phyer/myTestFreqAI
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

训练时数据长度不匹配问题

Browse Source
This commit is contained in:
zhangkun9038@dingtalk.com 2025-08-18 01:56:24 +08:00
parent c5b35e9c09
commit dee9cdb16a
1 changed files with 69 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

85
freqtrade/templates/freqaiprimer.py
View File

@ -411,7 +411,7 @@ class FreqaiPrimer(IStrategy):
logger.warning(f"[{pair}] 1h 数据为空或不足（{len(dataframe_1h)} 根K线），初始化空列")
for col in ['open_1h', 'high_1h', 'low_1h', 'close_1h', 'stochrsi_k_1h', 'stochrsi_d_1h',
'bb_upper_1h', 'bb_middle_1h', 'bb_lower_1h', 'volume_z_score_1h']:
dataframe[col] = np.nan
dataframe[col] = 0.0 # 使用0.0代替np.nan避免NaN值
else:
# 计算 1h 指标
if len(dataframe_1h) >= 20: # 确保有足够数据计算 rolling(20)
@ -430,7 +430,7 @@ class FreqaiPrimer(IStrategy):
dataframe_1h['volume_z_score_1h'] = dataframe_1h['volume_z_score_1h'].replace([np.inf, -np.inf], 0).ffill().fillna(0)
else:
logger.warning(f"[{pair}] 1h 数据不足以计算 volume_z_score_1h（{len(dataframe_1h)} 根K线，需至少20根）")
dataframe_1h['volume_z_score_1h'] = np.nan
dataframe_1h['volume_z_score_1h'] = 0.0
# 映射 1h 数据到主时间框架
for col in ['open', 'high', 'low', 'close', 'stochrsi_k', 'stochrsi_d', 'bb_upper_1h', 'bb_middle_1h', 'bb_lower_1h', 'volume_z_score_1h']:
@ -438,15 +438,30 @@ class FreqaiPrimer(IStrategy):
dataframe[col if col.endswith('_1h') else f"{col}_1h"] = dataframe_1h[col].reindex(dataframe.index, method='ffill').bfill()
else:
logger.warning(f"[{pair}] 1h 数据缺少列 {col}，初始化为空")
dataframe[col if col.endswith('_1h') else f"{col}_1h"] = np.nan
dataframe[col if col.endswith('_1h') else f"{col}_1h"] = 0.0
# 数据清理：处理 NaN 和无穷值
for col in ["ema200", "bb_upperband", "bb_middleband", "bb_lowerband", "rsi", "volume_z_score",
"&-price_value_divergence", "price_value_divergence", "open_1h", "high_1h", "low_1h",
"close_1h", "stochrsi_k_1h", "stochrsi_d_1h", "bb_upper_1h", "bb_middle_1h", "bb_lower_1h",
"volume_z_score_1h"]:
# 数据清理：处理 NaN 和无穷值，确保数据完整性
critical_columns = ["ema200", "bb_upperband", "bb_middleband", "bb_lowerband", "rsi", "volume_z_score",
"&-price_value_divergence", "price_value_divergence", "open_1h", "high_1h", "low_1h",
"close_1h", "stochrsi_k_1h", "stochrsi_d_1h", "bb_upper_1h", "bb_middle_1h", "bb_lower_1h",
"volume_z_score_1h", "stochrsi_k", "stochrsi_d", "close", "volume"]
for col in critical_columns:
if col in dataframe.columns:
dataframe[col] = dataframe[col].replace([np.inf, -np.inf], 0).ffill().fillna(0)
# 处理无穷值和NaN值
dataframe[col] = dataframe[col].replace([np.inf, -np.inf], 0)
# 前向填充，然后用0填充剩余NaN
dataframe[col] = dataframe[col].ffill().fillna(0)
# 确保数据类型正确
dataframe[col] = pd.to_numeric(dataframe[col], errors='coerce').fillna(0)
# 验证数据长度一致性
expected_length = len(dataframe)
for col in critical_columns:
if col in dataframe.columns and len(dataframe[col]) != expected_length:
logger.warning(f"[{pair}] 列 {col} 长度不匹配: {len(dataframe[col])} != {expected_length}")
# 重新索引确保长度一致
dataframe[col] = dataframe[col].reindex(dataframe.index, method='ffill').fillna(0)
# 调用 FreqAI 预测 - 使用单一回归模型
if not hasattr(self, 'freqai') or self.freqai is None:
