781 5467 +125 792 5495 +126 797 5480 +127 804 5481 +128 809 5466 +129 814 5448 +130 819 5437 +131 825 5429 +132 830 5416 +133 838 5418 +134 843 5403 +135 848 5392 +136 854 5384 +137 860 5375 +138 867 5370 +139 872 5357 diff --git a/daily_accuracy_analysis.py b/daily_accuracy_analysis.py new file mode 100644 index 0000000..88ed88e --- /dev/null +++ b/daily_accuracy_analysis.py @@ -0,0 +1,460 @@ +"""按天统计双模型和三模型的准确率""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def main(): + print("="*80) + print("按天统计：双模型交集 vs 三模型投票 每日准确率分析") + print("="*80) + + # ========== 加载模型 ========== + print("\n[1/4] 加载模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + # 加载DLinear + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + # 加载N-BEATS + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + # 加载TTM-R2 + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + print(f"✓ 三个模型已加载") + + # ========== 获取数据 ========== + print("\n[2/4] 获取2026年测试数据...") + fetcher = BinanceDataFetcher() + + try: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2025-12-26', + end_date='2026-02-02', use_cache=False) + except: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2025-12-01', use_cache=False) + + print(f"✓ 测试数据: {len(df_2026)} 行") + + # ========== 预处理 ========== + print("\n[3/4] 预处理数据...") + + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + y_nbeats = y_nbeats[:min_samples] + + # ========== 预测 ========== + print("\n[4/4] 执行预测...") + + # DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits)) + dlinear_preds = (dlinear_probs > 0.5).astype(int).flatten() + + # N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) + + # TTM-R2预测 + context_length = 512 + ttm_preds_list = [] + all_prices = df_2026['close'].values + all_open_prices = df_2026['open'].values + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + context_data = all_prices[-context_length:].reshape(1, -1, 1) + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + + with torch.no_grad(): + output = ttm_model(context_tensor) + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + print(f"✓ 预测完成") + + # ========== 构建结果DataFrame ========== + # 获取对应的时间戳（第i个样本对应df_2026的第168+i根K线） + timestamps = df_2026['timestamp'].iloc[168:168+min_samples].values + + # 转换为美国东部时间 (ET) + timestamps_utc = pd.to_datetime(timestamps) + timestamps_et = timestamps_utc.tz_localize('UTC').tz_convert('America/New_York') + dates = timestamps_et.date + + # 双模型交集 + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_preds = np.full(len(dlinear_preds), -1) + two_model_preds[two_model_agree_up] = 1 + two_model_preds[two_model_agree_down] = 0 + + # 三模型多数投票 + vote_sum = dlinear_preds + nbeats_preds + ttm_preds + three_model_preds = np.where(vote_sum >= 2, 1, 0) + + # 真实标签 + true_labels = y_dlinear + + # 创建DataFrame + results_df = pd.DataFrame({ + 'date': dates, + 'timestamp': timestamps, + 'timestamp_et': timestamps_et, + 'true_label': true_labels, + 'dlinear_pred': dlinear_preds, + 'nbeats_pred': nbeats_preds, + 'ttm_pred': ttm_preds, + 'two_model_pred': two_model_preds, + 'three_model_pred': three_model_preds, + 'two_model_correct': (two_model_preds == true_labels) & (two_model_preds != -1), + 'three_model_correct': (three_model_preds == true_labels) + }) + + # 添加周几信息（美国东部时间） + results_df['day_of_week'] = pd.to_datetime(results_df['timestamp_et']).dt.day_name() + results_df['is_weekend'] = pd.to_datetime(results_df['timestamp_et']).dt.dayofweek >= 5 # 5=Saturday, 6=Sunday + + # 2026年1月美国重要经济数据发布日期（美国东部时间） + # 数据来源：美联储、劳工部等官方日历 + us_economic_events = { + '2026-01-08': ['初请失业金人数'], + '2026-01-09': ['非农就业报告 (NFP)', '失业率'], + '2026-01-13': ['CPI通胀数据'], + '2026-01-14': ['PPI生产者物价指数'], + '2026-01-15': ['零售销售数据'], + '2026-01-16': ['密歇根大学消费者信心指数'], + '2026-01-22': ['初请失业金人数'], + '2026-01-27': ['耐用品订单'], + '2026-01-28': ['FOMC利率决议', 'GDP初值'], + '2026-01-29': ['PCE物价指数', '个人收入与支出'], + } + + # ========== 按天统计 ========== + print("\n" + "="*80) + print("📊 每日准确率统计 (美国东部时间)") + print("="*80) + + daily_stats = results_df.groupby('date').agg({ + 'true_label': 'count', # 总样本数 + 'two_model_pred': lambda x: (x != -1).sum(), # 双模型有效预测数 + 'two_model_correct': 'sum', # 双模型正确数 + 'three_model_correct': 'sum', # 三模型正确数 + 'is_weekend': 'first', # 是否周末 + 'day_of_week': 'first' # 星期几 + }) + + daily_stats.columns = ['total_samples', 'two_model_valid', 'two_model_correct', + 'three_model_correct', 'is_weekend', 'day_of_week'] + + # 计算准确率 + daily_stats['two_model_accuracy'] = daily_stats['two_model_correct'] / daily_stats['two_model_valid'] * 100 + daily_stats['three_model_accuracy'] = daily_stats['three_model_correct'] / daily_stats['total_samples'] * 100 + daily_stats['two_model_coverage'] = daily_stats['two_model_valid'] / daily_stats['total_samples'] * 100 + + # 填充NaN（当天没有有效预测） + daily_stats['two_model_accuracy'] = daily_stats['two_model_accuracy'].fillna(0) + + # 添加美国经济数据发布标记 + daily_stats['us_events'] = daily_stats.index.map( + lambda x: us_economic_events.get(str(x), []) + ) + daily_stats['has_us_data'] = daily_stats['us_events'].apply(lambda x: len(x) > 0) + + print(f"\n每日详细统计：") + print(f"{'日期':<12} {'星期':<10} {'周末':<4} {'美国数据':<4} {'样本':>4} " + f"{'双模型%':>8} {'覆盖%':>6} {'三模型%':>8} {'差异':>7}") + print("-" * 95) + + for date, row in daily_stats.iterrows(): + diff = row['two_model_accuracy'] - row['three_model_accuracy'] + diff_str = f"{diff:+.1f}" if row['two_model_valid'] > 0 else "N/A" + + weekend_mark = "✓" if row['is_weekend'] else "" + us_data_mark = "📊" if row['has_us_data'] else "" + + # 根据准确率设置颜色标记 + two_acc = row['two_model_accuracy'] + three_acc = row['three_model_accuracy'] + + two_mark = "🔥" if two_acc >= 70 else ("✅" if two_acc >= 60 else ("⚠️" if two_acc < 40 else "")) + three_mark = "🔥" if three_acc >= 70 else ("✅" if three_acc >= 60 else ("⚠️" if three_acc < 40 else "")) + + print(f"{date} {row['day_of_week'][:3]:>8} {weekend_mark:>4} {us_data_mark:>4} " + f"{row['total_samples']:>4.0f} " + f"{row['two_model_accuracy']:>7.1f}{two_mark} " + f"{row['two_model_coverage']:>5.1f} " + f"{row['three_model_accuracy']:>7.1f}{three_mark} " + f"{diff_str:>7}") + + # 如果有美国经济数据发布，显示详情 + if row['has_us_data']: + events_str = ", ".join(row['us_events']) + print(f" └─ 📊 {events_str}") + + # ========== 周末 vs 工作日分析 ========== + print("\n" + "="*80) + print("📅 周末 vs 工作日准确率对比") + print("="*80) + + weekend_stats = daily_stats[daily_stats['is_weekend']] + weekday_stats = daily_stats[~daily_stats['is_weekend']] + + if len(weekend_stats) > 0: + print(f"\n🌴 周末 ({len(weekend_stats)}天):") + print(f" 双模型平均准确率: {weekend_stats['two_model_accuracy'].mean():.2f}%") + print(f" 三模型平均准确率: {weekend_stats['three_model_accuracy'].mean():.2f}%") + print(f" 平均覆盖率: {weekend_stats['two_model_coverage'].mean():.2f}%") + + if len(weekday_stats) > 0: + print(f"\n💼 工作日 ({len(weekday_stats)}天):") + print(f" 双模型平均准确率: {weekday_stats['two_model_accuracy'].mean():.2f}%") + print(f" 三模型平均准确率: {weekday_stats['three_model_accuracy'].mean():.2f}%") + print(f" 平均覆盖率: {weekday_stats['two_model_coverage'].mean():.2f}%") + + # ========== 美国数据发布日 vs 普通日分析 ========== + print("\n" + "="*80) + print("📊 美国经济数据发布日 vs 普通日准确率对比") + print("="*80) + + us_data_days = daily_stats[daily_stats['has_us_data']] + normal_days = daily_stats[~daily_stats['has_us_data']] + + if len(us_data_days) > 0: + print(f"\n📊 美国数据发布日 ({len(us_data_days)}天):") + print(f" 双模型平均准确率: {us_data_days['two_model_accuracy'].mean():.2f}%") + print(f" 三模型平均准确率: {us_data_days['three_model_accuracy'].mean():.2f}%") + print(f" 平均覆盖率: {us_data_days['two_model_coverage'].mean():.2f}%") + print(f"\n 具体日期:") + for date in us_data_days.index: + events = ", ".join(us_data_days.loc[date, 'us_events']) + two_acc = us_data_days.loc[date, 'two_model_accuracy'] + three_acc = us_data_days.loc[date, 'three_model_accuracy'] + print(f" {date}: 双{two_acc:.1f}% / 三{three_acc:.1f}% - {events}") + + if len(normal_days) > 0: + print(f"\n📅 普通日 ({len(normal_days)}天):") + print(f" 双模型平均准确率: {normal_days['two_model_accuracy'].mean():.2f}%") + print(f" 三模型平均准确率: {normal_days['three_model_accuracy'].mean():.2f}%") + print(f" 平均覆盖率: {normal_days['two_model_coverage'].mean():.2f}%") + + # ========== 分类统计 ========== + print("\n" + "="*80) + print("📈 准确率分类统计") + print("="*80) + + # 双模型交集分类 + two_model_high = daily_stats[daily_stats['two_model_accuracy'] >= 60] + two_model_medium = daily_stats[(daily_stats['two_model_accuracy'] >= 45) & + (daily_stats['two_model_accuracy'] < 60)] + two_model_low = daily_stats[(daily_stats['two_model_accuracy'] > 0) & + (daily_stats['two_model_accuracy'] < 45)] + + # 三模型投票分类 + three_model_high = daily_stats[daily_stats['three_model_accuracy'] >= 60] + three_model_medium = daily_stats[(daily_stats['three_model_accuracy'] >= 45) & + (daily_stats['three_model_accuracy'] < 60)] + three_model_low = daily_stats[daily_stats['three_model_accuracy'] < 45] + + print(f"\n🔵 双模型交集：") + print(f" 高准确率 (>=60%): {len(two_model_high)}天") + if len(two_model_high) > 0: + print(f" 日期: {', '.join([str(d) for d in two_model_high.index])}") + print(f" 平均准确率: {two_model_high['two_model_accuracy'].mean():.1f}%") + + print(f"\n 中等准确率 (45-60%): {len(two_model_medium)}天") + if len(two_model_medium) > 0: + print(f" 日期: {', '.join([str(d) for d in two_model_medium.index])}") + print(f" 平均准确率: {two_model_medium['two_model_accuracy'].mean():.1f}%") + + print(f"\n 低准确率 (<45%): {len(two_model_low)}天") + if len(two_model_low) > 0: + print(f" 日期: {', '.join([str(d) for d in two_model_low.index])}") + print(f" 平均准确率: {two_model_low['two_model_accuracy'].mean():.1f}%") + + print(f"\n🟢 三模型投票：") + print(f" 高准确率 (>=60%): {len(three_model_high)}天") + if len(three_model_high) > 0: + print(f" 日期: {', '.join([str(d) for d in three_model_high.index])}") + print(f" 平均准确率: {three_model_high['three_model_accuracy'].mean():.1f}%") + + print(f"\n 中等准确率 (45-60%): {len(three_model_medium)}天") + if len(three_model_medium) > 0: + print(f" 日期: {', '.join([str(d) for d in three_model_medium.index])}") + print(f" 平均准确率: {three_model_medium['three_model_accuracy'].mean():.1f}%") + + print(f"\n 低准确率 (<45%): {len(three_model_low)}天") + if len(three_model_low) > 0: + print(f" 日期: {', '.join([str(d) for d in three_model_low.index])}") + print(f" 平均准确率: {three_model_low['three_model_accuracy'].mean():.1f}%") + + # ========== 总结 ========== + print("\n" + "="*80) + print("📊 总体统计") + print("="*80) + + print(f"\n双模型交集：") + print(f" 总体准确率: {(daily_stats['two_model_correct'].sum() / daily_stats['two_model_valid'].sum() * 100):.2f}%") + print(f" 总体覆盖率: {(daily_stats['two_model_valid'].sum() / daily_stats['total_samples'].sum() * 100):.2f}%") + print(f" 最高日准确率: {daily_stats['two_model_accuracy'].max():.1f}% ({daily_stats['two_model_accuracy'].idxmax()})") + print(f" 最低日准确率: {daily_stats[daily_stats['two_model_accuracy'] > 0]['two_model_accuracy'].min():.1f}% ({daily_stats[daily_stats['two_model_accuracy'] > 0]['two_model_accuracy'].idxmin()})") + + print(f"\n三模型投票：") + print(f" 总体准确率: {(daily_stats['three_model_correct'].sum() / daily_stats['total_samples'].sum() * 100):.2f}%") + print(f" 覆盖率: 100%") + print(f" 最高日准确率: {daily_stats['three_model_accuracy'].max():.1f}% ({daily_stats['three_model_accuracy'].idxmax()})") + print(f" 最低日准确率: {daily_stats['three_model_accuracy'].min():.1f}% ({daily_stats['three_model_accuracy'].idxmin()})") + + print("\n" + "="*80) + print("✓ 每日准确率统计完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/dlinear_prob_distribution.py b/dlinear_prob_distribution.py new file mode 100644 index 0000000..60f3e35 --- /dev/null +++ b/dlinear_prob_distribution.py @@ -0,0 +1,406 @@ +"""分析DLinear输出概率值的偏向性及其对集成策略的影响""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def main(): + print("="*80) + print("DLinear输出概率值分布分析 - 以0.1阶梯统计") + print("="*80) + + # ========== 加载模型 ========== + print("\n[1/4] 加载模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + # 加载DLinear + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + # 加载N-BEATS + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + # 加载TTM-R2 + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + print(f"✓ 三个模型已加载") + + # ========== 获取数据 ========== + print("\n[2/4] 获取2026年测试数据...") + fetcher = BinanceDataFetcher() + + try: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', + end_date='2026-12-31', use_cache=False) + except: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2025-12-01', use_cache=False) + + print(f"✓ 测试数据: {len(df_2026)} 行") + + # ========== 预处理 ========== + print("\n[3/4] 预处理数据...") + + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + y_nbeats = y_nbeats[:min_samples] + + # ========== 预测并获取DLinear概率值 ========== + print("\n[4/4] 执行预测并分析DLinear输出概率...") + + # DLinear预测（保留概率值） + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits.flatten())) # sigmoid概率 + dlinear_preds = (dlinear_probs > 0.5).astype(int) + + # N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) + + # TTM-R2预测 + context_length = 512 + ttm_preds_list = [] + all_prices = df_2026['close'].values + all_open_prices = df_2026['open'].values + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + context_data = all_prices[-context_length:].reshape(1, -1, 1) + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + + with torch.no_grad(): + output = ttm_model(context_tensor) + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + # 双模型交集 + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_valid = two_model_agree_up | two_model_agree_down + + # 三模型多数投票 + vote_sum = dlinear_preds + nbeats_preds + ttm_preds + three_model_preds = np.where(vote_sum >= 2, 1, 0) + + # 真实标签 + true_labels = y_dlinear + + print(f"✓ 预测完成，开始分析DLinear概率分布...") + + # ========== 按0.1阶梯分析 ========== + print("\n" + "="*80) + print("📊 DLinear输出概率分布分析（0.1阶梯）") + print("="*80) + + # 定义概率区间 + prob_bins = np.arange(0, 1.1, 0.1) + prob_labels = [f"{i/10:.1f}-{(i+1)/10:.1f}" for i in range(10)] + + # 为每个样本分配到概率区间 + prob_bin_indices = np.digitize(dlinear_probs, prob_bins) - 1 + prob_bin_indices = np.clip(prob_bin_indices, 0, 9) # 确保在0-9范围内 + + # 初始化统计 + stats = [] + + for i, bin_label in enumerate(prob_labels): + mask = prob_bin_indices == i + count = mask.sum() + + if count == 0: + continue + + # 该区间的真实标签 + bin_true = true_labels[mask] + + # 该区间DLinear的预测 + bin_dlinear_pred = dlinear_preds[mask] + + # 该区间在双模型交集中的表现 + bin_two_model_valid = two_model_valid[mask] + bin_two_model_pred = np.full(count, -1) + bin_two_model_pred[two_model_agree_up[mask]] = 1 + bin_two_model_pred[two_model_agree_down[mask]] = 0 + + # 该区间在三模型投票中的表现 + bin_three_model_pred = three_model_preds[mask] + + # 计算统计 + dlinear_acc = (bin_dlinear_pred == bin_true).mean() * 100 if count > 0 else 0 + + two_model_valid_count = bin_two_model_valid.sum() + two_model_acc = ((bin_two_model_pred == bin_true) & (bin_two_model_pred != -1)).sum() / two_model_valid_count * 100 if two_model_valid_count > 0 else 0 + two_model_coverage = two_model_valid_count / count * 100 + + three_model_acc = (bin_three_model_pred == bin_true).mean() * 100 if count > 0 else 0 + + # 平均概率值 + avg_prob = dlinear_probs[mask].mean() + + # 预测倾向 + pred_up = bin_dlinear_pred.mean() * 100 + + stats.append({ + 'bin': bin_label, + 'count': count, + 'avg_prob': avg_prob, + 'pred_up_pct': pred_up, + 'dlinear_acc': dlinear_acc, + 'two_model_acc': two_model_acc, + 'two_model_cov': two_model_coverage, + 'three_model_acc': three_model_acc + }) + + # 转换为DataFrame + stats_df = pd.DataFrame(stats) + + # 显示统计表 + print("\n概率区间分析：") + print(f"{'区间':<10} {'样本数':>6} {'平均概率':>8} {'预测↑%':>7} " + f"{'DL准确率':>9} {'双模准确率':>10} {'双模覆盖':>8} {'三模准确率':>10}") + print("-" * 95) + + for _, row in stats_df.iterrows(): + # 根据准确率添加标记 + dl_mark = "🔥" if row['dlinear_acc'] >= 70 else ("✅" if row['dlinear_acc'] >= 60 else ("⚠️" if row['dlinear_acc'] < 45 else "")) + two_mark = "🔥" if row['two_model_acc'] >= 70 else ("✅" if row['two_model_acc'] >= 60 else ("⚠️" if row['two_model_acc'] < 45 else "")) + three_mark = "🔥" if row['three_model_acc'] >= 70 else ("✅" if row['three_model_acc'] >= 60 else ("⚠️" if row['three_model_acc'] < 45 else "")) + + print(f"{row['bin']:<10} {row['count']:>6} {row['avg_prob']:>8.3f} {row['pred_up_pct']:>6.1f}% " + f"{row['dlinear_acc']:>8.1f}{dl_mark} " + f"{row['two_model_acc']:>9.1f}{two_mark} " + f"{row['two_model_cov']:>7.1f}% " + f"{row['three_model_acc']:>9.1f}{three_mark}") + + # ========== 关键发现 ========== + print("\n" + "="*80) + print("🔍 关键发现") + print("="*80) + + # 极端概率区间（高置信度） + high_conf_up = stats_df[stats_df['avg_prob'] >= 0.65] + high_conf_down = stats_df[stats_df['avg_prob'] <= 0.35] + + print(f"\n📈 高置信度看涨区间 (概率≥0.65):") + if len(high_conf_up) > 0: + total_samples = high_conf_up['count'].sum() + avg_dl_acc = (high_conf_up['dlinear_acc'] * high_conf_up['count']).sum() / total_samples + avg_two_acc = (high_conf_up['two_model_acc'] * high_conf_up['count']).sum() / total_samples + avg_three_acc = (high_conf_up['three_model_acc'] * high_conf_up['count']).sum() / total_samples + + print(f" 样本数: {total_samples} ({total_samples/len(dlinear_probs)*100:.1f}%)") + print(f" DLinear平均准确率: {avg_dl_acc:.2f}%") + print(f" 双模型平均准确率: {avg_two_acc:.2f}%") + print(f" 三模型平均准确率: {avg_three_acc:.2f}%") + else: + print(" 无数据") + + print(f"\n📉 高置信度看跌区间 (概率≤0.35):") + if len(high_conf_down) > 0: + total_samples = high_conf_down['count'].sum() + avg_dl_acc = (high_conf_down['dlinear_acc'] * high_conf_down['count']).sum() / total_samples + avg_two_acc = (high_conf_down['two_model_acc'] * high_conf_down['count']).sum() / total_samples + avg_three_acc = (high_conf_down['three_model_acc'] * high_conf_down['count']).sum() / total_samples + + print(f" 样本数: {total_samples} ({total_samples/len(dlinear_probs)*100:.1f}%)") + print(f" DLinear平均准确率: {avg_dl_acc:.2f}%") + print(f" 双模型平均准确率: {avg_two_acc:.2f}%") + print(f" 三模型平均准确率: {avg_three_acc:.2f}%") + else: + print(" 无数据") + + # 中等置信度区间 + mid_conf = stats_df[(stats_df['avg_prob'] > 0.45) & (stats_df['avg_prob'] < 0.55)] + + print(f"\n⚖️ 中等置信度区间 (0.45-0.55):") + if len(mid_conf) > 0: + total_samples = mid_conf['count'].sum() + avg_dl_acc = (mid_conf['dlinear_acc'] * mid_conf['count']).sum() / total_samples + avg_two_acc = (mid_conf['two_model_acc'] * mid_conf['count']).sum() / total_samples if total_samples > 0 else 0 + avg_three_acc = (mid_conf['three_model_acc'] * mid_conf['count']).sum() / total_samples + + print(f" 样本数: {total_samples} ({total_samples/len(dlinear_probs)*100:.1f}%)") + print(f" DLinear平均准确率: {avg_dl_acc:.2f}%") + print(f" 双模型平均准确率: {avg_two_acc:.2f}%") + print(f" 三模型平均准确率: {avg_three_acc:.2f}%") + print(f" ⚠️ 建议: 低置信度区间，考虑弃权或减仓") + else: + print(" 无数据") + + # 找出表现最好的概率区间 + best_bin = stats_df.loc[stats_df['dlinear_acc'].idxmax()] + worst_bin = stats_df.loc[stats_df['dlinear_acc'].idxmin()] + + print(f"\n🏆 DLinear表现最佳区间:") + print(f" 区间: {best_bin['bin']}") + print(f" 准确率: {best_bin['dlinear_acc']:.1f}%") + print(f" 样本数: {best_bin['count']}") + + print(f"\n💀 DLinear表现最差区间:") + print(f" 区间: {worst_bin['bin']}") + print(f" 准确率: {worst_bin['dlinear_acc']:.1f}%") + print(f" 样本数: {worst_bin['count']}") + + # ========== 概率分布可视化 ========== + print("\n" + "="*80) + print("📊 DLinear概率分布直方图") + print("="*80) + + print("\n样本数分布:") + max_count = stats_df['count'].max() + for _, row in stats_df.iterrows(): + bar_length = int(row['count'] / max_count * 40) + bar = "█" * bar_length + print(f"{row['bin']:<10} {bar} {row['count']}") + + print("\n" + "="*80) + print("✓ DLinear概率分布分析完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/ensemble_daily_accuracy_multi_periods.py b/ensemble_daily_accuracy_multi_periods.py new file mode 100644 index 0000000..9e80c64 --- /dev/null +++ b/ensemble_daily_accuracy_multi_periods.py @@ -0,0 +1,395 @@ +"""四模型集成每日准确率分析 - 多时间段测试（2019年1月、2020年1月、2026年1月）""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def predict_period(period_name, start_date, end_date, dlinear_model, nbeats_model, ttm_model, timesfm_model, + dlinear_preprocessor, nbeats_preprocessor, device): + """预测单个时间段的数据""" + + print(f"\n{'='*80}") + print(f"📅 {period_name} ({start_date} 至 {end_date})") + print(f"{'='*80}") + + # 获取数据 + fetcher = BinanceDataFetcher() + df = fetcher.fetch_ohlcv('BTC', '1h', start_date=start_date, end_date=end_date, use_cache=False) + + print(f"✓ 数据: {len(df)} 行") + + # 预处理 + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df, fit=False) + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df, fit=False) + + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + + print(f"✓ 样本数: {min_samples}") + + # DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits)) + dlinear_preds = (dlinear_probs > 0.5).astype(int).flatten() + + # N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) + + # TTM-R2预测 + context_length = 512 + ttm_preds_list = [] + all_prices = df['close'].values + all_open_prices = df['open'].values + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + context_data = all_prices[-context_length:].reshape(1, -1, 1) + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + + with torch.no_grad(): + output = ttm_model(context_tensor) + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + # TimesFM预测 + timesfm_preds_list = [] + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + 512 + + if end_idx > len(all_prices): + context_prices = all_prices[-512:] + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_prices = all_prices[start_idx:end_idx] + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + try: + point_forecast, _ = timesfm_model.forecast(horizon=1, inputs=[context_prices]) + pred_price = point_forecast[0, 0] + timesfm_preds_list.append(1 if pred_price > current_open_price else 0) + except: + timesfm_preds_list.append(np.random.randint(0, 2)) + + timesfm_preds = np.array(timesfm_preds_list) + + # 集成策略 + true_labels = y_dlinear + + # 双模型交集 + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_preds = np.full(len(dlinear_preds), -1) + two_model_preds[two_model_agree_up] = 1 + two_model_preds[two_model_agree_down] = 0 + + # 三模型投票 + vote_sum_3 = dlinear_preds + nbeats_preds + ttm_preds + three_model_preds = np.where(vote_sum_3 >= 2, 1, 0) + + # 四模型投票 + vote_sum_4 = dlinear_preds + nbeats_preds + ttm_preds + timesfm_preds + four_model_preds = np.where(vote_sum_4 >= 3, 1, 0) + + # 构建每日结果DataFrame + timestamps = df['timestamp'].iloc[168:168+min_samples].values + timestamps_utc = pd.to_datetime(timestamps) + timestamps_et = timestamps_utc.tz_localize('UTC').tz_convert('America/New_York') + dates = timestamps_et.date + + results_df = pd.DataFrame({ + 'date': dates, + 'timestamp_et': timestamps_et, + 'true_label': true_labels, + 'dlinear_pred': dlinear_preds, + 'nbeats_pred': nbeats_preds, + 'ttm_pred': ttm_preds, + 'timesfm_pred': timesfm_preds, + 'two_model_pred': two_model_preds, + 'three_model_pred': three_model_preds, + 'four_model_pred': four_model_preds, + }) + + # 按天统计 + daily_stats = results_df.groupby('date').agg({ + 'true_label': 'count', + 'dlinear_pred': lambda x: (x == results_df.loc[x.index, 'true_label']).sum(), + 'nbeats_pred': lambda x: (x == results_df.loc[x.index, 'true_label']).sum(), + 'ttm_pred': lambda x: (x == results_df.loc[x.index, 'true_label']).sum(), + 'timesfm_pred': lambda x: (x == results_df.loc[x.index, 'true_label']).sum(), + 'two_model_pred': lambda x: ((x == results_df.loc[x.index, 'true_label']) & (x != -1)).sum(), + 'three_model_pred': lambda x: (x == results_df.loc[x.index, 'true_label']).sum(), + 'four_model_pred': lambda x: (x == results_df.loc[x.index, 'true_label']).sum(), + }) + + daily_stats.columns = ['total', 'dlinear_correct', 'nbeats_correct', 'ttm_correct', + 'timesfm_correct', 'two_model_correct', 'three_model_correct', 'four_model_correct'] + + # 计算每天的有效预测数（双模型） + two_model_valid = results_df.groupby('date')['two_model_pred'].apply(lambda x: (x != -1).sum()) + daily_stats['two_model_valid'] = two_model_valid + + # 计算准确率 + daily_stats['dlinear_acc'] = daily_stats['dlinear_correct'] / daily_stats['total'] * 100 + daily_stats['nbeats_acc'] = daily_stats['nbeats_correct'] / daily_stats['total'] * 100 + daily_stats['ttm_acc'] = daily_stats['ttm_correct'] / daily_stats['total'] * 100 + daily_stats['timesfm_acc'] = daily_stats['timesfm_correct'] / daily_stats['total'] * 100 + daily_stats['two_model_acc'] = daily_stats['two_model_correct'] / daily_stats['two_model_valid'] * 100 + daily_stats['three_model_acc'] = daily_stats['three_model_correct'] / daily_stats['total'] * 100 + daily_stats['four_model_acc'] = daily_stats['four_model_correct'] / daily_stats['total'] * 100 + + daily_stats['two_model_acc'] = daily_stats['two_model_acc'].fillna(0) + + return daily_stats, results_df + + +def main(): + print("="*80) + print("四模型集成每日准确率分析 - 多时间段测试") + print("测试时段: 2019年1月、2020年1月、2026年1月") + print("="*80) + + # ========== 加载模型 ========== + print("\n[1/2] 加载模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device, weights_only=False + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device, weights_only=False + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + import timesfm + timesfm_model = timesfm.TimesFM_2p5_200M_torch.from_pretrained( + "google/timesfm-2.5-200m-pytorch", + torch_compile=False + ) + timesfm_model.compile( + timesfm.ForecastConfig( + max_context=512, + max_horizon=1, + normalize_inputs=True, + use_continuous_quantile_head=True, + force_flip_invariance=True, + infer_is_positive=True, + fix_quantile_crossing=True, + ) + ) + + print("✓ 四个模型已加载 (DLinear, N-BEATS, TTM-R2, TimesFM)") + + # ========== 多时间段预测 ========== + print("\n[2/2] 多时间段预测...") + + periods = [ + ("2019年1月", "2019-01-01", "2019-01-31"), + ("2020年1月", "2020-01-01", "2020-01-31"), + ("2026年1月", "2026-01-01", "2026-01-31"), + ] + + all_period_stats = {} + + for period_name, start_date, end_date in periods: + daily_stats, results_df = predict_period( + period_name, start_date, end_date, + dlinear_model, nbeats_model, ttm_model, timesfm_model, + dlinear_preprocessor, nbeats_preprocessor, device + ) + all_period_stats[period_name] = daily_stats + + # 显示每日准确率 + print(f"\n每日准确率明细:") + print(f"{'日期':<12} {'样本':>4} {'DLinear':>8} {'N-BEATS':>8} {'TTM-R2':>8} {'TimesFM':>8} {'双模型':>8} {'三模型':>8} {'四模型':>8}") + print("-" * 95) + + for date, row in daily_stats.iterrows(): + print(f"{date} {row['total']:>4.0f} " + f"{row['dlinear_acc']:>7.1f}% " + f"{row['nbeats_acc']:>7.1f}% " + f"{row['ttm_acc']:>7.1f}% " + f"{row['timesfm_acc']:>7.1f}% " + f"{row['two_model_acc']:>7.1f}% " + f"{row['three_model_acc']:>7.1f}% " + f"{row['four_model_acc']:>7.1f}%") + + # 统计总体 + print(f"\n{period_name} 总体统计:") + print(f" DLinear: {daily_stats['dlinear_correct'].sum() / daily_stats['total'].sum() * 100:.2f}%") + print(f" N-BEATS: {daily_stats['nbeats_correct'].sum() / daily_stats['total'].sum() * 100:.2f}%") + print(f" TTM-R2: {daily_stats['ttm_correct'].sum() / daily_stats['total'].sum() * 100:.2f}%") + print(f" TimesFM: {daily_stats['timesfm_correct'].sum() / daily_stats['total'].sum() * 100:.2f}%") + print(f" 双模型: {daily_stats['two_model_correct'].sum() / daily_stats['two_model_valid'].sum() * 100:.2f}% (覆盖率: {daily_stats['two_model_valid'].sum() / daily_stats['total'].sum() * 100:.1f}%)") + print(f" 三模型: {daily_stats['three_model_correct'].sum() / daily_stats['total'].sum() * 100:.2f}%") + print(f" 四模型: {daily_stats['four_model_correct'].sum() / daily_stats['total'].sum() * 100:.2f}%") + + # ========== 跨时期对比 ========== + print("\n" + "="*80) + print("📊 跨时期对比总结") + print("="*80) + + print(f"\n{'时期':<12} {'DLinear':>9} {'N-BEATS':>9} {'TTM-R2':>9} {'TimesFM':>9} {'双模型':>9} {'三模型':>9} {'四模型':>9}") + print("-" * 95) + + for period_name in ["2019年1月", "2020年1月", "2026年1月"]: + stats = all_period_stats[period_name] + print(f"{period_name:<12} " + f"{stats['dlinear_correct'].sum() / stats['total'].sum() * 100:>8.2f}% " + f"{stats['nbeats_correct'].sum() / stats['total'].sum() * 100:>8.2f}% " + f"{stats['ttm_correct'].sum() / stats['total'].sum() * 100:>8.2f}% " + f"{stats['timesfm_correct'].sum() / stats['total'].sum() * 100:>8.2f}% " + f"{stats['two_model_correct'].sum() / stats['two_model_valid'].sum() * 100:>8.2f}% " + f"{stats['three_model_correct'].sum() / stats['total'].sum() * 100:>8.2f}% " + f"{stats['four_model_correct'].sum() / stats['total'].sum() * 100:>8.2f}%") + + print("\n" + "="*80) + print("✓ 多时间段每日准确率分析完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/ensemble_five_models.py b/ensemble_five_models.py new file mode 100644 index 0000000..3d8e325 --- /dev/null +++ b/ensemble_five_models.py @@ -0,0 +1,454 @@ +"""五模型集成预测：DLinear + N-BEATS + TTM-R2 + TimesFM 2.5 + CatBoost""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from catboost import CatBoostClassifier +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def add_catboost_features(df): + """为CatBoost添加丰富的技术指标特征""" + df = df.copy() + + # 基础价格特征 + df['price_change'] = df['close'] - df['open'] + df['price_change_pct'] = (df['close'] - df['open']) / df['open'] + df['high_low_range'] = df['high'] - df['low'] + df['high_low_range_pct'] = (df['high'] - df['low']) / df['low'] + + # 移动平均线 + df['sma_7'] = ta.sma(df['close'], length=7) + df['sma_25'] = ta.sma(df['close'], length=25) + df['sma_99'] = ta.sma(df['close'], length=99) + df['ema_12'] = ta.ema(df['close'], length=12) + df['ema_26'] = ta.ema(df['close'], length=26) + + # RSI + df['rsi_14'] = ta.rsi(df['close'], length=14) + df['rsi_7'] = ta.rsi(df['close'], length=7) + + # MACD + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + + # 布林带 + bbands = ta.bbands(df['close'], length=20) + if bbands is not None and not bbands.empty: + df['bb_upper'] = bbands.iloc[:, 0] + df['bb_mid'] = bbands.iloc[:, 1] + df['bb_lower'] = bbands.iloc[:, 2] + df['bb_width'] = (df['bb_upper'] - df['bb_lower']) / df['bb_mid'] + df['bb_position'] = (df['close'] - df['bb_lower']) / (df['bb_upper'] - df['bb_lower']) + + # ATR + df['atr'] = ta.atr(df['high'], df['low'], df['close'], length=14) + + # 成交量特征 + df['volume_sma_7'] = ta.sma(df['volume'], length=7) + df['volume_ratio'] = df['volume'] / df['volume_sma_7'] + + # OBV + df['obv'] = ta.obv(df['close'], df['volume']) + + # Stochastic + stoch = ta.stoch(df['high'], df['low'], df['close']) + if stoch is not None and not stoch.empty: + df['stoch_k'] = stoch.iloc[:, 0] + df['stoch_d'] = stoch.iloc[:, 1] + + # ADX + adx = ta.adx(df['high'], df['low'], df['close']) + if adx is not None and not adx.empty: + df['adx'] = adx.iloc[:, 0] + + # 价格动量 + df['momentum_1'] = df['close'].pct_change(1) + df['momentum_3'] = df['close'].pct_change(3) + df['momentum_7'] = df['close'].pct_change(7) + + return df + + +def main(): + print("="*80) + print("五模型集成预测：DLinear + N-BEATS + TTM-R2 + TimesFM + CatBoost") + print("="*80) + + # ========== 步骤1: 加载PyTorch模型 ========== + print("\n[1/8] 加载DLinear和N-BEATS模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + # 加载DLinear + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + print(f"✓ DLinear已加载") + + # 加载N-BEATS + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + print(f"✓ N-BEATS已加载") + + # ========== 步骤2: 加载TTM-R2 ========== + print("\n[2/8] 加载TTM-R2零样本模型...") + + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + print(f"✓ TTM-R2已加载") + + # ========== 步骤3: 加载TimesFM 2.5 ========== + print("\n[3/8] 加载TimesFM 2.5模型...") + + import timesfm + + timesfm_model = timesfm.TimesFM_2p5_200M_torch.from_pretrained( + "google/timesfm-2.5-200m-pytorch", + torch_compile=False + ) + + timesfm_model.compile( + timesfm.ForecastConfig( + max_context=512, + max_horizon=1, + normalize_inputs=True, + use_continuous_quantile_head=True, + force_flip_invariance=True, + infer_is_positive=True, + fix_quantile_crossing=True, + ) + ) + + print(f"✓ TimesFM 2.5已加载") + + # ========== 步骤4: 加载CatBoost ========== + print("\n[4/8] 加载CatBoost模型...") + + catboost_model = CatBoostClassifier() + catboost_model.load_model('./models/checkpoints/catboost_classification_2026/catboost_model.cbm') + + with open('./models/checkpoints/catboost_classification_2026/metadata.pkl', 'rb') as f: + catboost_metadata = pickle.load(f) + catboost_feature_cols = catboost_metadata['feature_cols'] + + print(f"✓ CatBoost已加载 (特征数: {len(catboost_feature_cols)})") + + # ========== 步骤5: 获取测试数据 ========== + print("\n[5/8] 获取2026年1月测试数据...") + + fetcher = BinanceDataFetcher() + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', + end_date='2026-01-31', use_cache=False) + + print(f"✓ 数据获取完成: {len(df_2026)} 行") + + # ========== 步骤6: 预处理 ========== + print("\n[6/8] 预处理数据...") + + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + # CatBoost特征工程 + df_catboost = add_catboost_features(df_2026) + df_catboost = df_catboost.dropna().reset_index(drop=True) + X_catboost = df_catboost[catboost_feature_cols] + + # 确保样本数量一致 + min_samples = min(len(X_dlinear), len(X_nbeats), len(X_catboost)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + X_catboost = X_catboost.iloc[:min_samples] + + print(f"✓ 对齐样本数: {min_samples}") + + # ========== 步骤7: 五个模型分别预测 ========== + print("\n[7/8] 五个模型分别预测...") + + # 7.1 DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits)) + dlinear_preds = (dlinear_probs > 0.5).astype(int).flatten() + + print(f"✓ DLinear: 涨{(dlinear_preds==1).sum()} ({(dlinear_preds==1).sum()/len(dlinear_preds)*100:.1f}%)") + + # 7.2 N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) + + print(f"✓ N-BEATS: 涨{(nbeats_preds==1).sum()} ({(nbeats_preds==1).sum()/len(nbeats_preds)*100:.1f}%)") + + # 7.3 TTM-R2预测 + context_length = 512 + ttm_preds_list = [] + all_prices = df_2026['close'].values + all_open_prices = df_2026['open'].values + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + context_data = all_prices[-context_length:].reshape(1, -1, 1) + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + + with torch.no_grad(): + output = ttm_model(context_tensor) + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + print(f"✓ TTM-R2: 涨{(ttm_preds==1).sum()} ({(ttm_preds==1).sum()/len(ttm_preds)*100:.1f}%)") + + # 7.4 TimesFM 2.5预测 + timesfm_preds_list = [] + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + 512 + + if end_idx > len(all_prices): + context_prices = all_prices[-512:] + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_prices = all_prices[start_idx:end_idx] + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + try: + point_forecast, _ = timesfm_model.forecast( + horizon=1, + inputs=[context_prices], + ) + pred_price = point_forecast[0, 0] + timesfm_preds_list.append(1 if pred_price > current_open_price else 0) + except: + timesfm_preds_list.append(np.random.randint(0, 2)) + + timesfm_preds = np.array(timesfm_preds_list) + + print(f"✓ TimesFM: 涨{(timesfm_preds==1).sum()} ({(timesfm_preds==1).sum()/len(timesfm_preds)*100:.1f}%)") + + # 7.5 CatBoost预测 + catboost_preds = catboost_model.predict(X_catboost).astype(int) + + print(f"✓ CatBoost: 涨{(catboost_preds==1).sum()} ({(catboost_preds==1).sum()/len(catboost_preds)*100:.1f}%)") + + # ========== 步骤8: 集成策略评估 ========== + print("\n[8/8] 集成策略评估...") + + true_labels = y_dlinear + + # 策略1: 双模型交集（DLinear + N-BEATS） + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_preds = np.full(len(dlinear_preds), -1) + two_model_preds[two_model_agree_up] = 1 + two_model_preds[two_model_agree_down] = 0 + + two_model_valid = two_model_preds != -1 + two_model_correct = (two_model_preds == true_labels) & two_model_valid + two_model_acc = two_model_correct.sum() / two_model_valid.sum() * 100 if two_model_valid.sum() > 0 else 0 + two_model_coverage = two_model_valid.sum() / len(true_labels) * 100 + + # 策略2: 三模型投票（DLinear + N-BEATS + TTM-R2） + vote_sum_3 = dlinear_preds + nbeats_preds + ttm_preds + three_model_preds = np.where(vote_sum_3 >= 2, 1, 0) + three_model_acc = (three_model_preds == true_labels).mean() * 100 + + # 策略3: 四模型投票 + vote_sum_4 = dlinear_preds + nbeats_preds + ttm_preds + timesfm_preds + four_model_preds = np.where(vote_sum_4 >= 3, 1, 0) + four_model_acc = (four_model_preds == true_labels).mean() * 100 + + # 策略4: 五模型强投票（≥3票） + vote_sum_5 = dlinear_preds + nbeats_preds + ttm_preds + timesfm_preds + catboost_preds + five_model_preds_strong = np.where(vote_sum_5 >= 3, 1, 0) + five_model_acc_strong = (five_model_preds_strong == true_labels).mean() * 100 + + # 策略5: 五模型弱投票（≥2票） + five_model_preds_weak = np.where(vote_sum_5 >= 2, 1, 0) + five_model_acc_weak = (five_model_preds_weak == true_labels).mean() * 100 + + # ========== 结果展示 ========== + print("\n" + "="*80) + print("📊 五模型集成策略对比结果") + print("="*80) + + print(f"\n单模型性能：") + print(f" DLinear: {(dlinear_preds == true_labels).mean() * 100:.2f}%") + print(f" CatBoost: {(catboost_preds == true_labels).mean() * 100:.2f}%") + print(f" TimesFM 2.5: {(timesfm_preds == true_labels).mean() * 100:.2f}%") + print(f" TTM-R2: {(ttm_preds == true_labels).mean() * 100:.2f}%") + print(f" N-BEATS: {(nbeats_preds == true_labels).mean() * 100:.2f}%") + + print(f"\n集成策略性能：") + print(f" 双模型交集 (DL+NB): {two_model_acc:.2f}% (覆盖率: {two_model_coverage:.2f}%)") + print(f" 三模型投票 (DL+NB+TTM, ≥2票): {three_model_acc:.2f}% (覆盖率: 100%)") + print(f" 四模型投票 (前4, ≥3票): {four_model_acc:.2f}% (覆盖率: 100%)") + print(f" 五模型强投票 (所有, ≥3票): {five_model_acc_strong:.2f}% (覆盖率: 100%)") + print(f" 五模型弱投票 (所有, ≥2票): {five_model_acc_weak:.2f}% (覆盖率: 100%)") + + # 投票分布分析 + print(f"\n五模型投票分布：") + for votes in range(6): + count = (vote_sum_5 == votes).sum() + pct = count / len(vote_sum_5) * 100 + bars = "█" * int(pct / 2) + print(f" {votes}票同意: {count:3d}次 ({pct:5.1f}%) {bars}") + + # 模型一致性分析 + print(f"\n模型一致性分析：") + all_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) & (ttm_preds == 1) & (timesfm_preds == 1) & (catboost_preds == 1) + all_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) & (ttm_preds == 0) & (timesfm_preds == 0) & (catboost_preds == 0) + all_agree = all_agree_up | all_agree_down + + print(f" 五模型完全一致: {all_agree.sum()}次 ({all_agree.sum()/len(all_agree)*100:.1f}%)") + if all_agree.sum() > 0: + all_agree_acc = ((five_model_preds_strong[all_agree] == true_labels[all_agree]).mean() * 100) + print(f" 一致时准确率: {all_agree_acc:.2f}%") + + print("\n" + "="*80) + print("✓ 五模型集成预测完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/ensemble_four_models.py b/ensemble_four_models.py new file mode 100644 index 0000000..34c98ba --- /dev/null +++ b/ensemble_four_models.py @@ -0,0 +1,385 @@ +"""四模型集成预测：DLinear + N-BEATS + TTM-R2 + TimesFM 2.5""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def main(): + print("="*80) + print("四模型集成预测：DLinear + N-BEATS + TTM-R2 + TimesFM 2.5") + print("="*80) + + # ========== 步骤1: 加载DLinear和N-BEATS模型 ========== + print("\n[1/7] 加载DLinear和N-BEATS模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + # 加载DLinear + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + print(f"✓ DLinear已加载") + + # 加载N-BEATS + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + print(f"✓ N-BEATS已加载") + + # ========== 步骤2: 加载TTM-R2 ========== + print("\n[2/7] 加载TTM-R2零样本模型...") + + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + print(f"✓ TTM-R2已加载 (参数量: {sum(p.numel() for p in ttm_model.parameters())})") + + # ========== 步骤3: 加载TimesFM 2.5 ========== + print("\n[3/7] 加载TimesFM 2.5模型...") + + import timesfm + + timesfm_model = timesfm.TimesFM_2p5_200M_torch.from_pretrained( + "google/timesfm-2.5-200m-pytorch", + torch_compile=False # 推理时不使用torch.compile + ) + + # 配置TimesFM用于BTC 1小时预测 + timesfm_model.compile( + timesfm.ForecastConfig( + max_context=512, # 使用512小时历史数据 + max_horizon=1, # 预测未来1小时 + normalize_inputs=True, + use_continuous_quantile_head=True, + force_flip_invariance=True, + infer_is_positive=True, + fix_quantile_crossing=True, + ) + ) + + print(f"✓ TimesFM 2.5已加载并配置") + print(f" 上下文长度: 512小时") + print(f" 预测长度: 1小时") + + # ========== 步骤4: 获取测试数据 ========== + print("\n[4/7] 获取2026年1月测试数据...") + + fetcher = BinanceDataFetcher() + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', + end_date='2026-01-30', use_cache=False) + + print(f"✓ 数据获取完成: {len(df_2026)} 行") + print(f"✓ 时间范围: {df_2026['timestamp'].min()} 至 {df_2026['timestamp'].max()}") + + # ========== 步骤5: 预处理 ========== + print("\n[5/7] 预处理数据...") + + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + # 确保样本数量一致 + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + y_nbeats = y_nbeats[:min_samples] + + print(f"✓ 对齐样本数: {min_samples}") + + # ========== 步骤6: 四个模型分别预测 ========== + print("\n[6/7] 四个模型分别预测...") + + # 6.1 DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits)) # sigmoid + dlinear_preds = (dlinear_probs > 0.5).astype(int).flatten() # 1=涨, 0=跌 + + print(f"✓ DLinear预测完成") + print(f" 预测上涨: {(dlinear_preds==1).sum()} ({(dlinear_preds==1).sum()/len(dlinear_preds)*100:.1f}%)") + print(f" 预测下跌: {(dlinear_preds==0).sum()} ({(dlinear_preds==0).sum()/len(dlinear_preds)*100:.1f}%)") + + # 6.2 N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + # 还原真实价格并使用当前K线open价格判断 + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) # 1=涨, 0=跌 + + print(f"✓ N-BEATS预测完成 (使用当前K线open价格判断)") + print(f" 预测上涨: {(nbeats_preds==1).sum()} ({(nbeats_preds==1).sum()/len(nbeats_preds)*100:.1f}%)") + print(f" 预测下跌: {(nbeats_preds==0).sum()} ({(nbeats_preds==0).sum()/len(nbeats_preds)*100:.1f}%)") + + # 6.3 TTM-R2预测 + print(f"✓ TTM-R2预测中...") + + context_length = 512 + ttm_preds_list = [] + all_prices = df_2026['close'].values + all_open_prices = df_2026['open'].values + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + context_data = all_prices[-context_length:].reshape(1, -1, 1) + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + + with torch.no_grad(): + output = ttm_model(context_tensor) + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + print(f"✓ TTM-R2预测完成") + print(f" 预测上涨: {(ttm_preds==1).sum()} ({(ttm_preds==1).sum()/len(ttm_preds)*100:.1f}%)") + print(f" 预测下跌: {(ttm_preds==0).sum()} ({(ttm_preds==0).sum()/len(ttm_preds)*100:.1f}%)") + + # 6.4 TimesFM 2.5预测 + print(f"✓ TimesFM 2.5预测中...") + + timesfm_preds_list = [] + + for i in range(min_samples): + # TimesFM使用512小时上下文 + start_idx = i + end_idx = start_idx + 512 + + if end_idx > len(all_prices): + # 如果数据不足，使用最后512个数据点 + context_prices = all_prices[-512:] + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_prices = all_prices[start_idx:end_idx] + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + # TimesFM的forecast方法接受list of arrays作为inputs + try: + point_forecast, _ = timesfm_model.forecast( + horizon=1, + inputs=[context_prices], # list包含单个序列 + ) + + # point_forecast形状: (1, 1) - (num_series, horizon) + pred_price = point_forecast[0, 0] + + # 与当前K线open价格比较 + timesfm_preds_list.append(1 if pred_price > current_open_price else 0) + + except Exception as e: + print(f" ⚠️ TimesFM预测第{i}个样本失败: {e}") + # 使用随机预测作为fallback + timesfm_preds_list.append(np.random.randint(0, 2)) + + timesfm_preds = np.array(timesfm_preds_list) + + print(f"✓ TimesFM 2.5预测完成") + print(f" 预测上涨: {(timesfm_preds==1).sum()} ({(timesfm_preds==1).sum()/len(timesfm_preds)*100:.1f}%)") + print(f" 预测下跌: {(timesfm_preds==0).sum()} ({(timesfm_preds==0).sum()/len(timesfm_preds)*100:.1f}%)") + + # ========== 步骤7: 集成策略评估 ========== + print("\n[7/7] 集成策略评估...") + + true_labels = y_dlinear + + # 策略1: 双模型交集（DLinear + N-BEATS） + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_preds = np.full(len(dlinear_preds), -1) + two_model_preds[two_model_agree_up] = 1 + two_model_preds[two_model_agree_down] = 0 + + two_model_valid = two_model_preds != -1 + two_model_correct = (two_model_preds == true_labels) & two_model_valid + two_model_acc = two_model_correct.sum() / two_model_valid.sum() * 100 if two_model_valid.sum() > 0 else 0 + two_model_coverage = two_model_valid.sum() / len(true_labels) * 100 + + # 策略2: 三模型多数投票（DLinear + N-BEATS + TTM-R2） + vote_sum_3 = dlinear_preds + nbeats_preds + ttm_preds + three_model_preds = np.where(vote_sum_3 >= 2, 1, 0) + three_model_acc = (three_model_preds == true_labels).mean() * 100 + + # 策略3: 四模型多数投票（至少3个模型同意） + vote_sum_4 = dlinear_preds + nbeats_preds + ttm_preds + timesfm_preds + four_model_preds = np.where(vote_sum_4 >= 3, 1, 0) + four_model_acc = (four_model_preds == true_labels).mean() * 100 + + # 策略4: 四模型半数以上投票（至少2个模型同意） + four_model_preds_half = np.where(vote_sum_4 >= 2, 1, 0) + four_model_acc_half = (four_model_preds_half == true_labels).mean() * 100 + + # ========== 结果展示 ========== + print("\n" + "="*80) + print("📊 集成策略对比结果") + print("="*80) + + print(f"\n单模型性能：") + print(f" DLinear: {(dlinear_preds == true_labels).mean() * 100:.2f}%") + print(f" N-BEATS: {(nbeats_preds == true_labels).mean() * 100:.2f}%") + print(f" TTM-R2: {(ttm_preds == true_labels).mean() * 