ai

freqtrade/templates/freqaiprimer.py

@@ -4,6 +4,7 @@ warnings.filterwarnings("ignore", category=UserWarning, module="pandas_ta")
 import logging
 from freqtrade.strategy import IStrategy
 from pandas import DataFrame
+import pandas as pd
 import pandas_ta as ta
 from freqtrade.persistence import Trade
 import numpy as np
@@ -198,10 +199,24 @@ class FreqaiPrimer(IStrategy):
             # 确保波动率列存在
             if "%-volatility" not in dataframe.columns:
                 logger.warning("缺少 %-volatility 列，强制重新生成")
-                dataframe["%-volatility"] = dataframe["close"].pct_change().rolling(20, min_periods=1).std()
-                dataframe["%-volatility"] = dataframe["%-volatility"].replace([np.inf, -np.inf], 0)
-                dataframe["%-volatility"] = dataframe["%-volatility"].ffill()
-                dataframe["%-volatility"] = dataframe["%-volatility"].fillna(0)
+                # 计算健壮的波动率指标，避免MinMaxScaler 0样本错误
+                def calculate_robust_volatility(df, window=20):
+                    # 计算变化率
+                    returns = df['close'].pct_change()
+                    # 添加小的epsilon避免全零
+                    returns = returns + 1e-10
+                    # 使用较高的最小周期要求以确保足够的数据
+                    volatility = returns.rolling(window, min_periods=5).std()
+                    # 确保有足够的有效数据
+                    if volatility.isna().sum() > len(volatility) * 0.8:
+                        # 如果大多数数据无效，使用替代计算方法
+                        return pd.Series([0.001]*len(df), index=df.index)
+                    # 再填充和异常值处理
+                    volatility = volatility.replace([np.inf, -np.inf], np.nan)
+                    volatility = volatility.ffill().bfill().fillna(0.001)  # 使用0.001而不是0作为默认值
+                    return volatility
+                
+                dataframe["%-volatility"] = calculate_robust_volatility(dataframe)
             
             # 生成买入RSI阈值目标
             dataframe["&-buy_rsi"] = ta.rsi(dataframe['close'], length=14)
@@ -341,13 +356,31 @@ class FreqaiPrimer(IStrategy):
                 dataframe[col] = dataframe[col].ffill().bfill().fillna(0)
         
         # 启动FreqAI处理
-        try:
-            logger.info(f"[{metadata['pair']}] 开始FreqAI处理，数据量: {len(dataframe)}行")
-            # 创建FreqAI处理的数据副本，确保原始数据不被修改
-            freqai_data = dataframe.copy()
-            
-            # 增强FreqAI处理前的数据完整性检查，避免MinMaxScaler 0样本错误
-            valid_rows = freqai_data.dropna(subset=['close'])
+            try:
+                logger.info(f"[{metadata['pair']}] 开始FreqAI处理，数据量: {len(dataframe)}行")
+                # 创建FreqAI处理的数据副本，确保原始数据不被修改
+                freqai_data = dataframe.copy()
+                
+                # 增强FreqAI处理前的数据完整性检查，避免MinMaxScaler 0样本错误
+                valid_rows = freqai_data.dropna(subset=['close'])
+                
+                # 在FreqAI处理前添加目标变量验证，确保所有变量都能被正确标准化
+                target_cols = ['&-buy_rsi', '&-sell_rsi', '&-roi_0', '&-stoploss', '%-volatility']
+                for target_col in target_cols:
+                    if target_col in freqai_data.columns:
+                        # 检查唯一值数量和分布
+                        unique_vals = freqai_data[target_col].nunique()
+                        if unique_vals < 3:
+                            logger.warning(f"[{metadata['pair']}] {target_col} 唯一值太少({unique_vals}), 可能导致标准化问题")
+                            # 添加微小噪声以增加唯一性
+                            freqai_data[target_col] += np.random.normal(0, 1e-7, size=len(freqai_data))
+                            # 确保噪声添加后值仍然合理
+                            if target_col.startswith('&-'):
+                                if target_col == '&-buy_rsi' or target_col == '&-sell_rsi':
+                                    freqai_data[target_col] = freqai_data[target_col].clip(0, 100)
+                                else:
+                                    # 限制ROI和止损的范围
+                                    freqai_data[target_col] = freqai_data[target_col].clip(-0.2, 0.2)
             if len(valid_rows) < 30:
                 logger.warning(f"[{metadata['pair']}] 有效收盘价数据不足({len(valid_rows)}), 跳过FreqAI处理")
                 dataframe = original_dataframe.copy()