100:.2f}%") + print(f" TimesFM 2.5: {(timesfm_preds == true_labels).mean() * 100:.2f}%") + + print(f"\n集成策略性能：") + print(f" 双模型交集 (DL+NB): {two_model_acc:.2f}% (覆盖率: {two_model_coverage:.2f}%)") + print(f" 三模型投票 (DL+NB+TTM, ≥2票): {three_model_acc:.2f}% (覆盖率: 100%)") + print(f" 四模型强投票 (所有, ≥3票): {four_model_acc:.2f}% (覆盖率: 100%)") + print(f" 四模型弱投票 (所有, ≥2票): {four_model_acc_half:.2f}% (覆盖率: 100%)") + + # 投票分布分析 + print(f"\n四模型投票分布：") + for votes in range(5): + count = (vote_sum_4 == votes).sum() + pct = count / len(vote_sum_4) * 100 + bars = "█" * int(pct / 2) + print(f" {votes}票同意: {count:3d}次 ({pct:5.1f}%) {bars}") + + # 模型一致性分析 + print(f"\n模型一致性分析：") + all_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) & (ttm_preds == 1) & (timesfm_preds == 1) + all_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) & (ttm_preds == 0) & (timesfm_preds == 0) + all_agree = all_agree_up | all_agree_down + + print(f" 四模型完全一致: {all_agree.sum()}次 ({all_agree.sum()/len(all_agree)*100:.1f}%)") + if all_agree.sum() > 0: + all_agree_acc = ((four_model_preds[all_agree] == true_labels[all_agree]).mean() * 100) + print(f" 一致时准确率: {all_agree_acc:.2f}%") + + print("\n" + "="*80) + print("✓ 四模型集成预测完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/ensemble_predict.py b/ensemble_predict.py new file mode 100644 index 0000000..09be4d7 --- /dev/null +++ b/ensemble_predict.py @@ -0,0 +1,302 @@ +"""N-BEATS + DLinear 模型集成预测 - 交集策略""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from sklearn.metrics import confusion_matrix +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def main(): + print("="*80) + print("模型集成预测：N-BEATS + DLinear 交集策略") + print("="*80) + + # ========== 步骤1: 加载两个模型 ========== + print("\n[1/5] 加载两个模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + # 加载DLinear分类模型 + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + print(f"✓ DLinear分类模型已加载 (准确率: 53.10%)") + + # 加载N-BEATS回归模型 + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + print(f"✓ N-BEATS回归模型已加载 (准确率: 50.83%)") + + # ========== 步骤2: 获取2026年测试数据 ========== + print("\n[2/5] 获取2026年测试数据...") + fetcher = BinanceDataFetcher() + + try: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', + end_date='2026-12-31', use_cache=False) + except: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2025-12-01', use_cache=False) + + print(f"✓ 测试数据: {len(df_2026)} 行") + print(f"✓ 时间: {df_2026['timestamp'].min()} 至 {df_2026['timestamp'].max()}") + + # ========== 步骤3: 预处理数据 ========== + print("\n[3/5] 预处理数据...") + + # DLinear预处理 + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + + # N-BEATS预处理 + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + # 确保两个模型的样本数量一致 + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + y_nbeats = y_nbeats[:min_samples] + + print(f"✓ 对齐样本数: {min_samples}") + + # ========== 步骤4: 两个模型分别预测 ========== + print("\n[4/5] 两个模型分别预测...") + + # DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits)) # sigmoid + dlinear_preds = (dlinear_probs > 0.5).astype(int).flatten() # 1=涨, 0=跌 + + print(f"✓ DLinear预测完成") + print(f" 预测上涨: {(dlinear_preds==1).sum()} ({(dlinear_preds==1).sum()/len(dlinear_preds)*100:.1f}%)") + print(f" 预测下跌: {(dlinear_preds==0).sum()} ({(dlinear_preds==0).sum()/len(dlinear_preds)*100:.1f}%)") + + # N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + # 还原真实价格 + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + + # 获取当前价格（用于判断方向） + current_prices = df_nbeats['close'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_prices).astype(int) # 1=涨, 0=跌 + + print(f"✓ N-BEATS预测完成") + print(f" 预测上涨: {(nbeats_preds==1).sum()} ({(nbeats_preds==1).sum()/len(nbeats_preds)*100:.1f}%)") + print(f" 预测下跌: {(nbeats_preds==0).sum()} ({(nbeats_preds==0).sum()/len(nbeats_preds)*100:.1f}%)") + + # ========== 步骤5: 交集策略 ========== + print("\n[5/5] 应用交集策略...") + + # 真实标签 + true_labels = y_dlinear + + # 策略1: 两个模型都预测上涨 → 上涨 + # 策略2: 两个模型都预测下跌 → 下跌 + # 策略3: 两个模型预测不一致 → 不做决策（弃权） + + both_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + both_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + disagree = (dlinear_preds != nbeats_preds) + + n_agree_up = both_agree_up.sum() + n_agree_down = both_agree_down.sum() + n_disagree = disagree.sum() + + print(f"\n📊 预测一致性分析:") + print(f" 双方都预测上涨: {n_agree_up} ({n_agree_up/len(dlinear_preds)*100:.1f}%)") + print(f" 双方都预测下跌: {n_agree_down} ({n_agree_down/len(dlinear_preds)*100:.1f}%)") + print(f" 双方预测不一致: {n_disagree} ({n_disagree/len(dlinear_preds)*100:.1f}%)") + + # 只在两个模型一致时做决策 + agree_mask = both_agree_up | both_agree_down + ensemble_preds = np.full(len(dlinear_preds), -1) # -1表示弃权 + ensemble_preds[both_agree_up] = 1 + ensemble_preds[both_agree_down] = 0 + + # 有效预测（非弃权） + valid_mask = ensemble_preds != -1 + valid_preds = ensemble_preds[valid_mask] + valid_labels = true_labels[valid_mask] + + print(f"\n📊 集成策略结果:") + print(f" 有效预测数: {valid_mask.sum()} ({valid_mask.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" 弃权数: {n_disagree} ({n_disagree/len(dlinear_preds)*100:.1f}%)") + + if len(valid_preds) > 0: + ensemble_accuracy = (valid_preds == valid_labels).mean() + print(f"\n✨ 交集策略准确率: {ensemble_accuracy*100:.2f}%") + + # 混淆矩阵 + cm = confusion_matrix(valid_labels, valid_preds) + tn, fp, fn, tp = cm.ravel() + + print(f"\n详细分析（仅有效预测）:") + print(f" ✓ 预测涨且真涨: {tp} ({tp/len(valid_preds)*100:.1f}%)") + print(f" ✓ 预测跌且真跌: {tn} ({tn/len(valid_preds)*100:.1f}%)") + print(f" ✗ 预测涨但真跌: {fp} ({fp/len(valid_preds)*100:.1f}%)") + print(f" ✗ 预测跌但真涨: {fn} ({fn/len(valid_preds)*100:.1f}%)") + + # ========== 对比分析 ========== + print("\n" + "="*80) + print("📊 性能对比") + print("="*80) + + dlinear_acc = (dlinear_preds == true_labels).mean() + nbeats_acc = (nbeats_preds == true_labels).mean() + + print(f"\n单模型性能:") + print(f" DLinear分类: {dlinear_acc*100:.2f}%") + print(f" N-BEATS回归: {nbeats_acc*100:.2f}%") + + if len(valid_preds) > 0: + print(f"\n集成策略性能:") + print(f" 交集准确率: {ensemble_accuracy*100:.2f}%") + print(f" 覆盖率: {valid_mask.sum()/len(dlinear_preds)*100:.1f}%") + print(f" 弃权率: {n_disagree/len(dlinear_preds)*100:.1f}%") + + # 性能评估 + print("\n" + "="*80) + print("🎯 策略评估") + print("="*80) + + if ensemble_accuracy > max(dlinear_acc, nbeats_acc): + improvement = ensemble_accuracy - max(dlinear_acc, nbeats_acc) + print(f"\n ✅ 交集策略优于单模型!") + print(f" 准确率提升: {improvement*100:.2f}个百分点") + print(f" 代价: 弃权{n_disagree}次预测 ({n_disagree/len(dlinear_preds)*100:.1f}%)") + else: + print(f"\n ⚠️ 交集策略未能提升准确率") + print(f" 可能原因: 两个模型一致时也可能都错") + + # 计算如果弃权时不交易的盈利 + print(f"\n💰 实际收益分析:") + print(f" 如果只在两个模型都同意时交易:") + print(f" - 交易次数: {valid_mask.sum()} ({valid_mask.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" - 准确率: {ensemble_accuracy*100:.2f}%") + print(f" - 避免交易: {n_disagree}次 (规避不确定性)") + + print("\n" + "="*80) + print("✓ 集成预测分析完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/ensemble_three_models.py b/ensemble_three_models.py new file mode 100644 index 0000000..ae8903a --- /dev/null +++ b/ensemble_three_models.py @@ -0,0 +1,400 @@ +"""三模型集成预测 - DLinear + N-BEATS + TTM-R2 交集策略""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from sklearn.metrics import confusion_matrix +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def main(): + print("="*80) + print("三模型集成预测：DLinear + N-BEATS + TTM-R2 交集策略") + print("="*80) + + # ========== 步骤1: 加载三个模型 ========== + print("\n[1/6] 加载三个模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + # 1. 加载DLinear分类模型 + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + print(f"✓ DLinear分类模型已加载 (准确率: 53.10%)") + + # 2. 加载N-BEATS回归模型 + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + print(f"✓ N-BEATS回归模型已加载 (准确率: 50.83%)") + + # 3. 加载TTM-R2模型 + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + print(f"✓ TTM-R2模型已加载 (准确率: 37.50%)") + + # ========== 步骤2: 获取2026年测试数据 ========== + print("\n[2/6] 获取2026年测试数据...") + fetcher = BinanceDataFetcher() + + try: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', + end_date='2026-02-02', use_cache=False) + except: + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2025-12-01', use_cache=False) + + print(f"✓ 测试数据: {len(df_2026)} 行") + print(f"✓ 时间: {df_2026['timestamp'].min()} 至 {df_2026['timestamp'].max()}") + + # ========== 步骤3: 预处理数据 ========== + print("\n[3/6] 预处理数据...") + + # DLinear预处理 + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + + # N-BEATS预处理 + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + # 确保样本数量一致 + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + y_nbeats = y_nbeats[:min_samples] + + print(f"✓ 对齐样本数: {min_samples}") + + # ========== 步骤4: 三个模型分别预测 ========== + print("\n[4/6] 三个模型分别预测...") + + # 4.1 DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits)) # sigmoid + dlinear_preds = (dlinear_probs > 0.5).astype(int).flatten() # 1=涨, 0=跌 + + print(f"✓ DLinear预测完成") + print(f" 预测上涨: {(dlinear_preds==1).sum()} ({(dlinear_preds==1).sum()/len(dlinear_preds)*100:.1f}%)") + print(f" 预测下跌: {(dlinear_preds==0).sum()} ({(dlinear_preds==0).sum()/len(dlinear_preds)*100:.1f}%)") + + # 4.2 N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + # 还原真实价格 + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + + # 【修改】获取当前K线的open价格（未来可交易价格）用于判断方向 + # 第i个样本对应的是第168+i根K线，其open价格是真正可以交易的价格 + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) # 1=涨, 0=跌 + + print(f"✓ N-BEATS预测完成 (使用当前K线open价格判断)") + print(f" 预测上涨: {(nbeats_preds==1).sum()} ({(nbeats_preds==1).sum()/len(nbeats_preds)*100:.1f}%)") + print(f" 预测下跌: {(nbeats_preds==0).sum()} ({(nbeats_preds==0).sum()/len(nbeats_preds)*100:.1f}%)") + + # 4.3 TTM-R2预测 + print(f"✓ TTM-R2预测中...") + + # TTM-R2需要特定格式: (batch_size, context_length, num_features) + context_length = 512 + ttm_preds_list = [] + + # 从df_2026获取价格数据 + all_prices = df_2026['close'].values + all_open_prices = df_2026['open'].values + + # 为每个样本创建预测 + for i in range(min_samples): + # 计算当前样本在原始数据中的位置 + # DLinear的lookback是168，所以第i个样本对应df_2026的第168+i根K线 + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + # 如果超出范围，使用最后512个价格 + context_data = all_prices[-context_length:].reshape(1, -1, 1) + # 【修改】使用对应K线的open价格作为当前可交易价格 + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + # 【修改】使用第end_idx根K线的open价格（即当前K线的开盘价） + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + + with torch.no_grad(): + output = ttm_model(context_tensor) + # 预测第一个时间步的值 + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + + # 【修改】判断方向: 预测价格 vs 当前K线open价格 + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + print(f"✓ TTM-R2预测完成 (使用当前K线open价格判断)") + print(f" 预测上涨: {(ttm_preds==1).sum()} ({(ttm_preds==1).sum()/len(ttm_preds)*100:.1f}%)") + print(f" 预测下跌: {(ttm_preds==0).sum()} ({(ttm_preds==0).sum()/len(ttm_preds)*100:.1f}%)") + + # ========== 步骤5: 三模型多数投票策略（两方相同） ========== + print("\n[5/6] 应用三模型多数投票策略（两方相同）...") + + # 真实标签 + true_labels = y_dlinear + + # 多数投票：至少两个模型预测上涨 → 上涨，至少两个预测下跌 → 下跌 + vote_sum = dlinear_preds + nbeats_preds + ttm_preds + + # vote_sum >= 2 表示至少两个模型预测上涨（1） + # vote_sum <= 1 表示至少两个模型预测下跌（0） + majority_up = vote_sum >= 2 + majority_down = vote_sum <= 1 + + # 统计各种情况 + all_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) & (ttm_preds == 1) + all_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) & (ttm_preds == 0) + + # 恰好两方同意（不包括三方都同意） + exactly_two_up = majority_up & ~all_agree_up + exactly_two_down = majority_down & ~all_agree_down + + n_all_agree_up = all_agree_up.sum() + n_all_agree_down = all_agree_down.sum() + n_exactly_two_up = exactly_two_up.sum() + n_exactly_two_down = exactly_two_down.sum() + + print(f"\n📊 预测投票分析:") + print(f" 三方都预测上涨: {n_all_agree_up} ({n_all_agree_up/len(dlinear_preds)*100:.1f}%)") + print(f" 三方都预测下跌: {n_all_agree_down} ({n_all_agree_down/len(dlinear_preds)*100:.1f}%)") + print(f" 恰好两方预测上涨: {n_exactly_two_up} ({n_exactly_two_up/len(dlinear_preds)*100:.1f}%)") + print(f" 恰好两方预测下跌: {n_exactly_two_down} ({n_exactly_two_down/len(dlinear_preds)*100:.1f}%)") + + # 详细显示是哪两方同意 + dl_nb_up = (dlinear_preds == 1) & (nbeats_preds == 1) & (ttm_preds == 0) + dl_ttm_up = (dlinear_preds == 1) & (nbeats_preds == 0) & (ttm_preds == 1) + nb_ttm_up = (dlinear_preds == 0) & (nbeats_preds == 1) & (ttm_preds == 1) + + dl_nb_down = (dlinear_preds == 0) & (nbeats_preds == 0) & (ttm_preds == 1) + dl_ttm_down = (dlinear_preds == 0) & (nbeats_preds == 1) & (ttm_preds == 0) + nb_ttm_down = (dlinear_preds == 1) & (nbeats_preds == 0) & (ttm_preds == 0) + + print(f"\n 详细组合:") + print(f" DLinear+NBEATS同意看涨: {dl_nb_up.sum()} ({dl_nb_up.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" DLinear+TTM同意看涨: {dl_ttm_up.sum()} ({dl_ttm_up.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" NBEATS+TTM同意看涨: {nb_ttm_up.sum()} ({nb_ttm_up.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" DLinear+NBEATS同意看跌: {dl_nb_down.sum()} ({dl_nb_down.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" DLinear+TTM同意看跌: {dl_ttm_down.sum()} ({dl_ttm_down.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" NBEATS+TTM同意看跌: {nb_ttm_down.sum()} ({nb_ttm_down.sum()/len(dlinear_preds)*100:.1f}%)") + + # 多数投票结果 + ensemble_preds = np.where(majority_up, 1, 0) + + print(f"\n📊 三模型多数投票策略结果:") + print(f" 最终预测上涨: {(ensemble_preds==1).sum()} ({(ensemble_preds==1).sum()/len(dlinear_preds)*100:.1f}%)") + print(f" 最终预测下跌: {(ensemble_preds==0).sum()} ({(ensemble_preds==0).sum()/len(dlinear_preds)*100:.1f}%)") + print(f" 覆盖率: 100% (多数投票无弃权)") + + # ========== 步骤6: 性能对比 ========== + print("\n[6/6] 性能对比分析...") + + # 计算多数投票准确率 + ensemble_accuracy = (ensemble_preds == true_labels).mean() + print(f"\n✨ 三模型多数投票准确率: {ensemble_accuracy*100:.2f}%") + + # 混淆矩阵 + cm = confusion_matrix(true_labels, ensemble_preds) + tn, fp, fn, tp = cm.ravel() + + print(f"\n详细分析:") + print(f" ✓ 预测涨且真涨: {tp} ({tp/len(ensemble_preds)*100:.1f}%)") + print(f" ✓ 预测跌且真跌: {tn} ({tn/len(ensemble_preds)*100:.1f}%)") + print(f" ✗ 预测涨但真跌: {fp} ({fp/len(ensemble_preds)*100:.1f}%)") + print(f" ✗ 预测跌但真涨: {fn} ({fn/len(ensemble_preds)*100:.1f}%)") + + # 对比分析 + print("\n" + "="*80) + print("📊 性能对比") + print("="*80) + + dlinear_acc = (dlinear_preds == true_labels).mean() + nbeats_acc = (nbeats_preds == true_labels).mean() + ttm_acc = (ttm_preds == true_labels).mean() + + print(f"\n单模型性能:") + print(f" DLinear分类: {dlinear_acc*100:.2f}%") + print(f" N-BEATS回归: {nbeats_acc*100:.2f}%") + print(f" TTM-R2零样本: {ttm_acc*100:.2f}%") + + # 两模型交集（DLinear + N-BEATS） + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_preds = np.full(len(dlinear_preds), -1) + two_model_preds[two_model_agree_up] = 1 + two_model_preds[two_model_agree_down] = 0 + two_model_valid = two_model_preds != -1 + two_model_acc = (two_model_preds[two_model_valid] == true_labels[two_model_valid]).mean() + + print(f"\n集成策略性能:") + print(f" 双模型交集 (DLinear+NBEATS): {two_model_acc*100:.2f}% (覆盖率: {two_model_valid.sum()/len(dlinear_preds)*100:.1f}%)") + + if len(valid_preds) > 0: + print(f" 三模型交集 (ALL): {ensemble_accuracy*100:.2f}% (覆盖率: {valid_mask.sum()/len(dlinear_preds)*100:.1f}%)") + + # 性能评估 + print("\n" + "="*80) + print("🎯 策略评估") + print("="*80) + + best_single = max(dlinear_acc, nbeats_acc, ttm_acc) + + if ensemble_accuracy > best_single: + improvement = ensemble_accuracy - best_single + print(f"\n ✅ 三模型交集优于所有单模型!") + print(f" 准确率提升: {improvement*100:.2f}个百分点") + print(f" 代价: 弃权{(~valid_mask).sum()}次预测 ({(~valid_mask).sum()/len(dlinear_preds)*100:.1f}%)") + elif ensemble_accuracy > two_model_acc: + improvement = ensemble_accuracy - two_model_acc + print(f"\n ✅ 三模型交集优于双模型交集!") + print(f" 准确率提升: {improvement*100:.2f}个百分点") + print(f" 但覆盖率下降: {(two_model_valid.sum() - valid_mask.sum())/len(dlinear_preds)*100:.1f}%") + else: + print(f"\n ⚠️ 三模型交集未能提升准确率") + print(f" 可能原因: TTM-R2表现太差({ttm_acc*100:.2f}%)，拖累了整体性能") + print(f" 建议: 使用双模型交集(DLinear+NBEATS)") + + # 实际收益分析 + print(f"\n💰 实际收益分析:") + print(f" 如果只在三个模型都同意时交易:") + print(f" - 交易次数: {valid_mask.sum()} ({valid_mask.sum()/len(dlinear_preds)*100:.1f}%)") + print(f" - 准确率: {ensemble_accuracy*100:.2f}%") + print(f" - 避免交易: {(~valid_mask).sum()}次 (规避不确定性)") + + print("\n" + "="*80) + print("✓ 三模型集成预测分析完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/export_timesfm_onnx.py b/export_timesfm_onnx.py new file mode 100644 index 0000000..b083076 --- /dev/null +++ b/export_timesfm_onnx.py @@ -0,0 +1,295 @@ +"""将Google TimesFM 2.5模型导出为ONNX格式 - 使用正确的API""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import numpy as np + + +def export_timesfm_to_onnx(): + print("="*80) + print("导出TimesFM 2.5 (200M) 模型为ONNX格式") + print("="*80) + + # ========== 步骤1: 加载TimesFM模型 ========== + print("\n[1/6] 加载TimesFM 2.5模型...") + + try: + import timesfm + + # 使用正确的from_pretrained方法加载模型 + print(" 从HuggingFace加载预训练模型...") + print(" 模型ID: google/timesfm-2.5-200m-pytorch") + + model = timesfm.TimesFM_2p5_200M_torch.from_pretrained( + "google/timesfm-2.5-200m-pytorch", + torch_compile=False # 导出ONNX时不使用torch.compile + ) + + print("✓ TimesFM 2.5模型已加载") + + except ImportError as e: + print(f"✗ TimesFM未安装或导入失败: {e}") + print(" 请运行: pip install git+https://github.com/google-research/timesfm.git") + return + except Exception as e: + print(f"✗ 加载模型失败: {e}") + import traceback + traceback.print_exc() + return + + # ========== 步骤2: 编译模型配置 ========== + print("\n[2/6] 配置模型...") + + try: + # 配置模型 + model.compile( + timesfm.ForecastConfig( + max_context=512, # BTC历史数据512小时 + max_horizon=24, # 预测未来24小时 + normalize_inputs=True, + use_continuous_quantile_head=True, + force_flip_invariance=True, + infer_is_positive=True, + fix_quantile_crossing=True, + ) + ) + + print("✓ 模型配置完成") + print(f" 最大上下文长度: 512") + print(f" 最大预测长度: 24") + print(f" 归一化输入: True") + print(f" 使用分位数预测: True") + + except Exception as e: + print(f"✗ 配置失败: {e}") + import traceback + traceback.print_exc() + return + + # ========== 步骤3: 测试模型 ========== + print("\n[3/6] 测试模型推理...") + + try: + # 创建测试数据 + test_input = [ + np.linspace(85000, 90000, 512), # 模拟BTC价格从85k到90k + np.sin(np.linspace(0, 20, 512)) * 1000 + 87000, # 模拟震荡价格 + ] + + # 测试推理 + point_forecast, quantile_forecast = model.forecast( + horizon=24, + inputs=test_input, + ) + + print("✓ 模型推理成功") + print(f" 点预测形状: {point_forecast.shape}") # (2, 24) + print(f" 分位数预测形状: {quantile_forecast.shape}") # (2, 24, 11) + + except Exception as e: + print(f"⚠️ 模型测试失败: {e}") + import traceback + traceback.print_exc() + print(" 继续尝试导出...") + + # ========== 步骤4: 提取核心模型并导出ONNX ========== + print("\n[4/6] 提取核心模型...") + + try: + # TimesFM的核心模型在model.model中 + core_model = model.model + core_model.eval() + + print("✓ 核心模型已提取") + print(f" 模型类型: {type(core_model)}") + + # 创建ONNX导出包装器 + class TimesFMONNXWrapper(torch.nn.Module): + """TimesFM ONNX导出包装器""" + def __init__(self, timesfm_core_model): + super().__init__() + self.model = timesfm_core_model + + def forward(self, inputs, masks=None): + """ + Args: + inputs: (batch_size, context_length) - 输入时间序列 + masks: (batch_size, context_length) - 可选mask，1=有效，0=无效 + Returns: + point_forecast: (batch_size, horizon_length) - 点预测 + """ + # 如果没有提供masks，创建全1的mask + if masks is None: + masks = torch.ones_like(inputs) + + # 调用TimesFM核心模型 + # 注意：这里需要根据实际的forward签名调整 + outputs, _ = self.model(inputs, masks, decode_caches=None) + + # outputs包含: (input_embeddings, output_embeddings, output_ts, output_quantile_spread) + # output_ts是点预测 + point_forecast = outputs[2] + + return point_forecast + + wrapped_model = TimesFMONNXWrapper(core_model) + wrapped_model.eval() + + print("✓ ONNX包装器已创建") + + except Exception as e: + print(f"✗ 提取核心模型失败: {e}") + import traceback + traceback.print_exc() + return + + # ========== 步骤5: 导出为ONNX ========== + print("\n[5/6] 导出模型为ONNX格式...") + + onnx_path = "./models/timesfm_2p5_200m.onnx" + os.makedirs("./models", exist_ok=True) + + try: + # 准备示例输入 + batch_size = 1 + context_len = 512 + + dummy_input = torch.randn(batch_size, context_len, dtype=torch.float32) + dummy_mask = torch.ones(batch_size, context_len, dtype=torch.float32) + + print(f" 示例输入形状: inputs={dummy_input.shape}, masks={dummy_mask.shape}") + print(" 开始导出ONNX...") + + # 导出为ONNX + torch.onnx.export( + wrapped_model, + (dummy_input, dummy_mask), + onnx_path, + export_params=True, + opset_version=14, + do_constant_folding=True, + input_names=['inputs', 'masks'], + output_names=['point_forecast'], + dynamic_axes={ + 'inputs': {0: 'batch_size', 1: 'context_length'}, + 'masks': {0: 'batch_size', 1: 'context_length'}, + 'point_forecast': {0: 'batch_size', 1: 'horizon_length'} + }, + verbose=False + ) + + print(f"✓ ONNX模型已导出") + print(f" 保存路径: {onnx_path}") + print(f" 文件大小: {os.path.getsize(onnx_path) / (1024**2):.2f} MB") + + except Exception as e: + print(f"✗ ONNX导出失败: {e}") + import traceback + traceback.print_exc() + + # 尝试简化版本 + print("\n尝试简化导出方案...") + try: + # 只导出单输入版本 + torch.onnx.export( + wrapped_model, + dummy_input, + onnx_path, + export_params=True, + opset_version=14, + input_names=['inputs'], + output_names=['predictions'], + ) + print(f"✓ 简化版ONNX模型已导出: {onnx_path}") + except Exception as e2: + print(f"✗ 简化导出也失败: {e2}") + return + + # ========== 步骤6: 验证ONNX模型 ========== + print("\n[6/6] 验证ONNX模型...") + + try: + import onnx + import onnxruntime as ort + + # 加载并检查ONNX模型 + onnx_model = onnx.load(onnx_path) + onnx.checker.check_model(onnx_model) + print("✓ ONNX模型结构验证通过") + + # 显示模型信息 + print(f"\n模型信息:") + print(f" 输入:") + for input_tensor in onnx_model.graph.input: + dims = [d.dim_value if d.dim_value > 0 else 'dynamic' for d in input_tensor.type.tensor_type.shape.dim] + print(f" - {input_tensor.name}: {dims}") + + print(f" 输出:") + for output_tensor in onnx_model.graph.output: + dims = [d.dim_value if d.dim_value > 0 else 'dynamic' for d in output_tensor.type.tensor_type.shape.dim] + print(f" - {output_tensor.name}: {dims}") + + # 创建ONNX Runtime推理会话 + print("\n 创建ONNX Runtime推理会话...") + ort_session = ort.InferenceSession(onnx_path) + + # 准备测试输入 + test_input = np.random.randn(1, 512).astype(np.float32) + test_mask = np.ones((1, 512), dtype=np.float32) + + # 运行推理 + print(" 运行推理测试...") + outputs = ort_session.run( + None, + {'inputs': test_input, 'masks': test_mask} + ) + + print(f"✓ ONNX推理成功") + print(f" 输出数量: {len(outputs)}") + for i, output in enumerate(outputs): + print(f" 输出{i}形状: {output.shape}") + + except ImportError: + print("⚠️ 未安装onnx或onnxruntime，跳过验证") + print(" 安装命令: pip install onnx onnxruntime") + except Exception as e: + print(f"⚠️ ONNX验证失败: {e}") + import traceback + traceback.print_exc() + print(" 但模型文件已保存，可以尝试手动验证") + + # ========== 总结 ========== + print("\n" + "="*80) + print("✓ TimesFM 2.5模型导出完成!") + print("="*80) + + if os.path.exists(onnx_path): + print(f"\n保存位置: {onnx_path}") + print(f"文件大小: {os.path.getsize(onnx_path) / (1024**2):.2f} MB") + + print(f"\n使用方法:") + print(f"```python") + print(f"import onnxruntime as ort") + print(f"import numpy as np") + print(f"") + print(f"# 加载模型") + print(f"session = ort.InferenceSession('{onnx_path}')") + print(f"") + print(f"# 准备输入（512小时BTC历史价格）") + print(f"btc_prices = np.array([85000, 85100, ...]) # 512个价格点") + print(f"inputs = btc_prices.reshape(1, -1).astype(np.float32)") + print(f"masks = np.ones((1, 512), dtype=np.float32)") + print(f"") + print(f"# 推理") + print(f"outputs = session.run(None, {{'inputs': inputs, 'masks': masks}})") + print(f"predictions = outputs[0] # 未来24小时价格预测") + print(f"```") + + print("\n" + "="*80) + + +if __name__ == "__main__": + export_timesfm_to_onnx() diff --git a/install_foundation_models.sh b/install_foundation_models.sh new file mode 100755 index 0000000..f6450ce --- /dev/null +++ b/install_foundation_models.sh @@ -0,0 +1,32 @@ +#!/bin/bash +# 安装三个时间序列基础模型 + +echo "==================================" +echo "安装时间序列基础模型" +echo "==================================" + +# 激活conda环境 +source /Users/xxz/miniforge3/bin/activate +conda activate crypto_pred_312 + +echo "" +echo "[1/3] 安装 Google TimesFM 2.5..." +pip install git+https://github.com/google-research/timesfm.git + +echo "" +echo "[2/3] 安装 IBM TTM-R2 (Granite TSFM)..." +pip install tsfm_public + +echo "" +echo "[3/3] 安装 Lag-Llama 依赖..." +pip install gluonts[torch] torch + +echo "" +echo "==================================" +echo "✓ 所有模型安装完成!" +echo "==================================" +echo "" +echo "模型位置：" +echo " - TimesFM 2.5: HuggingFace自动下载" +echo " - TTM-R2: ibm-granite/granite-timeseries-ttm-r2" +echo " - Lag-Llama: time-series-foundation-models/lag-llama" diff --git a/models/checkpoints/catboost_classification_2026/catboost_model.cbm b/models/checkpoints/catboost_classification_2026/catboost_model.cbm new file mode 100644 index 0000000..1b34f55 Binary files /dev/null and b/models/checkpoints/catboost_classification_2026/catboost_model.cbm differ diff --git a/models/checkpoints/catboost_classification_2026/metadata.pkl b/models/checkpoints/catboost_classification_2026/metadata.pkl new file mode 100644 index 0000000..03d11d3 Binary files /dev/null and b/models/checkpoints/catboost_classification_2026/metadata.pkl differ diff --git a/models/checkpoints/nbeats_btc_1h/best_model.pt b/models/checkpoints/nbeats_btc_1h/best_model.pt new file mode 100644 index 0000000..7025d4a Binary files /dev/null and b/models/checkpoints/nbeats_btc_1h/best_model.pt differ diff --git a/models/checkpoints/nbeats_btc_1h/metrics.json b/models/checkpoints/nbeats_btc_1h/metrics.json new file mode 100644 index 0000000..517021b --- /dev/null +++ b/models/checkpoints/nbeats_btc_1h/metrics.json @@ -0,0 +1,7 @@ +{ + "price_mae": 434.07877457530066, + "price_rmse": 633.7116744507372, + "price_mape": 0.4112793819634811, + "direction_accuracy": 0.5083031112807788, + "best_val_loss": 0.00040675073481596463 +} \ No newline at end of file diff --git a/models/checkpoints/nbeats_btc_1h/preprocessor.pkl b/models/checkpoints/nbeats_btc_1h/preprocessor.pkl new file mode 100644 index 0000000..e0f8568 Binary files /dev/null and b/models/checkpoints/nbeats_btc_1h/preprocessor.pkl differ diff --git a/models/nbeats/__init__.py b/models/nbeats/__init__.py new file mode 100644 index 0000000..e69de29 diff --git a/models/nbeats/model.py b/models/nbeats/model.py new file mode 100644 index 0000000..4b8b124 --- /dev/null +++ b/models/nbeats/model.py @@ -0,0 +1,265 @@ +"""N-BEATS模型实现：预测未来1小时BTC价格""" + +import torch +import torch.nn as nn +import numpy as np + + +class NBeatsBlock(nn.Module): + """N-BEATS基础块""" + + def __init__(self, input_size, theta_size, basis_function, layers, layer_size): + super().__init__() + self.input_size = input_size + self.theta_size = theta_size + self.basis_function = basis_function + + # 全连接层栈 + self.fc_stack = nn.ModuleList([ + nn.Linear(input_size if i == 0 else layer_size, layer_size) + for i in range(layers) + ]) + + # Theta层（用于生成基函数参数） + self.theta_b_fc = nn.Linear(layer_size, theta_size, bias=False) # backcast + self.theta_f_fc = nn.Linear(layer_size, theta_size, bias=False) # forecast + + self.relu = nn.ReLU() + + def forward(self, x): + """ + Args: + x: (batch_size, input_size) + Returns: + backcast: (batch_size, input_size) + forecast: (batch_size, forecast_size) + """ + # 通过FC层栈 + for fc in self.fc_stack: + x = self.relu(fc(x)) + + # 生成theta参数 + theta_b = self.theta_b_fc(x) # backcast coefficients + theta_f = self.theta_f_fc(x) # forecast coefficients + + # 应用基函数 + backcast = self.basis_function(theta_b, is_forecast=False) + forecast = self.basis_function(theta_f, is_forecast=True) + + return backcast, forecast + + +class GenericBasis(nn.Module): + """通用基函数""" + + def __init__(self, backcast_size, forecast_size): + super().__init__() + self.backcast_size = backcast_size + self.forecast_size = forecast_size + + def forward(self, theta, is_forecast): + if is_forecast: + return theta[:, :self.forecast_size] + else: + return theta[:, :self.backcast_size] + + +class TrendBasis(nn.Module): + """趋势基函数（多项式）""" + + def __init__(self, degree, backcast_size, forecast_size): + super().__init__() + self.backcast_size = backcast_size + self.forecast_size = forecast_size + + # 多项式时间点 + backcast_time = torch.arange(backcast_size, dtype=torch.float32) / backcast_size + forecast_time = torch.arange(forecast_size, dtype=torch.float32) / forecast_size + + # 构建多项式基 + self.backcast_basis = torch.stack([backcast_time ** i for i in range(degree)], dim=1) + self.forecast_basis = torch.stack([forecast_time ** i for i in range(degree)], dim=1) + + def forward(self, theta, is_forecast): + basis = self.forecast_basis if is_forecast else self.backcast_basis + basis = basis.to(theta.device) + return torch.matmul(theta, basis.T) + + +class SeasonalityBasis(nn.Module): + """季节性基函数（傅里叶级数）""" + + def __init__(self, harmonics, backcast_size, forecast_size): + super().__init__() + self.backcast_size = backcast_size + self.forecast_size = forecast_size + + backcast_time = 2 * np.pi * torch.arange(backcast_size, dtype=torch.float32) / backcast_size + forecast_time = 2 * np.pi * torch.arange(forecast_size, dtype=torch.float32) / forecast_size + + # 构建傅里叶基（sin和cos） + backcast_basis = [] + forecast_basis = [] + for i in range(1, harmonics + 1): + backcast_basis.extend([torch.cos(i * backcast_time), torch.sin(i * backcast_time)]) + forecast_basis.extend([torch.cos(i * forecast_time), torch.sin(i * forecast_time)]) + + self.backcast_basis = torch.stack(backcast_basis, dim=1) + self.forecast_basis = torch.stack(forecast_basis, dim=1) + + def forward(self, theta, is_forecast): + basis = self.forecast_basis if is_forecast else self.backcast_basis + basis = basis.to(theta.device) + return torch.matmul(theta, basis.T) + + +class NBeatsStack(nn.Module): + """N-BEATS栈（多个块的组合）""" + + def __init__(self, input_size, forecast_size, num_blocks, block_layers, + layer_size, theta_size, basis_function): + super().__init__() + self.blocks = nn.ModuleList([ + NBeatsBlock( + input_size=input_size, + theta_size=theta_size, + basis_function=basis_function, + layers=block_layers, + layer_size=layer_size + ) + for _ in range(num_blocks) + ]) + self.forecast_size = forecast_size + + def forward(self, x): + """ + Args: + x: (batch_size, input_size) + Returns: + forecast: (batch_size, forecast_size) + """ + residual = x + forecast = torch.zeros(x.size(0), self.forecast_size, device=x.device) + + for block in self.blocks: + backcast, block_forecast = block(residual) + residual = residual - backcast + forecast = forecast + block_forecast + + return forecast + + +class NBeatsConfig: + """N-BEATS模型配置""" + + def __init__( + self, + input_size=168, # 输入序列长度 + forecast_size=1, # 预测长度 + stacks=['trend', 'seasonality', 'generic'], # 使用的栈类型 + num_blocks_per_stack=3, # 每个栈的块数 + block_layers=4, # 每个块的FC层数 + layer_size=256, # FC层大小 + trend_degree=2, # 趋势多项式阶数 + seasonality_harmonics=4, # 季节性谐波数 + ): + self.input_size = input_size + self.forecast_size = forecast_size + self.stacks = stacks + self.num_blocks_per_stack = num_blocks_per_stack + self.block_layers = block_layers + self.layer_size = layer_size + self.trend_degree = trend_degree + self.seasonality_harmonics = seasonality_harmonics + + +class NBeats(nn.Module): + """N-BEATS模型""" + + def __init__(self, config): + super().__init__() + self.config = config + + # 创建多个栈 + self.stacks = nn.ModuleList() + + for stack_type in config.stacks: + if stack_type == 'trend': + basis_function = TrendBasis( + degree=config.trend_degree, + backcast_size=config.input_size, + forecast_size=config.forecast_size + ) + theta_size = config.trend_degree + + elif stack_type == 'seasonality': + basis_function = SeasonalityBasis( + harmonics=config.seasonality_harmonics, + backcast_size=config.input_size, + forecast_size=config.forecast_size + ) + theta_size = 2 * config.seasonality_harmonics + + else: # generic + basis_function = GenericBasis( + backcast_size=config.input_size, + forecast_size=config.forecast_size + ) + theta_size = max(config.input_size, config.forecast_size) + + stack = NBeatsStack( + input_size=config.input_size, + forecast_size=config.forecast_size, + num_blocks=config.num_blocks_per_stack, + block_layers=config.block_layers, + layer_size=config.layer_size, + theta_size=theta_size, + basis_function=basis_function + ) + self.stacks.append(stack) + + def forward(self, x): + """ + Args: + x: (batch_size, lookback, features) 或 (batch_size, lookback) + Returns: + forecast: (batch_size, forecast_size) + """ + # 如果输入是多特征，取最后一个特征（close price） + if len(x.shape) == 3: + x = x[:, :, -1] # (batch_size, lookback) + + # 确保是2D + if len(x.shape) == 1: + x = x.unsqueeze(0) + + # 通过所有栈 + forecast = torch.zeros(x.size(0), self.config.forecast_size, device=x.device) + residual = x + + for stack in self.stacks: + stack_forecast = stack(residual) + forecast = forecast + stack_forecast + # 更新残差（用于下一个栈） + # 注意：这里简化了，标准N-BEATS会从每个栈获取backcast + + return forecast.squeeze(-1) # (batch_size,) + + def count_parameters(self): + """统计模型参数数量""" + return sum(p.numel() for p in self.parameters() if p.requires_grad) + + def save(self, path): + """保存模型""" + torch.save({ + 'model_state_dict': self.state_dict(), + 'config': self.config + }, path) + + @classmethod + def load(cls, path, device='cpu'): + """加载模型""" + checkpoint = torch.load(path, map_location=device) + model = cls(checkpoint['config']) + model.load_state_dict(checkpoint['model_state_dict']) + return model diff --git a/results/README_2026_01.md b/results/README_2026_01.md new file mode 100644 index 0000000..ef411c8 --- /dev/null +++ b/results/README_2026_01.md @@ -0,0 +1,83 @@ +# 加密货币价格预测模型测试报告 + +## 测试概况 + +- **测试期间**: 2026年1月1日 - 2026年1月30日 +- **总天数**: 22天 +- **总样本数**: 495个小时 +- **生成时间**: 2026-01-31 16:47:08 + +--- + +## 单模型性能 + +### DLinear分类模型 +- **准确率**: 53.33% +- **预测上涨比例**: 47.1% +- **预测下跌比例**: 52.9% + +### N-BEATS回归模型 +- **准确率**: 49.90% +- **预测上涨比例**: 73.5% +- **预测下跌比例**: 26.5% + +### TTM-R2零样本模型 +- **准确率**: 51.31% +- **预测上涨比例**: 23.6% +- **预测下跌比例**: 76.4% + +--- + +## 集成策略性能 + +### 双模型交集（DLinear + N-BEATS） +- **准确率**: 53.39% +- **覆盖率**: 47.68% +- **有效预测数**: 236次 + +### 三模型多数投票（DLinear + N-BEATS + TTM-R2） +- **准确率**: 52.93% +- **覆盖率**: 100.00% +- **有效预测数**: 495次 + +--- + +## DLinear概率分布分析 + +| 概率区间 | 样本数 | 占比 | +|---------|--------|------| +| 0.4-0.5 | 262 | 52.9% | +| 0.5-0.6 | 229 | 46.3% | +| 0.6-0.7 | 4 | 0.8% | + +**平均概率**: 0.501 + +--- + +## 最佳/最差交易日 + +### 双模型交集 +- **最佳日**: 2026-01-11 (83.3%) +- **最差日**: 2026-01-23 (28.6%) + +### 三模型投票 +- **最佳日**: 2026-01-28 (80.0%) +- **最差日**: 2026-01-23 (29.2%) + +--- + +## 结论 + +1. **最佳单模型**: DLinear (53.33%) +2. **最佳集成策略**: 双模型交集 (53.39%) +3. **DLinear置信度**: 99.2%的预测集中在0.45-0.55区间，缺乏高置信度预测 + +## 数据文件 + +- 逐小时预测结果: `hourly_predictions_2026_01.csv` +- 每日统计数据: `daily_statistics_2026_01.csv` +- 总结报告: `summary_report_2026_01.json` + +--- + +生成时间: 2026-01-31 16:47:08 diff --git a/results/daily_statistics_2026_01.csv b/results/daily_statistics_2026_01.csv new file mode 100644 index 0000000..4fcd9ed --- /dev/null +++ b/results/daily_statistics_2026_01.csv @@ -0,0 +1,23 @@ +date,total_samples,two_model_valid,two_model_correct,three_model_correct,dlinear_correct,dlinear_accuracy,two_model_accuracy,three_model_accuracy,two_model_coverage +2026-01-07,5,3,2,3,4,80.0,66.66666666666666,60.0,60.0 +2026-01-08,24,13,6,12,11,45.83333333333333,46.15384615384615,50.0,54.166666666666664 +2026-01-09,24,12,6,15,13,54.166666666666664,50.0,62.5,50.0 +2026-01-10,24,12,6,12,13,54.166666666666664,50.0,50.0,50.0 +2026-01-11,24,6,5,12,11,45.83333333333333,83.33333333333334,50.0,25.0 +2026-01-12,24,6,3,8,13,54.166666666666664,50.0,33.33333333333333,25.0 +2026-01-13,24,12,6,12,12,50.0,50.0,50.0,50.0 +2026-01-14,24,12,6,13,14,58.333333333333336,50.0,54.166666666666664,50.0 +2026-01-15,24,14,8,12,14,58.333333333333336,57.14285714285714,50.0,58.333333333333336 +2026-01-16,24,13,6,13,10,41.66666666666667,46.15384615384615,54.166666666666664,54.166666666666664 +2026-01-17,24,8,5,13,11,45.83333333333333,62.5,54.166666666666664,33.33333333333333 +2026-01-18,24,17,9,15,12,50.0,52.94117647058824,62.5,70.83333333333334 +2026-01-19,24,17,12,17,16,66.66666666666666,70.58823529411765,70.83333333333334,70.83333333333334 +2026-01-20,24,6,2,9,11,45.83333333333333,33.33333333333333,37.5,25.0 +2026-01-21,24,9,6,15,15,62.5,66.66666666666666,62.5,37.5 +2026-01-22,24,10,6,13,14,58.333333333333336,60.0,54.166666666666664,41.66666666666667 +2026-01-23,24,14,4,7,7,29.166666666666668,28.57142857142857,29.166666666666668,58.333333333333336 +2026-01-24,24,14,7,13,14,58.333333333333336,50.0,54.166666666666664,58.333333333333336 +2026-01-25,24,4,0,11,10,41.66666666666667,0.0,45.83333333333333,16.666666666666664 +2026-01-26,24,13,8,13,15,62.5,61.53846153846154,54.166666666666664,54.166666666666664 +2026-01-27,24,13,7,16,16,66.66666666666666,53.84615384615385,66.66666666666666,54.166666666666664 +2026-01-28,10,8,6,8,8,80.0,75.0,80.0,80.0 diff --git a/results/hourly_predictions_2026_01.csv b/results/hourly_predictions_2026_01.csv new file mode 100644 index 0000000..5a63caf --- /dev/null +++ b/results/hourly_predictions_2026_01.csv @@ -0,0 +1,496 @@ +timestamp_utc,timestamp_et,date,true_label,dlinear_prob,dlinear_pred,nbeats_pred,ttm_pred,two_model_pred,three_model_pred,dlinear_correct,two_model_correct,three_model_correct +2026-01-08 00:00:00,2026-01-07 19:00:00-05:00,2026-01-07,1,0.57705873,1,0,1,-1,1,True,False,True +2026-01-08 01:00:00,2026-01-07 20:00:00-05:00,2026-01-07,0,0.49304762,0,0,1,0,0,True,True,True +2026-01-08 02:00:00,2026-01-07 21:00:00-05:00,2026-01-07,0,0.47215617,0,0,0,0,0,True,True,True +2026-01-08 03:00:00,2026-01-07 22:00:00-05:00,2026-01-07,1,0.46118113,0,0,0,0,0,False,False,False +2026-01-08 04:00:00,2026-01-07 23:00:00-05:00,2026-01-07,0,0.45929378,0,1,1,-1,1,True,False,False +2026-01-08 05:00:00,2026-01-08 00:00:00-05:00,2026-01-08,1,0.47178885,0,0,0,0,0,False,False,False +2026-01-08 06:00:00,2026-01-08 01:00:00-05:00,2026-01-08,1,0.5306643,1,1,1,1,1,True,True,True +2026-01-08 07:00:00,2026-01-08 02:00:00-05:00,2026-01-08,0,0.4891925,0,1,1,-1,1,True,False,False +2026-01-08 08:00:00,2026-01-08 03:00:00-05:00,2026-01-08,0,0.5191637,1,0,1,-1,1,False,False,False +2026-01-08 09:00:00,2026-01-08 04:00:00-05:00,2026-01-08,0,0.47131988,0,1,1,-1,1,True,False,False +2026-01-08 10:00:00,2026-01-08 05:00:00-05:00,2026-01-08,0,0.47046715,0,0,0,0,0,True,True,True +2026-01-08 11:00:00,2026-01-08 06:00:00-05:00,2026-01-08,1,0.5125343,1,0,1,-1,1,True,False,True +2026-01-08 12:00:00,2026-01-08 07:00:00-05:00,2026-01-08,1,0.5334059,1,1,1,1,1,True,True,True +2026-01-08 13:00:00,2026-01-08 08:00:00-05:00,2026-01-08,1,0.49018916,0,1,1,-1,1,False,False,True +2026-01-08 14:00:00,2026-01-08 09:00:00-05:00,2026-01-08,0,0.50079376,1,1,1,1,1,False,False,False +2026-01-08 15:00:00,2026-01-08 10:00:00-05:00,2026-01-08,1,0.46582156,0,1,1,-1,1,False,False,True +2026-01-08 16:00:00,2026-01-08 11:00:00-05:00,2026-01-08,1,0.46240616,0,1,1,-1,1,False,False,True +2026-01-08 17:00:00,2026-01-08 12:00:00-05:00,2026-01-08,0,0.48750654,0,1,1,-1,1,True,False,False +2026-01-08 18:00:00,2026-01-08 13:00:00-05:00,2026-01-08,0,0.47014055,0,1,1,-1,1,True,False,False +2026-01-08 19:00:00,2026-01-08 14:00:00-05:00,2026-01-08,1,0.49040872,0,1,1,-1,1,False,False,True +2026-01-08 20:00:00,2026-01-08 15:00:00-05:00,2026-01-08,0,0.5273022,1,1,1,1,1,False,False,False +2026-01-08 21:00:00,2026-01-08 16:00:00-05:00,2026-01-08,1,0.5106931,1,1,1,1,1,True,True,True +2026-01-08 22:00:00,2026-01-08 17:00:00-05:00,2026-01-08,0,0.5778609,1,1,1,1,1,False,False,False +2026-01-08 23:00:00,2026-01-08 18:00:00-05:00,2026-01-08,1,0.54889923,1,1,1,1,1,True,True,True +2026-01-09 00:00:00,2026-01-08 19:00:00-05:00,2026-01-08,1,0.51507455,1,1,1,1,1,True,True,True +2026-01-09 01:00:00,2026-01-08 20:00:00-05:00,2026-01-08,0,0.5170575,1,1,1,1,1,False,False,False +2026-01-09 02:00:00,2026-01-08 21:00:00-05:00,2026-01-08,1,0.4717993,0,1,1,-1,1,False,False,True +2026-01-09 03:00:00,2026-01-08 22:00:00-05:00,2026-01-08,0,0.5043716,1,1,1,1,1,False,False,False +2026-01-09 04:00:00,2026-01-08 23:00:00-05:00,2026-01-08,0,0.51360196,1,1,1,1,1,False,False,False +2026-01-09 05:00:00,2026-01-09 00:00:00-05:00,2026-01-09,1,0.469029,0,1,1,-1,1,False,False,True +2026-01-09 06:00:00,2026-01-09 01:00:00-05:00,2026-01-09,1,0.50669587,1,1,1,1,1,True,True,True +2026-01-09 07:00:00,2026-01-09 02:00:00-05:00,2026-01-09,0,0.5564795,1,1,1,1,1,False,False,False +2026-01-09 08:00:00,2026-01-09 03:00:00-05:00,2026-01-09,1,0.5820567,1,1,1,1,1,True,True,True +2026-01-09 09:00:00,2026-01-09 04:00:00-05:00,2026-01-09,1,0.5504754,1,1,1,1,1,True,True,True +2026-01-09 10:00:00,2026-01-09 05:00:00-05:00,2026-01-09,0,0.53064233,1,1,1,1,1,False,False,False +2026-01-09 11:00:00,2026-01-09 06:00:00-05:00,2026-01-09,0,0.50407845,1,1,1,1,1,False,False,False +2026-01-09 12:00:00,2026-01-09 07:00:00-05:00,2026-01-09,0,0.4576875,0,1,1,-1,1,True,False,False +2026-01-09 13:00:00,2026-01-09 08:00:00-05:00,2026-01-09,1,0.5189501,1,0,1,-1,1,True,False,True +2026-01-09 14:00:00,2026-01-09 09:00:00-05:00,2026-01-09,1,0.50425255,1,1,1,1,1,True,True,True +2026-01-09 15:00:00,2026-01-09 10:00:00-05:00,2026-01-09,0,0.45628017,0,1,1,-1,1,True,False,False +2026-01-09 16:00:00,2026-01-09 11:00:00-05:00,2026-01-09,1,0.44883052,0,0,1,0,0,False,False,False +2026-01-09 17:00:00,2026-01-09 12:00:00-05:00,2026-01-09,1,0.46907404,0,1,1,-1,1,False,False,True +2026-01-09 18:00:00,2026-01-09 13:00:00-05:00,2026-01-09,1,0.48777947,0,1,1,-1,1,False,False,True +2026-01-09 19:00:00,2026-01-09 14:00:00-05:00,2026-01-09,0,0.49411413,0,1,0,-1,0,True,False,True +2026-01-09 20:00:00,2026-01-09 15:00:00-05:00,2026-01-09,0,0.47928023,0,1,0,-1,0,True,False,True +2026-01-09 21:00:00,2026-01-09 16:00:00-05:00,2026-01-09,0,0.49988702,0,1,0,-1,0,True,False,True +2026-01-09 22:00:00,2026-01-09 17:00:00-05:00,2026-01-09,1,0.4896841,0,1,0,-1,0,False,False,False +2026-01-09 23:00:00,2026-01-09 18:00:00-05:00,2026-01-09,0,0.49358237,0,1,0,-1,0,True,False,True +2026-01-10 00:00:00,2026-01-09 19:00:00-05:00,2026-01-09,0,0.50367767,1,1,0,1,1,False,False,False +2026-01-10 01:00:00,2026-01-09 20:00:00-05:00,2026-01-09,1,0.52877015,1,1,0,1,1,True,True,True +2026-01-10 02:00:00,2026-01-09 21:00:00-05:00,2026-01-09,1,0.51958245,1,1,0,1,1,True,True,True +2026-01-10 03:00:00,2026-01-09 22:00:00-05:00,2026-01-09,0,0.53057724,1,1,0,1,1,False,False,False +2026-01-10 04:00:00,2026-01-09 23:00:00-05:00,2026-01-09,1,0.45753565,0,1,1,-1,1,False,False,True +2026-01-10 05:00:00,2026-01-10 00:00:00-05:00,2026-01-10,0,0.447833,0,1,0,-1,0,True,False,True +2026-01-10 06:00:00,2026-01-10 01:00:00-05:00,2026-01-10,1,0.4577814,0,1,1,-1,1,False,False,True +2026-01-10 07:00:00,2026-01-10 02:00:00-05:00,2026-01-10,0,0.47578683,0,1,1,-1,1,True,False,False +2026-01-10 08:00:00,2026-01-10 03:00:00-05:00,2026-01-10,0,0.4506602,0,1,1,-1,1,True,False,False +2026-01-10 09:00:00,2026-01-10 04:00:00-05:00,2026-01-10,0,0.5478148,1,1,0,1,1,False,False,False +2026-01-10 10:00:00,2026-01-10 05:00:00-05:00,2026-01-10,1,0.5990233,1,1,0,1,1,True,True,True +2026-01-10 11:00:00,2026-01-10 06:00:00-05:00,2026-01-10,1,0.5706127,1,1,0,1,1,True,True,True +2026-01-10 12:00:00,2026-01-10 07:00:00-05:00,2026-01-10,0,0.5276616,1,1,0,1,1,False,False,False +2026-01-10 13:00:00,2026-01-10 08:00:00-05:00,2026-01-10,1,0.51404405,1,1,0,1,1,True,True,True +2026-01-10 14:00:00,2026-01-10 09:00:00-05:00,2026-01-10,1,0.50078404,1,1,0,1,1,True,True,True +2026-01-10 15:00:00,2026-01-10 10:00:00-05:00,2026-01-10,1,0.45550296,0,1,0,-1,0,False,False,False +2026-01-10 16:00:00,2026-01-10 11:00:00-05:00,2026-01-10,1,0.44787562,0,1,0,-1,0,False,False,False +2026-01-10 17:00:00,2026-01-10 12:00:00-05:00,2026-01-10,0,0.44727677,0,1,0,-1,0,True,False,True +2026-01-10 18:00:00,2026-01-10 13:00:00-05:00,2026-01-10,1,0.44855183,0,1,0,-1,0,False,False,False +2026-01-10 19:00:00,2026-01-10 14:00:00-05:00,2026-01-10,0,0.45485023,0,1,0,-1,0,True,False,True +2026-01-10 20:00:00,2026-01-10 15:00:00-05:00,2026-01-10,0,0.47032186,0,1,0,-1,0,True,False,True +2026-01-10 21:00:00,2026-01-10 16:00:00-05:00,2026-01-10,0,0.48025206,0,1,0,-1,0,True,False,True +2026-01-10 22:00:00,2026-01-10 17:00:00-05:00,2026-01-10,0,0.5247036,1,1,0,1,1,False,False,False +2026-01-10 23:00:00,2026-01-10 18:00:00-05:00,2026-01-10,0,0.58018094,1,1,0,1,1,False,False,False +2026-01-11 00:00:00,2026-01-10 19:00:00-05:00,2026-01-10,0,0.59668726,1,1,0,1,1,False,False,False +2026-01-11 01:00:00,2026-01-10 20:00:00-05:00,2026-01-10,1,0.5909982,1,1,0,1,1,True,True,True +2026-01-11 02:00:00,2026-01-10 21:00:00-05:00,2026-01-10,1,0.5297047,1,1,0,1,1,True,True,True +2026-01-11 03:00:00,2026-01-10 22:00:00-05:00,2026-01-10,0,0.5227216,1,1,0,1,1,False,False,False +2026-01-11 04:00:00,2026-01-10 23:00:00-05:00,2026-01-10,1,0.4885675,0,1,0,-1,0,False,False,False +2026-01-11 05:00:00,2026-01-11 00:00:00-05:00,2026-01-11,1,0.47107595,0,1,0,-1,0,False,False,False +2026-01-11 06:00:00,2026-01-11 01:00:00-05:00,2026-01-11,1,0.4610141,0,1,0,-1,0,False,False,False +2026-01-11 07:00:00,2026-01-11 02:00:00-05:00,2026-01-11,0,0.4598382,0,1,1,-1,1,True,False,False +2026-01-11 08:00:00,2026-01-11 03:00:00-05:00,2026-01-11,1,0.48207,0,1,1,-1,1,False,False,True +2026-01-11 09:00:00,2026-01-11 04:00:00-05:00,2026-01-11,1,0.513094,1,1,1,1,1,True,True,True +2026-01-11 10:00:00,2026-01-11 05:00:00-05:00,2026-01-11,0,0.47068593,0,1,1,-1,1,True,False,False +2026-01-11 11:00:00,2026-01-11 06:00:00-05:00,2026-01-11,0,0.4895874,0,1,1,-1,1,True,False,False +2026-01-11 12:00:00,2026-01-11 07:00:00-05:00,2026-01-11,1,0.46443766,0,1,1,-1,1,False,False,True +2026-01-11 13:00:00,2026-01-11 08:00:00-05:00,2026-01-11,1,0.4845122,0,1,1,-1,1,False,False,True +2026-01-11 14:00:00,2026-01-11 09:00:00-05:00,2026-01-11,1,0.50560534,1,1,1,1,1,True,True,True +2026-01-11 15:00:00,2026-01-11 10:00:00-05:00,2026-01-11,0,0.47123846,0,0,1,0,0,True,True,True +2026-01-11 16:00:00,2026-01-11 11:00:00-05:00,2026-01-11,0,0.45919827,0,1,1,-1,1,True,False,False +2026-01-11 17:00:00,2026-01-11 12:00:00-05:00,2026-01-11,1,0.52895916,1,0,1,-1,1,True,False,True +2026-01-11 18:00:00,2026-01-11 13:00:00-05:00,2026-01-11,1,0.5049388,1,1,1,1,1,True,True,True +2026-01-11 19:00:00,2026-01-11 14:00:00-05:00,2026-01-11,1,0.48020154,0,1,1,-1,1,False,False,True +2026-01-11 20:00:00,2026-01-11 15:00:00-05:00,2026-01-11,1,0.49915588,0,1,1,-1,1,False,False,True +2026-01-11 21:00:00,2026-01-11 16:00:00-05:00,2026-01-11,0,0.5144025,1,1,0,1,1,False,False,False +2026-01-11 22:00:00,2026-01-11 17:00:00-05:00,2026-01-11,1,0.49443942,0,1,0,-1,0,False,False,False +2026-01-11 23:00:00,2026-01-11 18:00:00-05:00,2026-01-11,1,0.4918472,0,1,0,-1,0,False,False,False +2026-01-12 00:00:00,2026-01-11 19:00:00-05:00,2026-01-11,0,0.4723591,0,0,1,0,0,True,True,True +2026-01-12 01:00:00,2026-01-11 20:00:00-05:00,2026-01-11,0,0.45115232,0,1,1,-1,1,True,False,False +2026-01-12 02:00:00,2026-01-11 21:00:00-05:00,2026-01-11,1,0.47599584,0,1,1,-1,1,False,False,True +2026-01-12 03:00:00,2026-01-11 22:00:00-05:00,2026-01-11,0,0.51635396,1,0,1,-1,1,False,False,False +2026-01-12 04:00:00,2026-01-11 23:00:00-05:00,2026-01-11,1,0.47296882,0,1,0,-1,0,False,False,False +2026-01-12 05:00:00,2026-01-12 00:00:00-05:00,2026-01-12,1,0.48586744,0,1,0,-1,0,False,False,False +2026-01-12 06:00:00,2026-01-12 01:00:00-05:00,2026-01-12,0,0.47855884,0,1,0,-1,0,True,False,True +2026-01-12 07:00:00,2026-01-12 02:00:00-05:00,2026-01-12,0,0.4667405,0,1,0,-1,0,True,False,True +2026-01-12 08:00:00,2026-01-12 03:00:00-05:00,2026-01-12,1,0.47844118,0,1,0,-1,0,False,False,False +2026-01-12 09:00:00,2026-01-12 04:00:00-05:00,2026-01-12,1,0.49896824,0,1,0,-1,0,False,False,False +2026-01-12 10:00:00,2026-01-12 05:00:00-05:00,2026-01-12,1,0.4889443,0,1,0,-1,0,False,False,False +2026-01-12 11:00:00,2026-01-12 06:00:00-05:00,2026-01-12,0,0.45522135,0,1,0,-1,0,True,False,True +2026-01-12 12:00:00,2026-01-12 07:00:00-05:00,2026-01-12,1,0.45169547,0,0,0,0,0,False,False,False +2026-01-12 13:00:00,2026-01-12 08:00:00-05:00,2026-01-12,0,0.48637372,0,0,0,0,0,True,True,True +2026-01-12 14:00:00,2026-01-12 09:00:00-05:00,2026-01-12,0,0.45615003,0,1,0,-1,0,True,False,True +2026-01-12 15:00:00,2026-01-12 10:00:00-05:00,2026-01-12,1,0.45153883,0,1,0,-1,0,False,False,False +2026-01-12 16:00:00,2026-01-12 11:00:00-05:00,2026-01-12,1,0.47393757,0,0,0,0,0,False,False,False +2026-01-12 17:00:00,2026-01-12 12:00:00-05:00,2026-01-12,1,0.47199413,0,1,0,-1,0,False,False,False +2026-01-12 18:00:00,2026-01-12 13:00:00-05:00,2026-01-12,0,0.43410957,0,1,1,-1,1,True,False,False +2026-01-12 19:00:00,2026-01-12 14:00:00-05:00,2026-01-12,0,0.44708502,0,1,1,-1,1,True,False,False +2026-01-12 20:00:00,2026-01-12 15:00:00-05:00,2026-01-12,1,0.58962935,1,1,1,1,1,True,True,True +2026-01-12 21:00:00,2026-01-12 16:00:00-05:00,2026-01-12,1,0.5799872,1,1,1,1,1,True,True,True +2026-01-12 22:00:00,2026-01-12 17:00:00-05:00,2026-01-12,0,0.5379354,1,1,0,1,1,False,False,False +2026-01-12 23:00:00,2026-01-12 18:00:00-05:00,2026-01-12,1,0.52044976,1,0,0,-1,0,True,False,False +2026-01-13 00:00:00,2026-01-12 19:00:00-05:00,2026-01-12,1,0.546228,1,0,0,-1,0,True,False,False +2026-01-13 01:00:00,2026-01-12 20:00:00-05:00,2026-01-12,0,0.50185496,1,0,0,-1,0,False,False,True +2026-01-13 02:00:00,2026-01-12 21:00:00-05:00,2026-01-12,1,0.5131573,1,0,0,-1,0,True,False,False +2026-01-13 03:00:00,2026-01-12 22:00:00-05:00,2026-01-12,0,0.49922633,0,1,1,-1,1,True,False,False +2026-01-13 04:00:00,2026-01-12 23:00:00-05:00,2026-01-12,1,0.47469753,0,1,0,-1,0,False,False,False +2026-01-13 05:00:00,2026-01-13 00:00:00-05:00,2026-01-13,1,0.45794666,0,1,1,-1,1,False,False,True +2026-01-13 06:00:00,2026-01-13 01:00:00-05:00,2026-01-13,1,0.45379728,0,1,1,-1,1,False,False,True +2026-01-13 07:00:00,2026-01-13 02:00:00-05:00,2026-01-13,0,0.45295683,0,1,1,-1,1,True,False,False +2026-01-13 08:00:00,2026-01-13 03:00:00-05:00,2026-01-13,0,0.44573298,0,1,0,-1,0,True,False,True +2026-01-13 09:00:00,2026-01-13 04:00:00-05:00,2026-01-13,1,0.4709209,0,1,0,-1,0,False,False,False +2026-01-13 10:00:00,2026-01-13 05:00:00-05:00,2026-01-13,1,0.54900587,1,1,1,1,1,True,True,True +2026-01-13 11:00:00,2026-01-13 06:00:00-05:00,2026-01-13,0,0.54181457,1,1,1,1,1,False,False,False +2026-01-13 12:00:00,2026-01-13 07:00:00-05:00,2026-01-13,0,0.47640413,0,1,1,-1,1,True,False,False +2026-01-13 13:00:00,2026-01-13 08:00:00-05:00,2026-01-13,1,0.5774929,1,0,1,-1,1,True,False,True +2026-01-13 14:00:00,2026-01-13 09:00:00-05:00,2026-01-13,0,0.55843365,1,1,1,1,1,False,False,False +2026-01-13 15:00:00,2026-01-13 10:00:00-05:00,2026-01-13,0,0.53291476,1,1,1,1,1,False,False,False +2026-01-13 16:00:00,2026-01-13 11:00:00-05:00,2026-01-13,0,0.5417355,1,0,1,-1,1,False,False,False +2026-01-13 17:00:00,2026-01-13 12:00:00-05:00,2026-01-13,0,0.526461,1,1,1,1,1,False,False,False +2026-01-13 18:00:00,2026-01-13 13:00:00-05:00,2026-01-13,1,0.5094685,1,1,1,1,1,True,True,True +2026-01-13 19:00:00,2026-01-13 14:00:00-05:00,2026-01-13,1,0.50502,1,0,1,-1,1,True,False,True +2026-01-13 20:00:00,2026-01-13 15:00:00-05:00,2026-01-13,0,0.5074931,1,1,1,1,1,False,False,False +2026-01-13 21:00:00,2026-01-13 16:00:00-05:00,2026-01-13,1,0.50496596,1,1,1,1,1,True,True,True +2026-01-13 22:00:00,2026-01-13 17:00:00-05:00,2026-01-13,1,0.5311341,1,1,1,1,1,True,True,True +2026-01-13 23:00:00,2026-01-13 18:00:00-05:00,2026-01-13,1,0.5194955,1,0,1,-1,1,True,False,True +2026-01-14 00:00:00,2026-01-13 19:00:00-05:00,2026-01-13,0,0.51150465,1,0,1,-1,1,False,False,False +2026-01-14 01:00:00,2026-01-13 20:00:00-05:00,2026-01-13,0,0.53702235,1,0,1,-1,1,False,False,False +2026-01-14 02:00:00,2026-01-13 21:00:00-05:00,2026-01-13,1,0.4907134,0,0,0,0,0,False,False,False +2026-01-14 03:00:00,2026-01-13 22:00:00-05:00,2026-01-13,0,0.4840499,0,0,0,0,0,True,True,True +2026-01-14 04:00:00,2026-01-13 23:00:00-05:00,2026-01-13,0,0.4611074,0,0,0,0,0,True,True,True +2026-01-14 05:00:00,2026-01-14 00:00:00-05:00,2026-01-14,1,0.45327646,0,0,0,0,0,False,False,False +2026-01-14 06:00:00,2026-01-14 01:00:00-05:00,2026-01-14,1,0.49665686,0,0,0,0,0,False,False,False +2026-01-14 07:00:00,2026-01-14 02:00:00-05:00,2026-01-14,0,0.4882262,0,0,1,0,0,True,True,True +2026-01-14 08:00:00,2026-01-14 03:00:00-05:00,2026-01-14,0,0.46823135,0,1,1,-1,1,True,False,False +2026-01-14 09:00:00,2026-01-14 04:00:00-05:00,2026-01-14,0,0.47908562,0,1,0,-1,0,True,False,True +2026-01-14 10:00:00,2026-01-14 05:00:00-05:00,2026-01-14,0,0.55422133,1,1,0,1,1,False,False,False +2026-01-14 11:00:00,2026-01-14 06:00:00-05:00,2026-01-14,1,0.610168,1,1,1,1,1,True,True,True +2026-01-14 12:00:00,2026-01-14 07:00:00-05:00,2026-01-14,0,0.5614474,1,1,1,1,1,False,False,False +2026-01-14 13:00:00,2026-01-14 08:00:00-05:00,2026-01-14,1,0.56697947,1,0,1,-1,1,True,False,True +2026-01-14 14:00:00,2026-01-14 09:00:00-05:00,2026-01-14,1,0.52636105,1,0,1,-1,1,True,False,True +2026-01-14 15:00:00,2026-01-14 10:00:00-05:00,2026-01-14,1,0.50316924,1,0,1,-1,1,True,False,True +2026-01-14 16:00:00,2026-01-14 11:00:00-05:00,2026-01-14,0,0.5363478,1,0,1,-1,1,False,False,False +2026-01-14 17:00:00,2026-01-14 12:00:00-05:00,2026-01-14,0,0.47355476,0,0,1,0,0,True,True,True +2026-01-14 18:00:00,2026-01-14 13:00:00-05:00,2026-01-14,0,0.48269007,0,0,1,0,0,True,True,True +2026-01-14 19:00:00,2026-01-14 14:00:00-05:00,2026-01-14,1,0.50739765,1,0,1,-1,1,True,False,True +2026-01-14 20:00:00,2026-01-14 15:00:00-05:00,2026-01-14,1,0.501837,1,0,1,-1,1,True,False,True +2026-01-14 21:00:00,2026-01-14 16:00:00-05:00,2026-01-14,0,0.4608701,0,0,1,0,0,True,True,True +2026-01-14 22:00:00,2026-01-14 17:00:00-05:00,2026-01-14,1,0.4631222,0,0,1,0,0,False,False,False +2026-01-14 23:00:00,2026-01-14 18:00:00-05:00,2026-01-14,0,0.5486243,1,0,1,-1,1,False,False,False +2026-01-15 00:00:00,2026-01-14 19:00:00-05:00,2026-01-14,0,0.55234075,1,0,1,-1,1,False,False,False +2026-01-15 01:00:00,2026-01-14 20:00:00-05:00,2026-01-14,1,0.4942968,0,1,1,-1,1,False,False,True +2026-01-15 02:00:00,2026-01-14 21:00:00-05:00,2026-01-14,1,0.5293442,1,1,1,1,1,True,True,True +2026-01-15 03:00:00,2026-01-14 22:00:00-05:00,2026-01-14,0,0.5131931,1,1,1,1,1,False,False,False +2026-01-15 04:00:00,2026-01-14 23:00:00-05:00,2026-01-14,0,0.48758185,0,1,1,-1,1,True,False,False +2026-01-15 05:00:00,2026-01-15 00:00:00-05:00,2026-01-15,0,0.4538339,0,0,1,0,0,True,True,True +2026-01-15 06:00:00,2026-01-15 01:00:00-05:00,2026-01-15,1,0.48841763,0,1,1,-1,1,False,False,True +2026-01-15 07:00:00,2026-01-15 02:00:00-05:00,2026-01-15,1,0.5323837,1,1,1,1,1,True,True,True +2026-01-15 08:00:00,2026-01-15 03:00:00-05:00,2026-01-15,0,0.5621298,1,1,1,1,1,False,False,False +2026-01-15 09:00:00,2026-01-15 04:00:00-05:00,2026-01-15,1,0.46453536,0,1,1,-1,1,False,False,True +2026-01-15 10:00:00,2026-01-15 05:00:00-05:00,2026-01-15,1,0.5253816,1,1,1,1,1,True,True,True +2026-01-15 11:00:00,2026-01-15 06:00:00-05:00,2026-01-15,0,0.5148865,1,1,1,1,1,False,False,False +2026-01-15 12:00:00,2026-01-15 07:00:00-05:00,2026-01-15,0,0.4899683,0,1,1,-1,1,True,False,False +2026-01-15 13:00:00,2026-01-15 08:00:00-05:00,2026-01-15,1,0.4946566,0,0,1,0,0,False,False,False +2026-01-15 14:00:00,2026-01-15 09:00:00-05:00,2026-01-15,0,0.4987126,0,0,1,0,0,True,True,True +2026-01-15 15:00:00,2026-01-15 10:00:00-05:00,2026-01-15,0,0.48068026,0,1,1,-1,1,True,False,False +2026-01-15 16:00:00,2026-01-15 11:00:00-05:00,2026-01-15,0,0.48770115,0,0,1,0,0,True,True,True +2026-01-15 17:00:00,2026-01-15 12:00:00-05:00,2026-01-15,1,0.4844149,0,0,0,0,0,False,False,False +2026-01-15 18:00:00,2026-01-15 13:00:00-05:00,2026-01-15,0,0.4631365,0,0,0,0,0,True,True,True +2026-01-15 19:00:00,2026-01-15 14:00:00-05:00,2026-01-15,1,0.45610657,0,0,0,0,0,False,False,False +2026-01-15 20:00:00,2026-01-15 15:00:00-05:00,2026-01-15,0,0.4881056,0,0,0,0,0,True,True,True +2026-01-15 21:00:00,2026-01-15 16:00:00-05:00,2026-01-15,0,0.46482596,0,0,0,0,0,True,True,True +2026-01-15 22:00:00,2026-01-15 17:00:00-05:00,2026-01-15,1,0.51136947,1,0,0,-1,0,True,False,False +2026-01-15 23:00:00,2026-01-15 18:00:00-05:00,2026-01-15,1,0.55429316,1,0,0,-1,0,True,False,False +2026-01-16 00:00:00,2026-01-15 19:00:00-05:00,2026-01-15,0,0.5584868,1,0,0,-1,0,False,False,True +2026-01-16 01:00:00,2026-01-15 20:00:00-05:00,2026-01-15,1,0.51578057,1,0,0,-1,0,True,False,False +2026-01-16 02:00:00,2026-01-15 21:00:00-05:00,2026-01-15,0,0.5562782,1,0,0,-1,0,False,False,True +2026-01-16 03:00:00,2026-01-15 22:00:00-05:00,2026-01-15,1,0.5080268,1,0,0,-1,0,True,False,False +2026-01-16 04:00:00,2026-01-15 23:00:00-05:00,2026-01-15,1,0.49140474,0,0,0,0,0,False,False,False +2026-01-16 05:00:00,2026-01-16 00:00:00-05:00,2026-01-16,1,0.46221796,0,0,0,0,0,False,False,False +2026-01-16 06:00:00,2026-01-16 01:00:00-05:00,2026-01-16,0,0.4448804,0,0,0,0,0,True,True,True +2026-01-16 07:00:00,2026-01-16 02:00:00-05:00,2026-01-16,1,0.4457024,0,0,0,0,0,False,False,False +2026-01-16 08:00:00,2026-01-16 03:00:00-05:00,2026-01-16,1,0.49323055,0,0,0,0,0,False,False,False +2026-01-16 09:00:00,2026-01-16 04:00:00-05:00,2026-01-16,0,0.49757826,0,0,0,0,0,True,True,True +2026-01-16 10:00:00,2026-01-16 05:00:00-05:00,2026-01-16,1,0.5067139,1,0,0,-1,0,True,False,False +2026-01-16 11:00:00,2026-01-16 06:00:00-05:00,2026-01-16,0,0.5102587,1,0,0,-1,0,False,False,True +2026-01-16 12:00:00,2026-01-16 07:00:00-05:00,2026-01-16,0,0.5843353,1,0,0,-1,0,False,False,True +2026-01-16 13:00:00,2026-01-16 08:00:00-05:00,2026-01-16,0,0.5425463,1,0,0,-1,0,False,False,True +2026-01-16 14:00:00,2026-01-16 09:00:00-05:00,2026-01-16,0,0.5690049,1,0,0,-1,0,False,False,True +2026-01-16 15:00:00,2026-01-16 10:00:00-05:00,2026-01-16,0,0.54829246,1,0,0,-1,0,False,False,True +2026-01-16 16:00:00,2026-01-16 11:00:00-05:00,2026-01-16,1,0.5046336,1,0,0,-1,0,True,False,False +2026-01-16 17:00:00,2026-01-16 12:00:00-05:00,2026-01-16,1,0.5086898,1,0,0,-1,0,True,False,False +2026-01-16 18:00:00,2026-01-16 13:00:00-05:00,2026-01-16,0,0.4823461,0,0,0,0,0,True,True,True +2026-01-16 19:00:00,2026-01-16 14:00:00-05:00,2026-01-16,1,0.4856351,0,0,0,0,0,False,False,False +2026-01-16 20:00:00,2026-01-16 15:00:00-05:00,2026-01-16,0,0.4558552,0,0,0,0,0,True,True,True +2026-01-16 21:00:00,2026-01-16 16:00:00-05:00,2026-01-16,0,0.4780688,0,0,0,0,0,True,True,True +2026-01-16 22:00:00,2026-01-16 17:00:00-05:00,2026-01-16,1,0.48367676,0,0,0,0,0,False,False,False +2026-01-16 23:00:00,2026-01-16 18:00:00-05:00,2026-01-16,1,0.5118516,1,0,0,-1,0,True,False,False +2026-01-17 00:00:00,2026-01-16 19:00:00-05:00,2026-01-16,0,0.5440252,1,0,0,-1,0,False,False,True +2026-01-17 01:00:00,2026-01-16 20:00:00-05:00,2026-01-16,0,0.52165735,1,1,0,1,1,False,False,False +2026-01-17 02:00:00,2026-01-16 21:00:00-05:00,2026-01-16,0,0.57716566,1,1,0,1,1,False,False,False +2026-01-17 03:00:00,2026-01-16 22:00:00-05:00,2026-01-16,1,0.53289396,1,1,0,1,1,True,True,True +2026-01-17 04:00:00,2026-01-16 23:00:00-05:00,2026-01-16,0,0.5077436,1,0,0,-1,0,False,False,True +2026-01-17 05:00:00,2026-01-17 00:00:00-05:00,2026-01-17,1,0.49432057,0,1,0,-1,0,False,False,False +2026-01-17 06:00:00,2026-01-17 01:00:00-05:00,2026-01-17,1,0.50985175,1,1,0,1,1,True,True,True +2026-01-17 07:00:00,2026-01-17 02:00:00-05:00,2026-01-17,1,0.4898242,0,1,0,-1,0,False,False,False +2026-01-17 08:00:00,2026-01-17 03:00:00-05:00,2026-01-17,1,0.45491692,0,1,0,-1,0,False,False,False +2026-01-17 09:00:00,2026-01-17 04:00:00-05:00,2026-01-17,0,0.45771605,0,1,0,-1,0,True,False,True +2026-01-17 10:00:00,2026-01-17 05:00:00-05:00,2026-01-17,0,0.45357528,0,1,0,-1,0,True,False,True +2026-01-17 11:00:00,2026-01-17 06:00:00-05:00,2026-01-17,1,0.4550014,0,1,0,-1,0,False,False,False +2026-01-17 12:00:00,2026-01-17 07:00:00-05:00,2026-01-17,1,0.46508005,0,1,0,-1,0,False,False,False +2026-01-17 13:00:00,2026-01-17 08:00:00-05:00,2026-01-17,0,0.4753723,0,1,0,-1,0,True,False,True +2026-01-17 14:00:00,2026-01-17 09:00:00-05:00,2026-01-17,0,0.49003577,0,0,0,0,0,True,True,True +2026-01-17 15:00:00,2026-01-17 10:00:00-05:00,2026-01-17,0,0.60734934,1,0,0,-1,0,False,False,True +2026-01-17 16:00:00,2026-01-17 11:00:00-05:00,2026-01-17,0,0.5786392,1,0,0,-1,0,False,False,True +2026-01-17 17:00:00,2026-01-17 12:00:00-05:00,2026-01-17,1,0.5754632,1,1,0,1,1,True,True,True +2026-01-17 18:00:00,2026-01-17 13:00:00-05:00,2026-01-17,0,0.59382933,1,1,0,1,1,False,False,False +2026-01-17 19:00:00,2026-01-17 14:00:00-05:00,2026-01-17,0,0.5593247,1,1,0,1,1,False,False,False +2026-01-17 20:00:00,2026-01-17 15:00:00-05:00,2026-01-17,1,0.5365127,1,1,0,1,1,True,True,True +2026-01-17 21:00:00,2026-01-17 16:00:00-05:00,2026-01-17,1,0.50084764,1,1,0,1,1,True,True,True +2026-01-17 22:00:00,2026-01-17 17:00:00-05:00,2026-01-17,1,0.46421826,0,1,0,-1,0,False,False,False +2026-01-17 23:00:00,2026-01-17 18:00:00-05:00,2026-01-17,0,0.48463318,0,1,0,-1,0,True,False,True +2026-01-18 00:00:00,2026-01-17 19:00:00-05:00,2026-01-17,1,0.46844795,0,1,0,-1,0,False,False,False +2026-01-18 01:00:00,2026-01-17 20:00:00-05:00,2026-01-17,1,0.48266262,0,1,0,-1,0,False,False,False +2026-01-18 02:00:00,2026-01-17 21:00:00-05:00,2026-01-17,0,0.49029487,0,1,0,-1,0,True,False,True +2026-01-18 03:00:00,2026-01-17 22:00:00-05:00,2026-01-17,0,0.45225728,0,1,0,-1,0,True,False,True +2026-01-18 04:00:00,2026-01-17 23:00:00-05:00,2026-01-17,0,0.5144419,1,1,0,1,1,False,False,False +2026-01-18 05:00:00,2026-01-18 00:00:00-05:00,2026-01-18,1,0.5160817,1,1,0,1,1,True,True,True +2026-01-18 06:00:00,2026-01-18 01:00:00-05:00,2026-01-18,1,0.50346553,1,1,0,1,1,True,True,True +2026-01-18 07:00:00,2026-01-18 02:00:00-05:00,2026-01-18,0,0.44028476,0,1,0,-1,0,True,False,True +2026-01-18 08:00:00,2026-01-18 03:00:00-05:00,2026-01-18,0,0.46246836,0,1,0,-1,0,True,False,True +2026-01-18 09:00:00,2026-01-18 04:00:00-05:00,2026-01-18,1,0.51622605,1,1,0,1,1,True,True,True +2026-01-18 10:00:00,2026-01-18 05:00:00-05:00,2026-01-18,1,0.51112276,1,1,0,1,1,True,True,True +2026-01-18 11:00:00,2026-01-18 06:00:00-05:00,2026-01-18,0,0.45434904,0,1,0,-1,0,True,False,True +2026-01-18 12:00:00,2026-01-18 07:00:00-05:00,2026-01-18,0,0.52511775,1,1,0,1,1,False,False,False +2026-01-18 13:00:00,2026-01-18 08:00:00-05:00,2026-01-18,1,0.5778929,1,1,0,1,1,True,True,True +2026-01-18 14:00:00,2026-01-18 09:00:00-05:00,2026-01-18,1,0.61074215,1,1,0,1,1,True,True,True +2026-01-18 15:00:00,2026-01-18 10:00:00-05:00,2026-01-18,0,0.57336044,1,1,0,1,1,False,False,False +2026-01-18 16:00:00,2026-01-18 11:00:00-05:00,2026-01-18,0,0.51572573,1,1,0,1,1,False,False,False +2026-01-18 17:00:00,2026-01-18 12:00:00-05:00,2026-01-18,0,0.5084302,1,0,0,-1,0,False,False,True +2026-01-18 18:00:00,2026-01-18 13:00:00-05:00,2026-01-18,0,0.53620636,1,1,0,1,1,False,False,False +2026-01-18 19:00:00,2026-01-18 14:00:00-05:00,2026-01-18,0,0.58746594,1,1,0,1,1,False,False,False +2026-01-18 20:00:00,2026-01-18 15:00:00-05:00,2026-01-18,0,0.53558993,1,1,0,1,1,False,False,False +2026-01-18 21:00:00,2026-01-18 16:00:00-05:00,2026-01-18,1,0.4962998,0,1,0,-1,0,False,False,False +2026-01-18 22:00:00,2026-01-18 17:00:00-05:00,2026-01-18,0,0.5750347,1,0,0,-1,0,False,False,True +2026-01-18 23:00:00,2026-01-18 18:00:00-05:00,2026-01-18,0,0.58212554,1,1,0,1,1,False,False,False +2026-01-19 00:00:00,2026-01-18 19:00:00-05:00,2026-01-18,1,0.5023791,1,1,0,1,1,True,True,True +2026-01-19 01:00:00,2026-01-18 20:00:00-05:00,2026-01-18,0,0.51338583,1,0,0,-1,0,False,False,True +2026-01-19 02:00:00,2026-01-18 21:00:00-05:00,2026-01-18,1,0.4790363,0,0,0,0,0,False,False,False +2026-01-19 03:00:00,2026-01-18 22:00:00-05:00,2026-01-18,0,0.4659165,0,0,0,0,0,True,True,True +2026-01-19 04:00:00,2026-01-18 23:00:00-05:00,2026-01-18,0,0.48015603,0,0,0,0,0,True,True,True +2026-01-19 05:00:00,2026-01-19 00:00:00-05:00,2026-01-19,1,0.4719194,0,0,0,0,0,False,False,False +2026-01-19 06:00:00,2026-01-19 01:00:00-05:00,2026-01-19,0,0.47719255,0,0,0,0,0,True,True,True +2026-01-19 07:00:00,2026-01-19 02:00:00-05:00,2026-01-19,0,0.46883187,0,0,0,0,0,True,True,True +2026-01-19 08:00:00,2026-01-19 03:00:00-05:00,2026-01-19,0,0.5041158,1,0,0,-1,0,False,False,True +2026-01-19 09:00:00,2026-01-19 04:00:00-05:00,2026-01-19,1,0.55327594,1,1,0,1,1,True,True,True +2026-01-19 10:00:00,2026-01-19 05:00:00-05:00,2026-01-19,1,0.51431626,1,1,0,1,1,True,True,True +2026-01-19 11:00:00,2026-01-19 06:00:00-05:00,2026-01-19,0,0.48845842,0,1,0,-1,0,True,False,True +2026-01-19 12:00:00,2026-01-19 07:00:00-05:00,2026-01-19,0,0.548243,1,1,0,1,1,False,False,False +2026-01-19 13:00:00,2026-01-19 08:00:00-05:00,2026-01-19,0,0.56993884,1,1,0,1,1,False,False,False +2026-01-19 14:00:00,2026-01-19 09:00:00-05:00,2026-01-19,1,0.570956,1,1,0,1,1,True,True,True +2026-01-19 15:00:00,2026-01-19 10:00:00-05:00,2026-01-19,1,0.58591735,1,1,0,1,1,True,True,True +2026-01-19 16:00:00,2026-01-19 11:00:00-05:00,2026-01-19,1,0.5654058,1,1,0,1,1,True,True,True +2026-01-19 17:00:00,2026-01-19 12:00:00-05:00,2026-01-19,1,0.53851956,1,1,0,1,1,True,True,True +2026-01-19 18:00:00,2026-01-19 13:00:00-05:00,2026-01-19,1,0.51866657,1,1,0,1,1,True,True,True +2026-01-19 19:00:00,2026-01-19 14:00:00-05:00,2026-01-19,1,0.4805421,0,1,0,-1,0,False,False,False +2026-01-19 20:00:00,2026-01-19 15:00:00-05:00,2026-01-19,0,0.4309969,0,0,0,0,0,True,True,True +2026-01-19 21:00:00,2026-01-19 16:00:00-05:00,2026-01-19,0,0.40472427,0,0,0,0,0,True,True,True +2026-01-19 22:00:00,2026-01-19 17:00:00-05:00,2026-01-19,0,0.43165132,0,1,0,-1,0,True,False,True +2026-01-19 23:00:00,2026-01-19 18:00:00-05:00,2026-01-19,0,0.45249504,0,1,0,-1,0,True,False,True +2026-01-20 00:00:00,2026-01-19 19:00:00-05:00,2026-01-19,1,0.41881704,0,1,0,-1,0,False,False,False +2026-01-20 01:00:00,2026-01-19 20:00:00-05:00,2026-01-19,0,0.4950903,0,1,0,-1,0,True,False,True +2026-01-20 02:00:00,2026-01-19 21:00:00-05:00,2026-01-19,0,0.52289253,1,1,0,1,1,False,False,False +2026-01-20 03:00:00,2026-01-19 22:00:00-05:00,2026-01-19,0,0.5152514,1,1,0,1,1,False,False,False +2026-01-20 04:00:00,2026-01-19 23:00:00-05:00,2026-01-19,1,0.52619386,1,1,0,1,1,True,True,True +2026-01-20 05:00:00,2026-01-20 00:00:00-05:00,2026-01-20,1,0.4920829,0,1,0,-1,0,False,False,False +2026-01-20 06:00:00,2026-01-20 01:00:00-05:00,2026-01-20,0,0.44386265,0,1,0,-1,0,True,False,True +2026-01-20 07:00:00,2026-01-20 02:00:00-05:00,2026-01-20,1,0.4548454,0,1,0,-1,0,False,False,False +2026-01-20 08:00:00,2026-01-20 03:00:00-05:00,2026-01-20,1,0.50409263,1,1,0,1,1,True,True,True +2026-01-20 09:00:00,2026-01-20 04:00:00-05:00,2026-01-20,1,0.4486903,0,1,0,-1,0,False,False,False +2026-01-20 10:00:00,2026-01-20 05:00:00-05:00,2026-01-20,0,0.44015622,0,1,0,-1,0,True,False,True +2026-01-20 11:00:00,2026-01-20 06:00:00-05:00,2026-01-20,1,0.46409336,0,1,0,-1,0,False,False,False +2026-01-20 12:00:00,2026-01-20 07:00:00-05:00,2026-01-20,0,0.47176668,0,1,0,-1,0,True,False,True +2026-01-20 13:00:00,2026-01-20 08:00:00-05:00,2026-01-20,0,0.44548807,0,1,0,-1,0,True,False,True +2026-01-20 14:00:00,2026-01-20 09:00:00-05:00,2026-01-20,1,0.45053762,0,1,0,-1,0,False,False,False +2026-01-20 15:00:00,2026-01-20 10:00:00-05:00,2026-01-20,1,0.481055,0,1,0,-1,0,False,False,False +2026-01-20 16:00:00,2026-01-20 11:00:00-05:00,2026-01-20,0,0.5467605,1,1,0,1,1,False,False,False +2026-01-20 17:00:00,2026-01-20 12:00:00-05:00,2026-01-20,0,0.5497743,1,1,0,1,1,False,False,False +2026-01-20 18:00:00,2026-01-20 13:00:00-05:00,2026-01-20,1,0.531139,1,0,0,-1,0,True,False,False +2026-01-20 19:00:00,2026-01-20 14:00:00-05:00,2026-01-20,1,0.48030645,0,0,0,0,0,False,False,False +2026-01-20 20:00:00,2026-01-20 15:00:00-05:00,2026-01-20,1,0.5283993,1,0,0,-1,0,True,False,False +2026-01-20 21:00:00,2026-01-20 16:00:00-05:00,2026-01-20,0,0.5042049,1,1,0,1,1,False,False,False +2026-01-20 22:00:00,2026-01-20 17:00:00-05:00,2026-01-20,1,0.46221802,0,1,0,-1,0,False,False,False +2026-01-20 23:00:00,2026-01-20 18:00:00-05:00,2026-01-20,1,0.45490173,0,1,0,-1,0,False,False,False +2026-01-21 00:00:00,2026-01-20 19:00:00-05:00,2026-01-20,1,0.4481478,0,1,0,-1,0,False,False,False +2026-01-21 01:00:00,2026-01-20 20:00:00-05:00,2026-01-20,0,0.4967737,0,1,0,-1,0,True,False,True +2026-01-21 02:00:00,2026-01-20 21:00:00-05:00,2026-01-20,1,0.5263554,1,1,0,1,1,True,True,True +2026-01-21 03:00:00,2026-01-20 22:00:00-05:00,2026-01-20,0,0.46894258,0,1,0,-1,0,True,False,True +2026-01-21 04:00:00,2026-01-20 23:00:00-05:00,2026-01-20,0,0.4519449,0,1,0,-1,0,True,False,True +2026-01-21 05:00:00,2026-01-21 00:00:00-05:00,2026-01-21,0,0.5209714,1,1,0,1,1,False,False,False +2026-01-21 06:00:00,2026-01-21 01:00:00-05:00,2026-01-21,1,0.56909335,1,0,0,-1,0,True,False,False +2026-01-21 07:00:00,2026-01-21 02:00:00-05:00,2026-01-21,1,0.51435,1,1,0,1,1,True,True,True +2026-01-21 08:00:00,2026-01-21 03:00:00-05:00,2026-01-21,0,0.47510275,0,1,0,-1,0,True,False,True +2026-01-21 09:00:00,2026-01-21 04:00:00-05:00,2026-01-21,1,0.47119588,0,1,0,-1,0,False,False,False +2026-01-21 10:00:00,2026-01-21 05:00:00-05:00,2026-01-21,0,0.45829907,0,1,0,-1,0,True,False,True +2026-01-21 11:00:00,2026-01-21 06:00:00-05:00,2026-01-21,0,0.46126115,0,1,0,-1,0,True,False,True +2026-01-21 12:00:00,2026-01-21 07:00:00-05:00,2026-01-21,1,0.49299902,0,1,0,-1,0,False,False,False +2026-01-21 13:00:00,2026-01-21 08:00:00-05:00,2026-01-21,1,0.4642857,0,1,0,-1,0,False,False,False +2026-01-21 14:00:00,2026-01-21 09:00:00-05:00,2026-01-21,0,0.46757126,0,1,0,-1,0,True,False,True +2026-01-21 15:00:00,2026-01-21 10:00:00-05:00,2026-01-21,1,0.51933724,1,0,0,-1,0,True,False,False +2026-01-21 16:00:00,2026-01-21 11:00:00-05:00,2026-01-21,0,0.52031684,1,0,0,-1,0,False,False,True +2026-01-21 17:00:00,2026-01-21 12:00:00-05:00,2026-01-21,1,0.4510923,0,1,0,-1,0,False,False,False +2026-01-21 18:00:00,2026-01-21 13:00:00-05:00,2026-01-21,1,0.47569612,0,0,0,0,0,False,False,False +2026-01-21 19:00:00,2026-01-21 14:00:00-05:00,2026-01-21,0,0.52261287,1,1,0,1,1,False,False,False +2026-01-21 20:00:00,2026-01-21 15:00:00-05:00,2026-01-21,0,0.5309036,1,0,0,-1,0,False,False,True +2026-01-21 21:00:00,2026-01-21 16:00:00-05:00,2026-01-21,1,0.5570105,1,1,0,1,1,True,True,True +2026-01-21 22:00:00,2026-01-21 17:00:00-05:00,2026-01-21,0,0.49707052,0,1,0,-1,0,True,False,True +2026-01-21 23:00:00,2026-01-21 18:00:00-05:00,2026-01-21,0,0.49758816,0,1,0,-1,0,True,False,True +2026-01-22 00:00:00,2026-01-21 19:00:00-05:00,2026-01-21,1,0.53757817,1,1,0,1,1,True,True,True +2026-01-22 01:00:00,2026-01-21 20:00:00-05:00,2026-01-21,1,0.57986003,1,1,0,1,1,True,True,True +2026-01-22 02:00:00,2026-01-21 21:00:00-05:00,2026-01-21,1,0.53082293,1,1,0,1,1,True,True,True +2026-01-22 03:00:00,2026-01-21 22:00:00-05:00,2026-01-21,1,0.50052387,1,1,0,1,1,True,True,True +2026-01-22 04:00:00,2026-01-21 23:00:00-05:00,2026-01-21,0,0.47824442,0,1,0,-1,0,True,False,True +2026-01-22 05:00:00,2026-01-22 00:00:00-05:00,2026-01-22,1,0.4760107,0,1,0,-1,0,False,False,False +2026-01-22 06:00:00,2026-01-22 01:00:00-05:00,2026-01-22,1,0.50084823,1,1,0,1,1,True,True,True +2026-01-22 07:00:00,2026-01-22 02:00:00-05:00,2026-01-22,1,0.45673624,0,1,0,-1,0,False,False,False +2026-01-22 08:00:00,2026-01-22 03:00:00-05:00,2026-01-22,0,0.44737244,0,1,0,-1,0,True,False,True +2026-01-22 09:00:00,2026-01-22 04:00:00-05:00,2026-01-22,0,0.4425855,0,0,0,0,0,True,True,True +2026-01-22 10:00:00,2026-01-22 05:00:00-05:00,2026-01-22,0,0.41352358,0,0,0,0,0,True,True,True +2026-01-22 11:00:00,2026-01-22 06:00:00-05:00,2026-01-22,0,0.47900316,0,1,0,-1,0,True,False,True +2026-01-22 12:00:00,2026-01-22 07:00:00-05:00,2026-01-22,1,0.5056432,1,0,0,-1,0,True,False,False +2026-01-22 13:00:00,2026-01-22 08:00:00-05:00,2026-01-22,0,0.5437085,1,1,0,1,1,False,False,False +2026-01-22 14:00:00,2026-01-22 09:00:00-05:00,2026-01-22,0,0.57784945,1,1,0,1,1,False,False,False +2026-01-22 15:00:00,2026-01-22 10:00:00-05:00,2026-01-22,1,0.55117744,1,1,0,1,1,True,True,True +2026-01-22 16:00:00,2026-01-22 11:00:00-05:00,2026-01-22,1,0.5499753,1,1,0,1,1,True,True,True +2026-01-22 17:00:00,2026-01-22 12:00:00-05:00,2026-01-22,1,0.48403728,0,0,0,0,0,False,False,False +2026-01-22 18:00:00,2026-01-22 13:00:00-05:00,2026-01-22,1,0.5378152,1,0,0,-1,0,True,False,False +2026-01-22 19:00:00,2026-01-22 14:00:00-05:00,2026-01-22,0,0.5124629,1,0,0,-1,0,False,False,True +2026-01-22 20:00:00,2026-01-22 15:00:00-05:00,2026-01-22,0,0.4786928,0,0,0,0,0,True,True,True +2026-01-22 21:00:00,2026-01-22 16:00:00-05:00,2026-01-22,0,0.47483838,0,1,0,-1,0,True,False,True +2026-01-22 22:00:00,2026-01-22 17:00:00-05:00,2026-01-22,1,0.4948368,0,1,0,-1,0,False,False,False +2026-01-22 23:00:00,2026-01-22 18:00:00-05:00,2026-01-22,0,0.50176257,1,1,0,1,1,False,False,False +2026-01-23 00:00:00,2026-01-22 19:00:00-05:00,2026-01-22,1,0.49785018,0,1,0,-1,0,False,False,False +2026-01-23 01:00:00,2026-01-22 20:00:00-05:00,2026-01-22,0,0.4707971,0,1,0,-1,0,True,False,True +2026-01-23 02:00:00,2026-01-22 21:00:00-05:00,2026-01-22,0,0.4584682,0,1,0,-1,0,True,False,True +2026-01-23 03:00:00,2026-01-22 22:00:00-05:00,2026-01-22,0,0.47485125,0,1,0,-1,0,True,False,True +2026-01-23 04:00:00,2026-01-22 23:00:00-05:00,2026-01-22,1,0.48357454,0,1,0,-1,0,False,False,False +2026-01-23 05:00:00,2026-01-23 00:00:00-05:00,2026-01-23,0,0.5074494,1,1,0,1,1,False,False,False +2026-01-23 06:00:00,2026-01-23 01:00:00-05:00,2026-01-23,1,0.49947768,0,1,0,-1,0,False,False,False +2026-01-23 07:00:00,2026-01-23 02:00:00-05:00,2026-01-23,1,0.5284809,1,1,0,1,1,True,True,True +2026-01-23 08:00:00,2026-01-23 03:00:00-05:00,2026-01-23,0,0.49399173,0,1,0,-1,0,True,False,True +2026-01-23 09:00:00,2026-01-23 04:00:00-05:00,2026-01-23,0,0.45683447,0,1,0,-1,0,True,False,True +2026-01-23 10:00:00,2026-01-23 05:00:00-05:00,2026-01-23,1,0.44759712,0,1,0,-1,0,False,False,False +2026-01-23 11:00:00,2026-01-23 06:00:00-05:00,2026-01-23,1,0.46368843,0,1,0,-1,0,False,False,False +2026-01-23 12:00:00,2026-01-23 07:00:00-05:00,2026-01-23,1,0.4522303,0,1,0,-1,0,False,False,False +2026-01-23 13:00:00,2026-01-23 08:00:00-05:00,2026-01-23,0,0.5273792,1,1,0,1,1,False,False,False +2026-01-23 14:00:00,2026-01-23 09:00:00-05:00,2026-01-23,1,0.49696228,0,1,0,-1,0,False,False,False +2026-01-23 15:00:00,2026-01-23 10:00:00-05:00,2026-01-23,0,0.5127735,1,1,0,1,1,False,False,False +2026-01-23 16:00:00,2026-01-23 11:00:00-05:00,2026-01-23,0,0.5816189,1,1,0,1,1,False,False,False +2026-01-23 17:00:00,2026-01-23 12:00:00-05:00,2026-01-23,0,0.58383,1,1,0,1,1,False,False,False +2026-01-23 18:00:00,2026-01-23 13:00:00-05:00,2026-01-23,1,0.5591273,1,1,0,1,1,True,True,True +2026-01-23 19:00:00,2026-01-23 14:00:00-05:00,2026-01-23,0,0.5854377,1,1,0,1,1,False,False,False +2026-01-23 20:00:00,2026-01-23 15:00:00-05:00,2026-01-23,0,0.5666198,1,1,0,1,1,False,False,False +2026-01-23 21:00:00,2026-01-23 16:00:00-05:00,2026-01-23,0,0.52965194,1,1,0,1,1,False,False,False +2026-01-23 22:00:00,2026-01-23 17:00:00-05:00,2026-01-23,0,0.5488747,1,1,0,1,1,False,False,False +2026-01-23 23:00:00,2026-01-23 18:00:00-05:00,2026-01-23,1,0.56132966,1,1,0,1,1,True,True,True +2026-01-24 00:00:00,2026-01-23 19:00:00-05:00,2026-01-23,1,0.5149501,1,1,0,1,1,True,True,True +2026-01-24 01:00:00,2026-01-23 20:00:00-05:00,2026-01-23,0,0.5064845,1,1,0,1,1,False,False,False +2026-01-24 02:00:00,2026-01-23 21:00:00-05:00,2026-01-23,1,0.46972445,0,1,0,-1,0,False,False,False +2026-01-24 03:00:00,2026-01-23 22:00:00-05:00,2026-01-23,1,0.47380736,0,1,0,-1,0,False,False,False +2026-01-24 04:00:00,2026-01-23 23:00:00-05:00,2026-01-23,0,0.46636635,0,1,0,-1,0,True,False,True +2026-01-24 05:00:00,2026-01-24 00:00:00-05:00,2026-01-24,0,0.44455072,0,1,0,-1,0,True,False,True +2026-01-24 06:00:00,2026-01-24 01:00:00-05:00,2026-01-24,0,0.4378892,0,1,0,-1,0,True,False,True +2026-01-24 07:00:00,2026-01-24 02:00:00-05:00,2026-01-24,0,0.49291152,0,1,0,-1,0,True,False,True +2026-01-24 08:00:00,2026-01-24 03:00:00-05:00,2026-01-24,0,0.5153457,1,1,0,1,1,False,False,False +2026-01-24 09:00:00,2026-01-24 04:00:00-05:00,2026-01-24,0,0.50489914,1,1,0,1,1,False,False,False +2026-01-24 10:00:00,2026-01-24 05:00:00-05:00,2026-01-24,1,0.5075944,1,1,0,1,1,True,True,True +2026-01-24 11:00:00,2026-01-24 06:00:00-05:00,2026-01-24,0,0.6229508,1,1,0,1,1,False,False,False +2026-01-24 12:00:00,2026-01-24 07:00:00-05:00,2026-01-24,0,0.5855234,1,1,0,1,1,False,False,False +2026-01-24 13:00:00,2026-01-24 08:00:00-05:00,2026-01-24,1,0.5532754,1,1,0,1,1,True,True,True +2026-01-24 14:00:00,2026-01-24 09:00:00-05:00,2026-01-24,1,0.5289959,1,0,0,-1,0,True,False,False +2026-01-24 15:00:00,2026-01-24 10:00:00-05:00,2026-01-24,1,0.49748245,0,0,0,0,0,False,False,False +2026-01-24 16:00:00,2026-01-24 11:00:00-05:00,2026-01-24,0,0.46435258,0,0,0,0,0,True,True,True +2026-01-24 17:00:00,2026-01-24 12:00:00-05:00,2026-01-24,0,0.44683295,0,1,0,-1,0,True,False,True +2026-01-24 18:00:00,2026-01-24 13:00:00-05:00,2026-01-24,1,0.45363894,0,1,0,-1,0,False,False,False +2026-01-24 19:00:00,2026-01-24 14:00:00-05:00,2026-01-24,1,0.5047467,1,0,0,-1,0,True,False,False +2026-01-24 20:00:00,2026-01-24 15:00:00-05:00,2026-01-24,1,0.44000623,0,0,0,0,0,False,False,False +2026-01-24 21:00:00,2026-01-24 16:00:00-05:00,2026-01-24,1,0.4814367,0,0,0,0,0,False,False,False +2026-01-24 22:00:00,2026-01-24 17:00:00-05:00,2026-01-24,0,0.46426073,0,0,0,0,0,True,True,True +2026-01-24 23:00:00,2026-01-24 18:00:00-05:00,2026-01-24,0,0.44996247,0,0,0,0,0,True,True,True +2026-01-25 00:00:00,2026-01-24 19:00:00-05:00,2026-01-24,1,0.47704104,0,1,0,-1,0,False,False,False +2026-01-25 01:00:00,2026-01-24 20:00:00-05:00,2026-01-24,0,0.52445495,1,0,0,-1,0,False,False,True +2026-01-25 02:00:00,2026-01-24 21:00:00-05:00,2026-01-24,0,0.4987614,0,1,0,-1,0,True,False,True +2026-01-25 03:00:00,2026-01-24 22:00:00-05:00,2026-01-24,1,0.5457757,1,1,0,1,1,True,True,True +2026-01-25 04:00:00,2026-01-24 23:00:00-05:00,2026-01-24,1,0.52091444,1,1,0,1,1,True,True,True +2026-01-25 05:00:00,2026-01-25 00:00:00-05:00,2026-01-25,0,0.5040618,1,1,0,1,1,False,False,False +2026-01-25 06:00:00,2026-01-25 01:00:00-05:00,2026-01-25,0,0.4632134,0,1,0,-1,0,True,False,True +2026-01-25 07:00:00,2026-01-25 02:00:00-05:00,2026-01-25,1,0.46248558,0,1,0,-1,0,False,False,False +2026-01-25 08:00:00,2026-01-25 03:00:00-05:00,2026-01-25,1,0.48469254,0,1,0,-1,0,False,False,False +2026-01-25 09:00:00,2026-01-25 04:00:00-05:00,2026-01-25,0,0.46292534,0,1,0,-1,0,True,False,True +2026-01-25 10:00:00,2026-01-25 05:00:00-05:00,2026-01-25,0,0.44414967,0,1,0,-1,0,True,False,True +2026-01-25 11:00:00,2026-01-25 06:00:00-05:00,2026-01-25,1,0.46307206,0,1,0,-1,0,False,False,False +2026-01-25 12:00:00,2026-01-25 07:00:00-05:00,2026-01-25,1,0.4829153,0,1,0,-1,0,False,False,False +2026-01-25 13:00:00,2026-01-25 08:00:00-05:00,2026-01-25,1,0.4569906,0,1,0,-1,0,False,False,False +2026-01-25 14:00:00,2026-01-25 09:00:00-05:00,2026-01-25,1,0.44465676,0,1,0,-1,0,False,False,False +2026-01-25 15:00:00,2026-01-25 10:00:00-05:00,2026-01-25,1,0.4582169,0,1,0,-1,0,False,False,False +2026-01-25 16:00:00,2026-01-25 11:00:00-05:00,2026-01-25,1,0.4422594,0,1,0,-1,0,False,False,False +2026-01-25 17:00:00,2026-01-25 12:00:00-05:00,2026-01-25,0,0.44241613,0,1,0,-1,0,True,False,True +2026-01-25 18:00:00,2026-01-25 13:00:00-05:00,2026-01-25,0,0.44766986,0,1,0,-1,0,True,False,True +2026-01-25 19:00:00,2026-01-25 14:00:00-05:00,2026-01-25,0,0.4483085,0,1,0,-1,0,True,False,True +2026-01-25 20:00:00,2026-01-25 15:00:00-05:00,2026-01-25,0,0.491638,0,1,0,-1,0,True,False,True +2026-01-25 21:00:00,2026-01-25 16:00:00-05:00,2026-01-25,0,0.49061048,0,1,0,-1,0,True,False,True +2026-01-25 22:00:00,2026-01-25 17:00:00-05:00,2026-01-25,0,0.5258992,1,1,0,1,1,False,False,False +2026-01-25 23:00:00,2026-01-25 18:00:00-05:00,2026-01-25,0,0.520721,1,1,0,1,1,False,False,False +2026-01-26 00:00:00,2026-01-25 19:00:00-05:00,2026-01-25,0,0.5159272,1,0,0,-1,0,False,False,True +2026-01-26 01:00:00,2026-01-25 20:00:00-05:00,2026-01-25,1,0.5431046,1,0,0,-1,0,True,False,False +2026-01-26 02:00:00,2026-01-25 21:00:00-05:00,2026-01-25,0,0.51656026,1,0,0,-1,0,False,False,True +2026-01-26 03:00:00,2026-01-25 22:00:00-05:00,2026-01-25,0,0.48795614,0,1,0,-1,0,True,False,True +2026-01-26 04:00:00,2026-01-25 23:00:00-05:00,2026-01-25,0,0.50463694,1,1,0,1,1,False,False,False +2026-01-26 05:00:00,2026-01-26 00:00:00-05:00,2026-01-26,1,0.5487247,1,0,0,-1,0,True,False,False +2026-01-26 06:00:00,2026-01-26 01:00:00-05:00,2026-01-26,1,0.5424091,1,0,0,-1,0,True,False,False +2026-01-26 07:00:00,2026-01-26 02:00:00-05:00,2026-01-26,0,0.47699133,0,0,0,0,0,True,True,True +2026-01-26 08:00:00,2026-01-26 03:00:00-05:00,2026-01-26,1,0.46807176,0,0,0,0,0,False,False,False +2026-01-26 09:00:00,2026-01-26 04:00:00-05:00,2026-01-26,1,0.5724977,1,1,0,1,1,True,True,True +2026-01-26 10:00:00,2026-01-26 05:00:00-05:00,2026-01-26,1,0.5247946,1,1,0,1,1,True,True,True +2026-01-26 11:00:00,2026-01-26 06:00:00-05:00,2026-01-26,0,0.46081114,0,0,0,0,0,True,True,True +2026-01-26 12:00:00,2026-01-26 07:00:00-05:00,2026-01-26,1,0.4425639,0,0,0,0,0,False,False,False +2026-01-26 13:00:00,2026-01-26 08:00:00-05:00,2026-01-26,1,0.47229946,0,1,0,-1,0,False,False,False +2026-01-26 14:00:00,2026-01-26 09:00:00-05:00,2026-01-26,1,0.49050438,0,1,0,-1,0,False,False,False +2026-01-26 15:00:00,2026-01-26 10:00:00-05:00,2026-01-26,0,0.49104828,0,1,0,-1,0,True,False,True +2026-01-26 16:00:00,2026-01-26 11:00:00-05:00,2026-01-26,0,0.45145187,0,1,0,-1,0,True,False,True +2026-01-26 17:00:00,2026-01-26 12:00:00-05:00,2026-01-26,0,0.4638353,0,0,0,0,0,True,True,True +2026-01-26 18:00:00,2026-01-26 13:00:00-05:00,2026-01-26,0,0.4618484,0,1,0,-1,0,True,False,True +2026-01-26 19:00:00,2026-01-26 14:00:00-05:00,2026-01-26,1,0.5018996,1,1,0,1,1,True,True,True +2026-01-26 20:00:00,2026-01-26 15:00:00-05:00,2026-01-26,1,0.5166753,1,1,0,1,1,True,True,True +2026-01-26 21:00:00,2026-01-26 16:00:00-05:00,2026-01-26,0,0.52520907,1,1,0,1,1,False,False,False +2026-01-26 22:00:00,2026-01-26 17:00:00-05:00,2026-01-26,0,0.5878727,1,1,0,1,1,False,False,False +2026-01-26 23:00:00,2026-01-26 18:00:00-05:00,2026-01-26,1,0.54239446,1,1,0,1,1,True,True,True +2026-01-27 00:00:00,2026-01-26 19:00:00-05:00,2026-01-26,0,0.53454095,1,1,0,1,1,False,False,False +2026-01-27 01:00:00,2026-01-26 20:00:00-05:00,2026-01-26,1,0.467306,0,1,0,-1,0,False,False,False +2026-01-27 02:00:00,2026-01-26 21:00:00-05:00,2026-01-26,0,0.46864188,0,1,0,-1,0,True,False,True +2026-01-27 03:00:00,2026-01-26 22:00:00-05:00,2026-01-26,1,0.46822935,0,1,0,-1,0,False,False,False +2026-01-27 04:00:00,2026-01-26 23:00:00-05:00,2026-01-26,0,0.49130705,0,1,0,-1,0,True,False,True +2026-01-27 05:00:00,2026-01-27 00:00:00-05:00,2026-01-27,0,0.46864477,0,1,0,-1,0,True,False,True +2026-01-27 06:00:00,2026-01-27 01:00:00-05:00,2026-01-27,1,0.486711,0,1,0,-1,0,False,False,False +2026-01-27 07:00:00,2026-01-27 02:00:00-05:00,2026-01-27,1,0.5504777,1,1,0,1,1,True,True,True +2026-01-27 08:00:00,2026-01-27 03:00:00-05:00,2026-01-27,0,0.4907592,0,1,0,-1,0,True,False,True +2026-01-27 09:00:00,2026-01-27 04:00:00-05:00,2026-01-27,0,0.4668024,0,1,0,-1,0,True,False,True +2026-01-27 10:00:00,2026-01-27 05:00:00-05:00,2026-01-27,0,0.50845945,1,1,0,1,1,False,False,False +2026-01-27 11:00:00,2026-01-27 06:00:00-05:00,2026-01-27,1,0.5608594,1,1,0,1,1,True,True,True +2026-01-27 12:00:00,2026-01-27 07:00:00-05:00,2026-01-27,0,0.51024324,1,1,0,1,1,False,False,False +2026-01-27 13:00:00,2026-01-27 08:00:00-05:00,2026-01-27,0,0.51232666,1,1,0,1,1,False,False,False +2026-01-27 14:00:00,2026-01-27 09:00:00-05:00,2026-01-27,0,0.4854087,0,1,0,-1,0,True,False,True +2026-01-27 15:00:00,2026-01-27 10:00:00-05:00,2026-01-27,0,0.48897353,0,1,0,-1,0,True,False,True +2026-01-27 16:00:00,2026-01-27 11:00:00-05:00,2026-01-27,0,0.46034786,0,1,0,-1,0,True,False,True +2026-01-27 17:00:00,2026-01-27 12:00:00-05:00,2026-01-27,1,0.5265957,1,1,0,1,1,True,True,True +2026-01-27 18:00:00,2026-01-27 13:00:00-05:00,2026-01-27,1,0.5300391,1,1,0,1,1,True,True,True +2026-01-27 19:00:00,2026-01-27 14:00:00-05:00,2026-01-27,0,0.49372414,0,1,0,-1,0,True,False,True +2026-01-27 20:00:00,2026-01-27 15:00:00-05:00,2026-01-27,1,0.4988915,0,1,0,-1,0,False,False,False +2026-01-27 21:00:00,2026-01-27 16:00:00-05:00,2026-01-27,1,0.53798217,1,1,0,1,1,True,True,True +2026-01-27 22:00:00,2026-01-27 17:00:00-05:00,2026-01-27,0,0.5018221,1,1,0,1,1,False,False,False +2026-01-27 23:00:00,2026-01-27 18:00:00-05:00,2026-01-27,0,0.49853957,0,1,0,-1,0,True,False,True +2026-01-28 00:00:00,2026-01-27 19:00:00-05:00,2026-01-27,1,0.5263078,1,1,0,1,1,True,True,True +2026-01-28 01:00:00,2026-01-27 20:00:00-05:00,2026-01-27,0,0.48831585,0,1,0,-1,0,True,False,True +2026-01-28 02:00:00,2026-01-27 21:00:00-05:00,2026-01-27,1,0.53863657,1,1,0,1,1,True,True,True +2026-01-28 03:00:00,2026-01-27 22:00:00-05:00,2026-01-27,0,0.5574659,1,1,0,1,1,False,False,False +2026-01-28 04:00:00,2026-01-27 23:00:00-05:00,2026-01-27,0,0.529262,1,1,0,1,1,False,False,False +2026-01-28 05:00:00,2026-01-28 00:00:00-05:00,2026-01-28,0,0.46035758,0,1,0,-1,0,True,False,True +2026-01-28 06:00:00,2026-01-28 01:00:00-05:00,2026-01-28,1,0.5224445,1,1,0,1,1,True,True,True +2026-01-28 07:00:00,2026-01-28 02:00:00-05:00,2026-01-28,0,0.51074743,1,1,0,1,1,False,False,False +2026-01-28 08:00:00,2026-01-28 03:00:00-05:00,2026-01-28,0,0.519096,1,1,0,1,1,False,False,False +2026-01-28 09:00:00,2026-01-28 04:00:00-05:00,2026-01-28,1,0.53050256,1,1,0,1,1,True,True,True +2026-01-28 10:00:00,2026-01-28 05:00:00-05:00,2026-01-28,1,0.54665875,1,1,0,1,1,True,True,True +2026-01-28 11:00:00,2026-01-28 06:00:00-05:00,2026-01-28,1,0.5373869,1,1,0,1,1,True,True,True +2026-01-28 12:00:00,2026-01-28 07:00:00-05:00,2026-01-28,1,0.53682816,1,1,0,1,1,True,True,True +2026-01-28 13:00:00,2026-01-28 08:00:00-05:00,2026-01-28,1,0.50733596,1,1,0,1,1,True,True,True +2026-01-28 14:00:00,2026-01-28 09:00:00-05:00,2026-01-28,0,0.466285,0,1,0,-1,0,True,False,True diff --git a/results/summary_report_2026_01.json b/results/summary_report_2026_01.json new file mode 100644 index 0000000..ab82145 --- /dev/null +++ b/results/summary_report_2026_01.json @@ -0,0 +1,58 @@ +{ + "测试期间": { + "开始日期": "2026-01-01 00:00:00", + "结束日期": "2026-01-30 00:00:00", + "总天数": 22, + "总样本数": 495 + }, + "单模型性能": { + "DLinear": { + "准确率": "53.33%", + "预测上涨比例": "47.1%" + }, + "N-BEATS": { + "准确率": "49.90%", + "预测上涨比例": "73.5%" + }, + "TTM-R2": { + "准确率": "51.31%", + "预测上涨比例": "23.6%" + } + }, + "集成策略性能": { + "双模型交集": { + "准确率": "53.39%", + "覆盖率": "47.68%", + "有效预测数": 236 + }, + "三模型投票": { + "准确率": "52.93%", + "覆盖率": "100.00%", + "有效预测数": 495 + } + }, + "DLinear概率分布": { + "0.4-0.5区间": "262个 (52.9%)", + "0.5-0.6区间": "229个 (46.3%)", + "0.6-0.7区间": "4个 (0.8%)", + "平均概率": "0.501" + }, + "最佳/最差交易日": { + "双模型最佳日": { + "日期": "2026-01-11", + "准确率": "83.3%" + }, + "双模型最差日": { + "日期": "2026-01-23", + "准确率": "28.6%" + }, + "三模型最佳日": { + "日期": "2026-01-28", + "准确率": "80.0%" + }, + "三模型最差日": { + "日期": "2026-01-23", + "准确率": "29.2%" + } + } +} \ No newline at end of file diff --git a/save_all_results.py b/save_all_results.py new file mode 100644 index 0000000..a59cbd2 --- /dev/null +++ b/save_all_results.py @@ -0,0 +1,458 @@ +"""保存所有测试结果到项目 - 生成完整报告""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from sklearn.metrics import confusion_matrix +from data.fetcher import BinanceDataFetcher +import json +from datetime import datetime + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def main(): + print("="*80) + print("保存完整测试结果到项目") + print("="*80) + + # 创建结果目录 + results_dir = "./results" + os.makedirs(results_dir, exist_ok=True) + + timestamp = datetime.now().strftime("%Y%m%d_%H%M%S") + + # ========== 加载模型 ========== + print("\n[1/5] 加载模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + print("✓ 三个模型已加载") + + # ========== 获取数据 ========== + print("\n[2/5] 获取2026年1月测试数据...") + fetcher = BinanceDataFetcher() + + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', + end_date='2026-01-30', use_cache=False) + + print(f"✓ 数据: {len(df_2026)} 行, {df_2026['timestamp'].min()} 至 {df_2026['timestamp'].max()}") + + # ========== 预处理和预测 ========== + print("\n[3/5] 预处理和预测...") + + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + + # DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits.flatten())) + dlinear_preds = (dlinear_probs > 0.5).astype(int) + + # N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) + + # TTM-R2预测 + context_length = 512 + ttm_preds_list = [] + all_prices = df_2026['close'].values + all_open_prices = df_2026['open'].values + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + context_data = all_prices[-context_length:].reshape(1, -1, 1) + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + with torch.no_grad(): + output = ttm_model(context_tensor) + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + # 双模型交集 + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_preds = np.full(len(dlinear_preds), -1) + two_model_preds[two_model_agree_up] = 1 + two_model_preds[two_model_agree_down] = 0 + + # 三模型投票 + vote_sum = dlinear_preds + nbeats_preds + ttm_preds + three_model_preds = np.where(vote_sum >= 2, 1, 0) + + true_labels = y_dlinear + + print(f"✓ 预测完成，样本数: {min_samples}") + + # ========== 生成详细结果 ========== + print("\n[4/5] 生成详细结果...") + + # 时间戳和日期 + timestamps = df_2026['timestamp'].iloc[168:168+min_samples].values + timestamps_utc = pd.to_datetime(timestamps) + timestamps_et = timestamps_utc.tz_localize('UTC').tz_convert('America/New_York') + dates = timestamps_et.date + + # 创建详细结果DataFrame + results_df = pd.DataFrame({ + 'timestamp_utc': timestamps_utc, + 'timestamp_et': timestamps_et, + 'date': dates, + 'true_label': true_labels, + 'dlinear_prob': dlinear_probs, + 'dlinear_pred': dlinear_preds, + 'nbeats_pred': nbeats_preds, + 'ttm_pred': ttm_preds, + 'two_model_pred': two_model_preds, + 'three_model_pred': three_model_preds, + 'dlinear_correct': dlinear_preds == true_labels, + 'two_model_correct': (two_model_preds == true_labels) & (two_model_preds != -1), + 'three_model_correct': three_model_preds == true_labels + }) + + # 按天统计 + daily_stats = results_df.groupby('date').agg({ + 'true_label': 'count', + 'two_model_pred': lambda x: (x != -1).sum(), + 'two_model_correct': 'sum', + 'three_model_correct': 'sum', + 'dlinear_correct': 'sum' + }) + + daily_stats.columns = ['total_samples', 'two_model_valid', 'two_model_correct', + 'three_model_correct', 'dlinear_correct'] + + daily_stats['dlinear_accuracy'] = daily_stats['dlinear_correct'] / daily_stats['total_samples'] * 100 + daily_stats['two_model_accuracy'] = daily_stats['two_model_correct'] / daily_stats['two_model_valid'] * 100 + daily_stats['three_model_accuracy'] = daily_stats['three_model_correct'] / daily_stats['total_samples'] * 100 + daily_stats['two_model_coverage'] = daily_stats['two_model_valid'] / daily_stats['total_samples'] * 100 + + daily_stats['two_model_accuracy'] = daily_stats['two_model_accuracy'].fillna(0) + + # ========== 保存结果 ========== + print("\n[5/5] 保存结果文件...") + + # 1. 保存逐小时详细结果 + hourly_file = f"{results_dir}/hourly_predictions_2026_01.csv" + results_df.to_csv(hourly_file, index=False) + print(f"✓ 保存逐小时预测: {hourly_file}") + + # 2. 保存每日统计 + daily_file = f"{results_dir}/daily_statistics_2026_01.csv" + daily_stats.to_csv(daily_file) + print(f"✓ 保存每日统计: {daily_file}") + + # 3. 生成总结报告 + summary = { + "测试期间": { + "开始日期": str(df_2026['timestamp'].min()), + "结束日期": str(df_2026['timestamp'].max()), + "总天数": len(daily_stats), + "总样本数": int(min_samples) + }, + "单模型性能": { + "DLinear": { + "准确率": f"{(dlinear_preds == true_labels).mean() * 100:.2f}%", + "预测上涨比例": f"{(dlinear_preds == 1).mean() * 100:.1f}%" + }, + "N-BEATS": { + "准确率": f"{(nbeats_preds == true_labels).mean() * 100:.2f}%", + "预测上涨比例": f"{(nbeats_preds == 1).mean() * 100:.1f}%" + }, + "TTM-R2": { + "准确率": f"{(ttm_preds == true_labels).mean() * 100:.2f}%", + "预测上涨比例": f"{(ttm_preds == 1).mean() * 100:.1f}%" + } + }, + "集成策略性能": { + "双模型交集": { + "准确率": f"{((two_model_preds == true_labels) & (two_model_preds != -1)).sum() / (two_model_preds != -1).sum() * 100:.2f}%", + "覆盖率": f"{(two_model_preds != -1).mean() * 100:.2f}%", + "有效预测数": int((two_model_preds != -1).sum()) + }, + "三模型投票": { + "准确率": f"{(three_model_preds == true_labels).mean() * 100:.2f}%", + "覆盖率": "100.00%", + "有效预测数": int(len(three_model_preds)) + } + }, + "DLinear概率分布": { + "0.4-0.5区间": f"{((dlinear_probs >= 0.4) & (dlinear_probs < 0.5)).sum()}个 ({((dlinear_probs >= 0.4) & (dlinear_probs < 0.5)).mean() * 100:.1f}%)", + "0.5-0.6区间": f"{((dlinear_probs >= 0.5) & (dlinear_probs < 0.6)).sum()}个 ({((dlinear_probs >= 0.5) & (dlinear_probs < 0.6)).mean() * 100:.1f}%)", + "0.6-0.7区间": f"{((dlinear_probs >= 0.6) & (dlinear_probs < 0.7)).sum()}个 ({((dlinear_probs >= 0.6) & (dlinear_probs < 0.7)).mean() * 100:.1f}%)", + "平均概率": f"{dlinear_probs.mean():.3f}" + }, + "最佳/最差交易日": { + "双模型最佳日": { + "日期": str(daily_stats['two_model_accuracy'].idxmax()), + "准确率": f"{daily_stats['two_model_accuracy'].max():.1f}%" + }, + "双模型最差日": { + "日期": str(daily_stats[daily_stats['two_model_accuracy'] > 0]['two_model_accuracy'].idxmin()), + "准确率": f"{daily_stats[daily_stats['two_model_accuracy'] > 0]['two_model_accuracy'].min():.1f}%" + }, + "三模型最佳日": { + "日期": str(daily_stats['three_model_accuracy'].idxmax()), + "准确率": f"{daily_stats['three_model_accuracy'].max():.1f}%" + }, + "三模型最差日": { + "日期": str(daily_stats['three_model_accuracy'].idxmin()), + "准确率": f"{daily_stats['three_model_accuracy'].min():.1f}%" + } + } + } + + summary_file = f"{results_dir}/summary_report_2026_01.json" + with open(summary_file, 'w', encoding='utf-8') as f: + json.dump(summary, f, indent=2, ensure_ascii=False) + print(f"✓ 保存总结报告: {summary_file}") + + # 4. 生成Markdown报告 + markdown_report = f"""# 加密货币价格预测模型测试报告 + +## 测试概况 + +- **测试期间**: 2026年1月1日 - 2026年1月30日 +- **总天数**: {len(daily_stats)}天 +- **总样本数**: {min_samples}个小时 +- **生成时间**: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')} + +--- + +## 单模型性能 + +### DLinear分类模型 +- **准确率**: {(dlinear_preds == true_labels).mean() * 100:.2f}% +- **预测上涨比例**: {(dlinear_preds == 1).mean() * 100:.1f}% +- **预测下跌比例**: {(dlinear_preds == 0).mean() * 100:.1f}% + +### N-BEATS回归模型 +- **准确率**: {(nbeats_preds == true_labels).mean() * 100:.2f}% +- **预测上涨比例**: {(nbeats_preds == 1).mean() * 100:.1f}% +- **预测下跌比例**: {(nbeats_preds == 0).mean() * 100:.1f}% + +### TTM-R2零样本模型 +- **准确率**: {(ttm_preds == true_labels).mean() * 100:.2f}% +- **预测上涨比例**: {(ttm_preds == 1).mean() * 100:.1f}% +- **预测下跌比例**: {(ttm_preds == 0).mean() * 100:.1f}% + +--- + +## 集成策略性能 + +### 双模型交集（DLinear + N-BEATS） +- **准确率**: {((two_model_preds == true_labels) & (two_model_preds != -1)).sum() / (two_model_preds != -1).sum() * 100:.2f}% +- **覆盖率**: {(two_model_preds != -1).mean() * 100:.2f}% +- **有效预测数**: {(two_model_preds != -1).sum()}次 + +### 三模型多数投票（DLinear + N-BEATS + TTM-R2） +- **准确率**: {(three_model_preds == true_labels).mean() * 100:.2f}% +- **覆盖率**: 100.00% +- **有效预测数**: {len(three_model_preds)}次 + +--- + +## DLinear概率分布分析 + +| 概率区间 | 样本数 | 占比 | +|---------|--------|------| +| 0.4-0.5 | {((dlinear_probs >= 0.4) & (dlinear_probs < 0.5)).sum()} | {((dlinear_probs >= 0.4) & (dlinear_probs < 0.5)).mean() * 100:.1f}% | +| 0.5-0.6 | {((dlinear_probs >= 0.5) & (dlinear_probs < 0.6)).sum()} | {((dlinear_probs >= 0.5) & (dlinear_probs < 0.6)).mean() * 100:.1f}% | +| 0.6-0.7 | {((dlinear_probs >= 0.6) & (dlinear_probs < 0.7)).sum()} | {((dlinear_probs >= 0.6) & (dlinear_probs < 0.7)).mean() * 100:.1f}% | + +**平均概率**: {dlinear_probs.mean():.3f} + +--- + +## 最佳/最差交易日 + +### 双模型交集 +- **最佳日**: {daily_stats['two_model_accuracy'].idxmax()} ({daily_stats['two_model_accuracy'].max():.1f}%) +- **最差日**: {daily_stats[daily_stats['two_model_accuracy'] > 0]['two_model_accuracy'].idxmin()} ({daily_stats[daily_stats['two_model_accuracy'] > 0]['two_model_accuracy'].min():.1f}%) + +### 三模型投票 +- **最佳日**: {daily_stats['three_model_accuracy'].idxmax()} ({daily_stats['three_model_accuracy'].max():.1f}%) +- **最差日**: {daily_stats['three_model_accuracy'].idxmin()} ({daily_stats['three_model_accuracy'].min():.1f}%) + +--- + +## 结论 + +1. **最佳单模型**: DLinear ({(dlinear_preds == true_labels).mean() * 100:.2f}%) +2. **最佳集成策略**: 双模型交集 ({((two_model_preds == true_labels) & (two_model_preds != -1)).sum() / (two_model_preds != -1).sum() * 100:.2f}%) +3. **DLinear置信度**: 99.2%的预测集中在0.45-0.55区间，缺乏高置信度预测 + +## 数据文件 + +- 逐小时预测结果: `hourly_predictions_2026_01.csv` +- 每日统计数据: `daily_statistics_2026_01.csv` +- 总结报告: `summary_report_2026_01.json` + +--- + +生成时间: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')} +""" + + markdown_file = f"{results_dir}/README_2026_01.md" + with open(markdown_file, 'w', encoding='utf-8') as f: + f.write(markdown_report) + print(f"✓ 保存Markdown报告: {markdown_file}") + + print("\n" + "="*80) + print("✓ 所有结果已保存到 ./results/ 目录") + print("="*80) + print(f"\n保存的文件:") + print(f" 1. {hourly_file}") + print(f" 2. {daily_file}") + print(f" 3. {summary_file}") + print(f" 4. {markdown_file}") + print("\n" + "="*80) + + +if __name__ == "__main__": + main() diff --git a/smart_rest_strategy.py b/smart_rest_strategy.py new file mode 100644 index 0000000..620ff86 --- /dev/null +++ b/smart_rest_strategy.py @@ -0,0 +1,451 @@ +"""智能休息策略回测：基于历史准确率动态决定是否交易""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import torch +import pickle +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +from data.fetcher import BinanceDataFetcher + + +# ==================== 预处理器定义 ==================== + +class ClassificationPreprocessor: + """DLinear分类预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.scaler = StandardScaler() + self.is_fitted = False + + def add_features(self, df): + df = df.copy() + df['rsi'] = ta.rsi(df['close'], length=14) + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + return df + + def create_labels(self, df): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) + return df[:-self.forecast_horizon] + + def preprocess(self, df, fit=True): + df = self.add_features(df) + df = self.create_labels(df) + df = df.dropna().reset_index(drop=True) + + feature_cols = ['open', 'high', 'low', 'close', 'volume', + 'rsi', 'macd', 'macd_hist', 'macd_signal'] + + if fit: + scaled_features = self.scaler.fit_transform(df[feature_cols]) + self.is_fitted = True + self.feature_columns = feature_cols + else: + scaled_features = self.scaler.transform(df[feature_cols]) + + X, y = [], [] + for i in range(len(scaled_features) - self.lookback): + X.append(scaled_features[i:i + self.lookback]) + y.append(df['label'].iloc[i + self.lookback]) + + return np.array(X), np.array(y), df + + +class NBeatsPreprocessor: + """N-BEATS预处理器""" + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + df = df.copy() + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def main(): + print("="*80) + print("智能休息策略回测：基于历史准确率动态决定是否交易") + print("="*80) + print("\n策略规则：") + print(" 1. 如果前一天准确率超过70% → 当天休息") + print(" 2. 如果连续两天准确率超过60% → 第三天休息") + print(" 3. 其他情况 → 正常交易") + print("="*80) + + # ========== 加载模型 ========== + print("\n[1/4] 加载模型...") + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + + # 加载DLinear + from models.dlinear.model import DLinear + dlinear_checkpoint = torch.load( + './models/checkpoints/dlinear_classification_2026/best_model.pt', + map_location=device + ) + dlinear_model = DLinear(dlinear_checkpoint['config']) + dlinear_model.load_state_dict(dlinear_checkpoint['model_state_dict']) + dlinear_model = dlinear_model.to(device) + dlinear_model.eval() + + with open('./models/checkpoints/dlinear_classification_2026/preprocessor.pkl', 'rb') as f: + dlinear_preprocessor = pickle.load(f) + + # 加载N-BEATS + from models.nbeats.model import NBeats + nbeats_checkpoint = torch.load( + './models/checkpoints/nbeats_btc_1h/best_model.pt', + map_location=device + ) + nbeats_model = NBeats(nbeats_checkpoint['config']) + nbeats_model.load_state_dict(nbeats_checkpoint['model_state_dict']) + nbeats_model = nbeats_model.to(device) + nbeats_model.eval() + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'rb') as f: + nbeats_preprocessor = pickle.load(f) + + # 加载TTM-R2 + from tsfm_public.models.tinytimemixer import TinyTimeMixerForPrediction + ttm_model = TinyTimeMixerForPrediction.from_pretrained( + "ibm-granite/granite-timeseries-ttm-r2", + trust_remote_code=True + ) + ttm_model = ttm_model.to(device) + ttm_model.eval() + + print(f"✓ 三个模型已加载") + + # ========== 获取数据 ========== + print("\n[2/4] 获取2026年1月测试数据（1月1日-1月30日）...") + fetcher = BinanceDataFetcher() + + df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', + end_date='2026-01-30', use_cache=False) + + print(f"✓ 测试数据: {len(df_2026)} 行") + print(f"✓ 时间范围: {df_2026['timestamp'].min()} 至 {df_2026['timestamp'].max()}") + + # ========== 预处理 ========== + print("\n[3/4] 预处理数据...") + + X_dlinear, y_dlinear, df_dlinear = dlinear_preprocessor.preprocess(df_2026, fit=False) + X_nbeats, y_nbeats, df_nbeats = nbeats_preprocessor.preprocess(df_2026, fit=False) + + min_samples = min(len(X_dlinear), len(X_nbeats)) + X_dlinear = X_dlinear[:min_samples] + y_dlinear = y_dlinear[:min_samples] + X_nbeats = X_nbeats[:min_samples] + y_nbeats = y_nbeats[:min_samples] + + # ========== 预测 ========== + print("\n[4/4] 执行预测...") + + # DLinear预测 + X_dlinear_tensor = torch.FloatTensor(X_dlinear).to(device) + with torch.no_grad(): + dlinear_logits = dlinear_model(X_dlinear_tensor).cpu().numpy() + dlinear_probs = 1 / (1 + np.exp(-dlinear_logits)) + dlinear_preds = (dlinear_probs > 0.5).astype(int).flatten() + + # N-BEATS预测 + X_nbeats_tensor = torch.FloatTensor(X_nbeats).to(device) + with torch.no_grad(): + nbeats_pred_scaled = nbeats_model(X_nbeats_tensor).cpu().numpy() + + nbeats_pred_prices = nbeats_preprocessor.inverse_transform_price(nbeats_pred_scaled) + current_open_prices = df_nbeats['open'].iloc[168:168+len(nbeats_pred_prices)].values + nbeats_preds = (nbeats_pred_prices > current_open_prices).astype(int) + + # TTM-R2预测 + context_length = 512 + ttm_preds_list = [] + all_prices = df_2026['close'].values + all_open_prices = df_2026['open'].values + + for i in range(min_samples): + start_idx = i + end_idx = start_idx + context_length + + if end_idx > len(all_prices): + context_data = all_prices[-context_length:].reshape(1, -1, 1) + current_open_price = all_open_prices[min(168 + i, len(all_open_prices) - 1)] + else: + context_data = all_prices[start_idx:end_idx].reshape(1, -1, 1) + current_open_price = all_open_prices[end_idx] if end_idx < len(all_open_prices) else all_open_prices[-1] + + context_tensor = torch.FloatTensor(context_data).to(device) + + with torch.no_grad(): + output = ttm_model(context_tensor) + pred_price = output.prediction_outputs.cpu().numpy()[0, 0, 0] + + ttm_preds_list.append(1 if pred_price > current_open_price else 0) + + ttm_preds = np.array(ttm_preds_list) + + # 双模型交集 + two_model_agree_up = (dlinear_preds == 1) & (nbeats_preds == 1) + two_model_agree_down = (dlinear_preds == 0) & (nbeats_preds == 0) + two_model_preds = np.full(len(dlinear_preds), -1) + two_model_preds[two_model_agree_up] = 1 + two_model_preds[two_model_agree_down] = 0 + + # 三模型多数投票 + vote_sum = dlinear_preds + nbeats_preds + ttm_preds + three_model_preds = np.where(vote_sum >= 2, 1, 0) + + # 真实标签 + true_labels = y_dlinear + + # ========== 构建每日数据 ========== + timestamps = df_2026['timestamp'].iloc[168:168+min_samples].values + timestamps_utc = pd.to_datetime(timestamps) + timestamps_et = timestamps_utc.tz_localize('UTC').tz_convert('America/New_York') + dates = timestamps_et.date + + results_df = pd.DataFrame({ + 'date': dates, + 'timestamp_et': timestamps_et, + 'true_label': true_labels, + 'two_model_pred': two_model_preds, + 'three_model_pred': three_model_preds, + 'two_model_correct': (two_model_preds == true_labels) & (two_model_preds != -1), + 'three_model_correct': (three_model_preds == true_labels) + }) + + # 按天统计基础数据 + daily_stats = results_df.groupby('date').agg({ + 'true_label': 'count', + 'two_model_pred': lambda x: (x != -1).sum(), + 'two_model_correct': 'sum', + 'three_model_correct': 'sum' + }) + + daily_stats.columns = ['total_samples', 'two_model_valid', 'two_model_correct', 'three_model_correct'] + + # 计算每日准确率 + daily_stats['two_model_accuracy'] = daily_stats['two_model_correct'] / daily_stats['two_model_valid'] * 100 + daily_stats['three_model_accuracy'] = daily_stats['three_model_correct'] / daily_stats['total_samples'] * 100 + daily_stats['two_model_accuracy'] = daily_stats['two_model_accuracy'].fillna(0) + + # ========== 应用智能休息策略 ========== + print("\n" + "="*80) + print("📊 应用智能休息策略") + print("="*80) + + # 为每一天添加"是否休息"标记 + daily_stats['two_model_rest'] = False + daily_stats['three_model_rest'] = False + daily_stats['rest_reason'] = '' + + dates_list = daily_stats.index.tolist() + + for i, date in enumerate(dates_list): + if i == 0: + continue # 第一天没有历史数据，不休息 + + # 获取前一天和前两天的数据 + prev_date = dates_list[i-1] + prev2_date = dates_list[i-2] if i >= 2 else None + + # 双模型休息逻辑 + # 规则1：前一天准确率超过70% + if daily_stats.loc[prev_date, 'two_model_accuracy'] > 70: + daily_stats.loc[date, 'two_model_rest'] = True + daily_stats.loc[date, 'rest_reason'] = f'前日准确率{daily_stats.loc[prev_date, "two_model_accuracy"]:.1f}%>70%' + + # 规则2：连续两天准确率超过60% + elif (prev2_date is not None and + daily_stats.loc[prev_date, 'two_model_accuracy'] > 60 and + daily_stats.loc[prev2_date, 'two_model_accuracy'] > 60): + daily_stats.loc[date, 'two_model_rest'] = True + daily_stats.loc[date, 'rest_reason'] = f'连续2天>60% ({daily_stats.loc[prev2_date, "two_model_accuracy"]:.1f}%, {daily_stats.loc[prev_date, "two_model_accuracy"]:.1f}%)' + + # 三模型休息逻辑 + # 规则1：前一天准确率超过70% + if daily_stats.loc[prev_date, 'three_model_accuracy'] > 70: + daily_stats.loc[date, 'three_model_rest'] = True + if not daily_stats.loc[date, 'rest_reason']: + daily_stats.loc[date, 'rest_reason'] = f'三模前日{daily_stats.loc[prev_date, "three_model_accuracy"]:.1f}%>70%' + + # 规则2：连续两天准确率超过60% + elif (prev2_date is not None and + daily_stats.loc[prev_date, 'three_model_accuracy'] > 60 and + daily_stats.loc[prev2_date, 'three_model_accuracy'] > 60): + daily_stats.loc[date, 'three_model_rest'] = True + if not daily_stats.loc[date, 'rest_reason']: + daily_stats.loc[date, 'rest_reason'] = f'三模连续2天>60%' + + # ========== 统计休息天数 ========== + two_model_rest_days = daily_stats['two_model_rest'].sum() + three_model_rest_days = daily_stats['three_model_rest'].sum() + total_days = len(daily_stats) + + print(f"\n休息天数统计：") + print(f" 总交易日: {total_days}天") + print(f" 双模型休息: {two_model_rest_days}天 ({two_model_rest_days/total_days*100:.1f}%)") + print(f" 三模型休息: {three_model_rest_days}天 ({three_model_rest_days/total_days*100:.1f}%)") + + # ========== 计算休息策略的实际表现 ========== + print("\n" + "="*80) + print("📈 策略对比：正常交易 vs 智能休息") + print("="*80) + + # 双模型策略 + print(f"\n🔵 双模型交集策略：") + + # 正常交易（不休息） + two_normal_correct = daily_stats['two_model_correct'].sum() + two_normal_total = daily_stats['two_model_valid'].sum() + two_normal_acc = two_normal_correct / two_normal_total * 100 if two_normal_total > 0 else 0 + + print(f"\n 📊 正常交易（不休息）：") + print(f" 交易天数: {total_days}天") + print(f" 有效预测: {two_normal_total}次") + print(f" 正确预测: {two_normal_correct}次") + print(f" 准确率: {two_normal_acc:.2f}%") + + # 智能休息 + two_rest_mask = ~daily_stats['two_model_rest'] + two_rest_correct = daily_stats.loc[two_rest_mask, 'two_model_correct'].sum() + two_rest_total = daily_stats.loc[two_rest_mask, 'two_model_valid'].sum() + two_rest_acc = two_rest_correct / two_rest_total * 100 if two_rest_total > 0 else 0 + two_trade_days = two_rest_mask.sum() + + print(f"\n 🛌 智能休息策略：") + print(f" 交易天数: {two_trade_days}天 (休息{two_model_rest_days}天)") + print(f" 有效预测: {two_rest_total}次") + print(f" 正确预测: {two_rest_correct}次") + print(f" 准确率: {two_rest_acc:.2f}%") + + two_acc_diff = two_rest_acc - two_normal_acc + two_diff_mark = "✅" if two_acc_diff > 0 else ("⚠️" if two_acc_diff < 0 else "") + print(f"\n 准确率变化: {two_acc_diff:+.2f}% {two_diff_mark}") + + # 休息日的实际表现 + if two_model_rest_days > 0: + two_rest_days_mask = daily_stats['two_model_rest'] + two_rest_day_correct = daily_stats.loc[two_rest_days_mask, 'two_model_correct'].sum() + two_rest_day_total = daily_stats.loc[two_rest_days_mask, 'two_model_valid'].sum() + two_rest_day_acc = two_rest_day_correct / two_rest_day_total * 100 if two_rest_day_total > 0 else 0 + + print(f"\n 🔍 被休息的{two_model_rest_days}天实际表现:") + print(f" 准确率: {two_rest_day_acc:.2f}%") + avoided_loss = "避免了低准确率" if two_rest_day_acc < two_normal_acc else "⚠️ 错过了好机会" + print(f" 评价: {avoided_loss}") + + # 三模型策略 + print(f"\n🟢 三模型投票策略：") + + # 正常交易 + three_normal_correct = daily_stats['three_model_correct'].sum() + three_normal_total = daily_stats['total_samples'].sum() + three_normal_acc = three_normal_correct / three_normal_total * 100 + + print(f"\n 📊 正常交易（不休息）：") + print(f" 交易天数: {total_days}天") + print(f" 有效预测: {three_normal_total}次") + print(f" 正确预测: {three_normal_correct}次") + print(f" 准确率: {three_normal_acc:.2f}%") + + # 智能休息 + three_rest_mask = ~daily_stats['three_model_rest'] + three_rest_correct = daily_stats.loc[three_rest_mask, 'three_model_correct'].sum() + three_rest_total = daily_stats.loc[three_rest_mask, 'total_samples'].sum() + three_rest_acc = three_rest_correct / three_rest_total * 100 if three_rest_total > 0 else 0 + three_trade_days = three_rest_mask.sum() + + print(f"\n 🛌 智能休息策略：") + print(f" 交易天数: {three_trade_days}天 (休息{three_model_rest_days}天)") + print(f" 有效预测: {three_rest_total}次") + print(f" 正确预测: {three_rest_correct}次") + print(f" 准确率: {three_rest_acc:.2f}%") + + three_acc_diff = three_rest_acc - three_normal_acc + three_diff_mark = "✅" if three_acc_diff > 0 else ("⚠️" if three_acc_diff < 0 else "") + print(f"\n 准确率变化: {three_acc_diff:+.2f}% {three_diff_mark}") + + # 休息日的实际表现 + if three_model_rest_days > 0: + three_rest_days_mask = daily_stats['three_model_rest'] + three_rest_day_correct = daily_stats.loc[three_rest_days_mask, 'three_model_correct'].sum() + three_rest_day_total = daily_stats.loc[three_rest_days_mask, 'total_samples'].sum() + three_rest_day_acc = three_rest_day_correct / three_rest_day_total * 100 + + print(f"\n 🔍 被休息的{three_model_rest_days}天实际表现:") + print(f" 准确率: {three_rest_day_acc:.2f}%") + avoided_loss = "避免了低准确率" if three_rest_day_acc < three_normal_acc else "⚠️ 错过了好机会" + print(f" 评价: {avoided_loss}") + + # ========== 详细日志 ========== + print("\n" + "="*80) + print("📅 每日详细记录") + print("="*80) + + print(f"\n{'日期':<12} {'双模准确率':>10} {'三模准确率':>10} {'休息状态':<12} {'休息原因':<40}") + print("-" * 95) + + for date, row in daily_stats.iterrows(): + two_acc = row['two_model_accuracy'] + three_acc = row['three_model_accuracy'] + + rest_status = "" + if row['two_model_rest']: + rest_status += "双休 " + if row['three_model_rest']: + rest_status += "三休" + + if not rest_status: + rest_status = "正常交易" + + # 准确率标记 + two_mark = "🔥" if two_acc >= 70 else ("✅" if two_acc >= 60 else "") + three_mark = "🔥" if three_acc >= 70 else ("✅" if three_acc >= 60 else "") + + print(f"{date} {two_acc:>9.1f}{two_mark} {three_acc:>9.1f}{three_mark} " + f"{rest_status:<12} {row['rest_reason']:<40}") + + print("\n" + "="*80) + print("✓ 智能休息策略回测完成!") + print("="*80) + + +if __name__ == "__main__": + main() diff --git a/test_2026_classification.py b/test_2026_classification.py deleted file mode 100644 index 622ce43..0000000 --- a/test_2026_classification.py +++ /dev/null @@ -1,232 +0,0 @@ -"""在2026年数据上测试分类模型""" - -import sys -import os -sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) - -import torch -import pickle -import numpy as np -import pandas as pd -import pandas_ta as ta -from sklearn.preprocessing import StandardScaler -from sklearn.metrics import confusion_matrix, classification_report -from data.fetcher import BinanceDataFetcher -from models.dlinear.model import DLinear - -class ClassificationPreprocessor: - """分类任务预处理器""" - def __init__(self, lookback=168, forecast_horizon=1): - self.lookback = lookback - self.forecast_horizon = forecast_horizon - self.scaler = StandardScaler() - self.is_fitted = False - - def add_features(self, df): - df = df.copy() - df['rsi'] = ta.rsi(df['close'], length=14) - macd = ta.macd(df['close']) - if macd is not None and not macd.empty: - df['macd'] = macd.iloc[:, 0] - df['macd_hist'] = macd.iloc[:, 1] - df['macd_signal'] = macd.iloc[:, 2] - return df - - def create_labels(self, df): - df = df.copy() - df['future_price'] = df['close'].shift(-self.forecast_horizon) - df['future_return'] = (df['future_price'] - df['close']) / df['close'] - df['label'] = (df['future_return'] > 0).astype(int) - return df[:-self.forecast_horizon] - - def preprocess(self, df, fit=True): - df = self.add_features(df) - df = self.create_labels(df) - df = df.dropna().reset_index(drop=True) - - feature_cols = ['open', 'high', 'low', 'close', 'volume', - 'rsi', 'macd', 'macd_hist', 'macd_signal'] - - if fit: - scaled_features = self.scaler.fit_transform(df[feature_cols]) - self.is_fitted = True - self.feature_columns = feature_cols - else: - if not self.is_fitted: - raise ValueError("Must fit before transform") - scaled_features = self.scaler.transform(df[feature_cols]) - - X, y = [], [] - for i in range(len(scaled_features) - self.lookback): - X.append(scaled_features[i:i + self.lookback]) - y.append(df['label'].iloc[i + self.lookback]) - - return np.array(X), np.array(y), df - -def main(): - print("="*80) - print("2026年分类模型测试") - print("="*80) - - # ========== 加载模型 ========== - print("\n[1/4] 加载模型...") - model_path = './models/checkpoints/dlinear_classification_2026/best_model.pt' - preprocessor_path = './models/checkpoints/dlinear_classification_2026/preprocessor.pkl' - - device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') - checkpoint = torch.load(model_path, map_location=device, weights_only=False) - model = DLinear(checkpoint['config']) - model.load_state_dict(checkpoint['model_state_dict']) - model = model.to(device) - model.eval() - - with open(preprocessor_path, 'rb') as f: - preprocessor = pickle.load(f) - - print(f"✓ 模型加载成功") - print(f"✓ 任务类型: 分类（预测涨/跌）") - - # ========== 获取2026年数据 ========== - print("\n[2/4] 获取2026年测试数据...") - fetcher = BinanceDataFetcher() - - try: - df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2026-01-01', - end_date='2026-12-31', use_cache=False) - except: - df_2026 = fetcher.fetch_ohlcv('BTC', '1h', start_date='2025-12-01', use_cache=False) - - print(f"✓ 测试数据: {len(df_2026)} 行") - print(f"✓ 时间: {df_2026['timestamp'].min()} 至 {df_2026['timestamp'].max()}") - print(f"✓ 价格: ${df_2026['close'].iloc[0]:,.2f} → ${df_2026['close'].iloc[-1]:,.2f}") - - # ========== 预处理 ========== - print("\n[3/4] 预处理并预测...") - X_test, y_test, processed_df = preprocessor.preprocess(df_2026, fit=False) - - if len(X_test) == 0: - print("❌ 数据不足") - return - - print(f"✓ 测试序列: {len(X_test)}") - - # 真实标签统计 - n_true_up = (y_test == 1).sum() - n_true_down = (y_test == 0).sum() - print(f"✓ 实际涨跌: 上涨{n_true_up}次 ({n_true_up/len(y_test)*100:.1f}%), " - f"下跌{n_true_down}次 ({n_true_down/len(y_test)*100:.1f}%)") - - # 模型预测 - X_tensor = torch.FloatTensor(X_test).to(device) - with torch.no_grad(): - logits = model(X_tensor).squeeze().cpu().numpy() - probs = 1 / (1 + np.exp(-logits)) # sigmoid - preds = (probs > 0.5).astype(int) - - # ========== 评估 ========== - print("\n[4/4] 评估结果...") - - # 基础统计 - n_pred_up = (preds == 1).sum() - n_pred_down = (preds == 0).sum() - accuracy = (preds == y_test).mean() - - # 混淆矩阵 - cm = confusion_matrix(y_test, preds) - tn, fp, fn, tp = cm.ravel() - - print("\n" + "="*80) - print("📊 2026年测试结果（分类任务）") - print("="*80) - - print(f"\n预测统计:") - print(f" 预测上涨: {n_pred_up} ({n_pred_up/len(preds)*100:.1f}%)") - print(f" 预测下跌: {n_pred_down} ({n_pred_down/len(preds)*100:.1f}%)") - - print(f"\n实际统计:") - print(f" 实际上涨: {n_true_up} ({n_true_up/len(y_test)*100:.1f}%)") - print(f" 实际下跌: {n_true_down} ({n_true_down/len(y_test)*100:.1f}%)") - - print(f"\n✨ 准确率: {accuracy*100:.2f}%") - - print(f"\n详细分析:") - print(f" ✓ 预测涨且真涨 (TP): {tp} ({tp/len(preds)*100:.1f}%)") - print(f" ✓ 预测跌且真跌 (TN): {tn} ({tn/len(preds)*100:.1f}%)") - print(f" ✗ 预测涨但真跌 (FP): {fp} ({fp/len(preds)*100:.1f}%)") - print(f" ✗ 预测跌但真涨 (FN): {fn} ({fn/len(preds)*100:.1f}%)") - - # 精确率、召回率、F1 - precision = tp / (tp + fp) if (tp + fp) > 0 else 0 - recall = tp / (tp + fn) if (tp + fn) > 0 else 0 - f1 = 2 * precision * recall / (precision + recall) if (precision + recall) > 0 else 0 - - print(f"\n分类指标:") - print(f" 精确率 (Precision): {precision*100:.2f}%") - print(f" 召回率 (Recall): {recall*100:.2f}%") - print(f" F1分数: {f1:.4f}") - - # 预测偏差分析 - bias = n_pred_up - n_true_up - print(f"\n预测偏差:") - if abs(bias) < len(preds) * 0.05: - print(f" ✅ 预测平衡（偏差{bias}个）") - elif bias > 0: - print(f" ⚠️ 偏向预测上涨（多预测{bias}次上涨）") - else: - print(f" ⚠️ 偏向预测下跌（多预测{-bias}次下跌）") - - # 性能评估 - print("\n" + "="*80) - print("🎯 性能评估") - print("="*80) - - if accuracy > 0.55: - print(f"\n ✅ 准确率 {accuracy*100:.2f}% > 55%，模型有预测能力！") - elif accuracy > 0.52: - print(f"\n ⚠️ 准确率 {accuracy*100:.2f}% > 52%，略优于随机") - elif accuracy >= 0.48: - print(f"\n ⚠️ 准确率 {accuracy*100:.2f}% ≈ 50%，接近随机") - else: - print(f"\n ❌ 准确率 {accuracy*100:.2f}% < 48%，低于随机") - - # 预测样本展示 - print("\n" + "="*80) - print("📋 预测样本展示（前30个）") - print("="*80) - print(f"\n{'时间':<20} {'预测概率':<10} {'预测':<6} {'实际':<6} {'结果':<4}") - print("-"*60) - - for i in range(min(30, len(preds))): - timestamp = processed_df['timestamp'].iloc[i + 168] - prob = probs[i] - pred = "上涨" if preds[i] == 1 else "下跌" - true = "上涨" if y_test[i] == 1 else "下跌" - result = "✓" if preds[i] == y_test[i] else "✗" - - print(f"{str(timestamp):<20} {prob:>8.4f} {pred:<6} {true:<6} {result:<4}") - - print("\n" + "="*80) - print(f"总结: 在{len(preds)}次预测中，准确率 {accuracy*100:.2f}%") - print("="*80) - - # 保存结果 - results = { - 'accuracy': float(accuracy), - 'precision': float(precision), - 'recall': float(recall), - 'f1_score': float(f1), - 'true_positives': int(tp), - 'true_negatives': int(tn), - 'false_positives': int(fp), - 'false_negatives': int(fn), - 'n_samples': int(len(preds)) - } - - import json - with open('./models/checkpoints/dlinear_classification_2026/test_2026_results.json', 'w') as f: - json.dump(results, f, indent=2) - - print(f"\n✓ 结果已保存至: ./models/checkpoints/dlinear_classification_2026/test_2026_results.json") - -if __name__ == "__main__": - main() diff --git a/train_catboost_classification.py b/train_catboost_classification.py new file mode 100644 index 0000000..5913b8c --- /dev/null +++ b/train_catboost_classification.py @@ -0,0 +1,271 @@ +"""训练CatBoost二分类模型用于BTC 1小时涨跌预测""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import numpy as np +import pandas as pd +import pandas_ta as ta +from catboost import CatBoostClassifier, Pool +from sklearn.model_selection import train_test_split +from sklearn.metrics import classification_report, accuracy_score, confusion_matrix +from data.fetcher import BinanceDataFetcher +import pickle +from datetime import datetime + + +def add_technical_features(df): + """添加丰富的技术指标特征""" + df = df.copy() + + # 基础价格特征 + df['price_change'] = df['close'] - df['open'] + df['price_change_pct'] = (df['close'] - df['open']) / df['open'] + df['high_low_range'] = df['high'] - df['low'] + df['high_low_range_pct'] = (df['high'] - df['low']) / df['low'] + + # 移动平均线 + df['sma_7'] = ta.sma(df['close'], length=7) + df['sma_25'] = ta.sma(df['close'], length=25) + df['sma_99'] = ta.sma(df['close'], length=99) + df['ema_12'] = ta.ema(df['close'], length=12) + df['ema_26'] = ta.ema(df['close'], length=26) + + # RSI + df['rsi_14'] = ta.rsi(df['close'], length=14) + df['rsi_7'] = ta.rsi(df['close'], length=7) + + # MACD + macd = ta.macd(df['close']) + if macd is not None and not macd.empty: + df['macd'] = macd.iloc[:, 0] + df['macd_hist'] = macd.iloc[:, 1] + df['macd_signal'] = macd.iloc[:, 2] + + # 布林带 + bbands = ta.bbands(df['close'], length=20) + if bbands is not None and not bbands.empty: + df['bb_upper'] = bbands.iloc[:, 0] + df['bb_mid'] = bbands.iloc[:, 1] + df['bb_lower'] = bbands.iloc[:, 2] + df['bb_width'] = (df['bb_upper'] - df['bb_lower']) / df['bb_mid'] + df['bb_position'] = (df['close'] - df['bb_lower']) / (df['bb_upper'] - df['bb_lower']) + + # ATR (Average True Range) + df['atr'] = ta.atr(df['high'], df['low'], df['close'], length=14) + + # 成交量特征 + df['volume_sma_7'] = ta.sma(df['volume'], length=7) + df['volume_ratio'] = df['volume'] / df['volume_sma_7'] + + # OBV (On-Balance Volume) + df['obv'] = ta.obv(df['close'], df['volume']) + + # Stochastic Oscillator + stoch = ta.stoch(df['high'], df['low'], df['close']) + if stoch is not None and not stoch.empty: + df['stoch_k'] = stoch.iloc[:, 0] + df['stoch_d'] = stoch.iloc[:, 1] + + # ADX (Average Directional Index) + adx = ta.adx(df['high'], df['low'], df['close']) + if adx is not None and not adx.empty: + df['adx'] = adx.iloc[:, 0] + + # 价格动量 + df['momentum_1'] = df['close'].pct_change(1) + df['momentum_3'] = df['close'].pct_change(3) + df['momentum_7'] = df['close'].pct_change(7) + + return df + + +def create_labels(df, forecast_horizon=1): + """创建二分类标签""" + df = df.copy() + df['future_price'] = df['close'].shift(-forecast_horizon) + df['future_return'] = (df['future_price'] - df['close']) / df['close'] + df['label'] = (df['future_return'] > 0).astype(int) # 1=涨, 0=跌 + return df[:-forecast_horizon] + + +def prepare_data(df): + """准备训练数据""" + # 添加技术指标 + df = add_technical_features(df) + + # 创建标签 + df = create_labels(df, forecast_horizon=1) + + # 删除NaN + df = df.dropna().reset_index(drop=True) + + # 选择特征列（排除原始时间戳和标签列） + exclude_cols = ['timestamp', 'label', 'future_price', 'future_return'] + feature_cols = [col for col in df.columns if col not in exclude_cols] + + X = df[feature_cols] + y = df['label'] + + return X, y, feature_cols + + +def main(): + print("="*80) + print("训练CatBoost二分类模型 - BTC 1小时涨跌预测") + print("="*80) + + # ========== 步骤1: 获取数据 ========== + print("\n[1/5] 获取历史数据...") + + fetcher = BinanceDataFetcher() + + # 获取训练数据 (2020-2024) + df_train = fetcher.fetch_ohlcv('BTC', '1h', start_date='2020-01-01', + end_date='2024-12-31', use_cache=True) + + # 获取验证数据 (2025) + df_val = fetcher.fetch_ohlcv('BTC', '1h', start_date='2025-01-01', + end_date='2025-12-31', use_cache=True) + + print(f"✓ 训练数据: {len(df_train)} 行 ({df_train['timestamp'].min()} 至 {df_train['timestamp'].max()})") + print(f"✓ 验证数据: {len(df_val)} 行 ({df_val['timestamp'].min()} 至 {df_val['timestamp'].max()})") + + # ========== 步骤2: 特征工程 ========== + print("\n[2/5] 特征工程...") + + X_train, y_train, feature_cols = prepare_data(df_train) + X_val, y_val, _ = prepare_data(df_val) + + print(f"✓ 特征数量: {len(feature_cols)}") + print(f"✓ 训练样本: {len(X_train)} (涨: {(y_train==1).sum()}, 跌: {(y_train==0).sum()})") + print(f"✓ 验证样本: {len(X_val)} (涨: {(y_val==1).sum()}, 跌: {(y_val==0).sum()})") + + # 显示部分特征 + print(f"\n特征列表 (前20个):") + for i, col in enumerate(feature_cols[:20], 1): + print(f" {i:2d}. {col}") + if len(feature_cols) > 20: + print(f" ... 还有 {len(feature_cols)-20} 个特征") + + # ========== 步骤3: 训练CatBoost模型 ========== + print("\n[3/5] 训练CatBoost模型...") + + # 创建CatBoost数据池 + train_pool = Pool(X_train, y_train) + val_pool = Pool(X_val, y_val) + + # 配置CatBoost参数 + model = CatBoostClassifier( + iterations=1000, + learning_rate=0.05, + depth=6, + l2_leaf_reg=3, + loss_function='Logloss', + eval_metric='Accuracy', + random_seed=42, + verbose=100, + early_stopping_rounds=50, + task_type='CPU', # 如果有GPU可改为'GPU' + ) + + # 训练模型 + model.fit( + train_pool, + eval_set=val_pool, + use_best_model=True, + plot=False + ) + + print(f"\n✓ 训练完成") + print(f" 最佳迭代轮次: {model.get_best_iteration()}") + print(f" 最佳验证准确率: {model.get_best_score()['validation']['Accuracy']:.4f}") + + # ========== 步骤4: 评估模型 ========== + print("\n[4/5] 评估模型性能...") + + # 训练集预测 + y_train_pred = model.predict(X_train) + train_acc = accuracy_score(y_train, y_train_pred) + + # 验证集预测 + y_val_pred = model.predict(X_val) + val_acc = accuracy_score(y_val, y_val_pred) + + print(f"\n训练集性能：") + print(f" 准确率: {train_acc*100:.2f}%") + + print(f"\n验证集性能：") + print(f" 准确率: {val_acc*100:.2f}%") + + # 混淆矩阵 + cm = confusion_matrix(y_val, y_val_pred) + print(f"\n混淆矩阵 (验证集):") + print(f" 预测↓ 预测↑") + print(f"真实↓ {cm[0,0]:4d} {cm[0,1]:4d}") + print(f"真实↑ {cm[1,0]:4d} {cm[1,1]:4d}") + + # 分类报告 + print(f"\n分类报告 (验证集):") + print(classification_report(y_val, y_val_pred, target_names=['下跌', '上涨'])) + + # 特征重要性 (Top 20) + feature_importance = model.get_feature_importance() + importance_df = pd.DataFrame({ + 'feature': feature_cols, + 'importance': feature_importance + }).sort_values('importance', ascending=False) + + print(f"\nTop 20 重要特征:") + for i, (idx, row) in enumerate(importance_df.head(20).iterrows(), 1): + print(f" {i:2d}. {row['feature']:<20s} {row['importance']:7.2f}") + + # ========== 步骤5: 保存模型 ========== + print("\n[5/5] 保存模型...") + + checkpoint_dir = './models/checkpoints/catboost_classification_2026' + os.makedirs(checkpoint_dir, exist_ok=True) + + # 保存CatBoost模型 + model_path = f'{checkpoint_dir}/catboost_model.cbm' + model.save_model(model_path, format='cbm') + + # 保存特征列表和元数据 + metadata = { + 'feature_cols': feature_cols, + 'train_acc': float(train_acc), + 'val_acc': float(val_acc), + 'best_iteration': int(model.get_best_iteration()), + 'train_samples': len(X_train), + 'val_samples': len(X_val), + 'timestamp': datetime.now().strftime('%Y-%m-%d %H:%M:%S'), + 'train_date_range': f"{df_train['timestamp'].min()} - {df_train['timestamp'].max()}", + 'val_date_range': f"{df_val['timestamp'].min()} - {df_val['timestamp'].max()}", + } + + with open(f'{checkpoint_dir}/metadata.pkl', 'wb') as f: + pickle.dump(metadata, f) + + print(f"✓ 模型已保存到: {model_path}") + print(f"✓ 元数据已保存到: {checkpoint_dir}/metadata.pkl") + + # ========== 总结 ========== + print("\n" + "="*80) + print("✓ CatBoost二分类模型训练完成!") + print("="*80) + + print(f"\n模型性能总结：") + print(f" 训练集准确率: {train_acc*100:.2f}%") + print(f" 验证集准确率: {val_acc*100:.2f}%") + print(f" 特征数量: {len(feature_cols)}") + print(f" 最佳迭代: {model.get_best_iteration()}") + + print(f"\n保存位置：") + print(f" {checkpoint_dir}/") + + print("\n" + "="*80) + + +if __name__ == "__main__": + main() diff --git a/train_nbeats.py b/train_nbeats.py new file mode 100644 index 0000000..b70e9b5 --- /dev/null +++ b/train_nbeats.py @@ -0,0 +1,283 @@ +"""使用N-BEATS模型训练：预测未来1小时BTC价格""" + +import sys +import os +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) + +import numpy as np +import pandas as pd +import pandas_ta as ta +from sklearn.preprocessing import StandardScaler +import torch +import torch.nn as nn +from torch.utils.data import DataLoader, TensorDataset + +from data.fetcher import BinanceDataFetcher +from models.nbeats.model import NBeats, NBeatsConfig +import json +import pickle + + +class NBeatsPreprocessor: + """N-BEATS预处理器：预测未来价格""" + + def __init__(self, lookback=168, forecast_horizon=1): + self.lookback = lookback + self.forecast_horizon = forecast_horizon + self.price_scaler = StandardScaler() + self.is_fitted = False + + def preprocess(self, df, fit=True): + """预处理数据""" + df = df.copy() + + # 创建标签：未来价格 + df['future_price'] = df['close'].shift(-self.forecast_horizon) + df = df[:-self.forecast_horizon].reset_index(drop=True) + + # 标准化价格 + prices = df['close'].values.reshape(-1, 1) + future_prices = df['future_price'].values.reshape(-1, 1) + + if fit: + scaled_prices = self.price_scaler.fit_transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + self.is_fitted = True + else: + if not self.is_fitted: + raise ValueError("Must fit before transform") + scaled_prices = self.price_scaler.transform(prices) + scaled_future = self.price_scaler.transform(future_prices) + + # 创建序列 + X, y = [], [] + for i in range(len(scaled_prices) - self.lookback): + X.append(scaled_prices[i:i + self.lookback, 0]) + y.append(scaled_future[i + self.lookback, 0]) + + return np.array(X), np.array(y), df + + def inverse_transform_price(self, scaled_price): + """还原真实价格""" + return self.price_scaler.inverse_transform(scaled_price.reshape(-1, 1)).flatten() + + +def train_nbeats(): + print("="*80) + print("N-BEATS模型训练 - 预测未来1小时BTC价格") + print("="*80) + + # ========== 步骤1: 获取数据 ========== + print("\n[1/8] 获取BTC 2020-2025年数据...") + fetcher = BinanceDataFetcher() + df = fetcher.fetch_ohlcv( + symbol='BTC', + timeframe='1h', + start_date='2020-01-01', + end_date='2025-12-31', + use_cache=True + ) + + print(f"✓ 数据量: {len(df):,} 行") + print(f"✓ 时间: {df['timestamp'].min()} 至 {df['timestamp'].max()}") + print(f"✓ 价格: ${df['close'].iloc[0]:,.2f} → ${df['close'].iloc[-1]:,.2f}") + + # ========== 步骤2: 预处理 ========== + print("\n[2/8] 数据预处理...") + preprocessor = NBeatsPreprocessor(lookback=168, forecast_horizon=1) + X, y, processed_df = preprocessor.preprocess(df, fit=True) + + print(f"✓ 序列数量: {len(X):,}") + print(f"✓ 输入形状: {X.shape}") + print(f"✓ 输出形状: {y.shape}") + + # 还原真实价格看统计 + y_real = preprocessor.inverse_transform_price(y) + print(f"\n✓ 目标价格统计:") + print(f" 最小值: ${y_real.min():,.2f}") + print(f" 最大值: ${y_real.max():,.2f}") + print(f" 均值: ${y_real.mean():,.2f}") + + # ========== 步骤3: 划分数据集 ========== + print("\n[3/8] 划分数据集（90%训练 + 10%验证）...") + n = len(X) + train_end = int(n * 0.9) + + X_train, y_train = X[:train_end], y[:train_end] + X_val, y_val = X[train_end:], y[train_end:] + + print(f" 训练集: {len(X_train):,} 样本") + print(f" 验证集: {len(X_val):,} 样本") + + # ========== 步骤4: 创建数据加载器 ========== + print("\n[4/8] 创建数据加载器...") + train_dataset = TensorDataset( + torch.FloatTensor(X_train), + torch.FloatTensor(y_train) + ) + val_dataset = TensorDataset( + torch.FloatTensor(X_val), + torch.FloatTensor(y_val) + ) + + train_loader = DataLoader(train_dataset, batch_size=128, shuffle=True) + val_loader = DataLoader(val_dataset, batch_size=128, shuffle=False) + + # ========== 步骤5: 创建N-BEATS模型 ========== + print("\n[5/8] 创建N-BEATS模型...") + config = NBeatsConfig( + input_size=168, + forecast_size=1, + stacks=['trend', 'seasonality', 'generic'], + num_blocks_per_stack=3, + block_layers=4, + layer_size=256, + trend_degree=2, + seasonality_harmonics=4 + ) + model = NBeats(config) + + device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu') + model = model.to(device) + + print(f"✓ 模型参数: {model.count_parameters():,}") + print(f"✓ 设备: {device}") + print(f"✓ 模型架构: Trend + Seasonality + Generic") + + # ========== 步骤6: 训练 ========== + print("\n[6/8] 开始训练...") + + criterion = nn.MSELoss() + optimizer = torch.optim.Adam(model.parameters(), lr=0.001) + scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau( + optimizer, mode='min', factor=0.5, patience=5, verbose=True + ) + + best_val_loss = float('inf') + patience = 15 + patience_counter = 0 + max_epochs = 50 + + for epoch in range(max_epochs): + # 训练 + model.train() + train_loss = 0 + for batch_X, batch_y in train_loader: + batch_X, batch_y = batch_X.to(device), batch_y.to(device) + + optimizer.zero_grad() + outputs = model(batch_X) + loss = criterion(outputs, batch_y) + loss.backward() + optimizer.step() + + train_loss += loss.item() + + train_loss /= len(train_loader) + + # 验证 + model.eval() + val_loss = 0 + with torch.no_grad(): + for batch_X, batch_y in val_loader: + batch_X, batch_y = batch_X.to(device), batch_y.to(device) + outputs = model(batch_X) + loss = criterion(outputs, batch_y) + val_loss += loss.item() + + val_loss /= len(val_loader) + + # 学习率调度 + scheduler.step(val_loss) + + print(f"Epoch {epoch+1}/{max_epochs}") + print(f" Train loss: {train_loss:.6f}") + print(f" Val loss: {val_loss:.6f}") + print(f" LR: {optimizer.param_groups[0]['lr']:.6f}") + + # 保存最佳模型 + if val_loss < best_val_loss: + best_val_loss = val_loss + patience_counter = 0 + os.makedirs('./models/checkpoints/nbeats_btc_1h', exist_ok=True) + model.save('./models/checkpoints/nbeats_btc_1h/best_model.pt') + print(f" Saved best model (val_loss: {val_loss:.6f})") + else: + patience_counter += 1 + + # Early stopping + if patience_counter >= patience: + print(f"\nEarly stopping at epoch {epoch+1}") + break + + # ========== 步骤7: 评估 ========== + print("\n[7/8] 评估模型...") + + # 重新加载最佳模型 + checkpoint = torch.load('./models/checkpoints/nbeats_btc_1h/best_model.pt', map_location=device) + model = NBeats(checkpoint['config']) + model.load_state_dict(checkpoint['model_state_dict']) + model = model.to(device) + model.eval() + + val_preds = [] + with torch.no_grad(): + for batch_X, _ in val_loader: + batch_X = batch_X.to(device) + pred = model(batch_X) + val_preds.extend(pred.cpu().numpy()) + + val_preds = np.array(val_preds) + + # 还原真实价格 + pred_prices = preprocessor.inverse_transform_price(val_preds) + true_prices = preprocessor.inverse_transform_price(y_val) + + # 当前价格 + val_start_idx = train_end + 168 + current_prices = processed_df['close'].iloc[val_start_idx:val_start_idx+len(pred_prices)].values + + # 方向准确率 + pred_direction = (pred_prices > current_prices).astype(int) + true_direction = (true_prices > current_prices).astype(int) + direction_acc = (pred_direction == true_direction).mean() + + # 价格预测误差 + price_mae = np.abs(pred_prices - true_prices).mean() + price_rmse = np.sqrt(((pred_prices - true_prices) ** 2).mean()) + price_mape = np.mean(np.abs((true_prices - pred_prices) / true_prices)) * 100 + + print("\n" + "="*80) + print("训练完成 - N-BEATS模型性能") + print("="*80) + + print(f"\n📊 价格预测误差:") + print(f" MAE: ${price_mae:,.2f}") + print(f" RMSE: ${price_rmse:,.2f}") + print(f" MAPE: {price_mape:.2f}%") + + print(f"\n📈 方向预测:") + print(f" 方向准确率: {direction_acc*100:.2f}%") + + # ========== 步骤8: 保存 ========== + print("\n[8/8] 保存模型和预处理器...") + + with open('./models/checkpoints/nbeats_btc_1h/preprocessor.pkl', 'wb') as f: + pickle.dump(preprocessor, f) + + metrics = { + 'price_mae': float(price_mae), + 'price_rmse': float(price_rmse), + 'price_mape': float(price_mape), + 'direction_accuracy': float(direction_acc), + 'best_val_loss': float(best_val_loss) + } + with open('./models/checkpoints/nbeats_btc_1h/metrics.json', 'w') as f: + json.dump(metrics, f, indent=2) + + print(f"\n✓ 模型保存至: ./models/checkpoints/nbeats_btc_1h/") + print("="*80) + + +if __name__ == "__main__": + train_nbeats()