@ -723,7 +738,20 @@ class FreqaiPrimer(IStrategy):
f"cond5: {cond5.mean():.2%}, "
f"buy_condition: {buy_condition.mean():.2%}")
# 记录 enter_long 信号统计
logger.info(f"[{pair}] enter_long 信号总数：{dataframe['enter_long'].sum() if 'enter_long' in dataframe.columns else 0}")
enter_long_count = dataframe['enter_long'].sum() if 'enter_long' in dataframe.columns else 0
logger.info(f"[{pair}] enter_long 信号总数：{enter_long_count}")
# 数据完整性检查
if len(dataframe) > 0:
nan_columns = [col for col in dataframe.columns if dataframe[col].isna().any()]
if nan_columns:
logger.warning(f"[{pair}] 发现NaN值的列: {nan_columns}")
for col in nan_columns:
nan_count = dataframe[col].isna().sum()
logger.warning(f"[{pair}] 列 {col} 有 {nan_count} 个NaN值")
# 验证数据长度
logger.info(f"[{pair}] 最终数据行数: {len(dataframe)}, 列数: {len(dataframe.columns)}")
return dataframe
@ -738,6 +766,11 @@ class FreqaiPrimer(IStrategy):
dataframe["adx"] = ta.ADX(dataframe, timeperiod=14)
# 计算短期价格涨幅（最近 5 根 K 线，约 15 分钟）
dataframe["short_term_return"] = dataframe["close"].pct_change(5, fill_method=None) * 100 # 百分比回报
# 清理新计算列的NaN值
for col in ["stochrsi_k", "adx", "short_term_return"]:
if col in dataframe.columns:
dataframe[col] = dataframe[col].replace([np.inf, -np.inf], 0).ffill().fillna(0)
# 获取市场趋势得分
trend_score = self.get_market_trend(dataframe=dataframe, metadata={'pair': pair})
@ -808,12 +841,12 @@ class FreqaiPrimer(IStrategy):
conditions.append(sell_condition)
# 调试日志
divergence_value = dataframe["&-price_value_divergence"].iloc[-1] if not dataframe["&-price_value_divergence"].isna().all() else np.nan
rsi_value = dataframe["rsi"].iloc[-1] if not dataframe["rsi"].isna().all() else np.nan
stochrsi_value = dataframe["stochrsi_k"].iloc[-1] if not dataframe["stochrsi_k"].isna().all() else np.nan
adx_value = dataframe["adx"].iloc[-1] if not dataframe["adx"].isna().all() else np.nan
short_term_return = dataframe["short_term_return"].iloc[-1] if not dataframe["short_term_return"].isna().all() else np.nan
# 调试日志 - 使用安全的值获取
divergence_value = dataframe["&-price_value_divergence"].iloc[-1] if len(dataframe) > 0 and not dataframe["&-price_value_divergence"].isna().iloc[-1] else 0.0
rsi_value = dataframe["rsi"].iloc[-1] if len(dataframe) > 0 and not dataframe["rsi"].isna().iloc[-1] else 0.0
stochrsi_value = dataframe["stochrsi_k"].iloc[-1] if len(dataframe) > 0 and not dataframe["stochrsi_k"].isna().iloc[-1] else 0.0
adx_value = dataframe["adx"].iloc[-1] if len(dataframe) > 0 and not dataframe["adx"].isna().iloc[-1] else 0.0
short_term_return = dataframe["short_term_return"].iloc[-1] if len(dataframe) > 0 and not dataframe["short_term_return"].isna().iloc[-1] else 0.0
logger.info(f"[{pair}] 卖出条件检查 - "
f"&-price_value_divergence={divergence_value:.6f} > {self.sell_threshold * float(self.exit_divergence_multiplier.value):.6f}: {cond1.iloc[-1]}, "
f"rsi={rsi_value:.2f} > {float(self.exit_rsi_threshold.value)} & stochrsi_k={stochrsi_value:.2f} > {float(self.exit_stochrsi_threshold.value)}: {cond2.iloc[-1]}, "
@ -828,6 +861,26 @@ class FreqaiPrimer(IStrategy):
else:
logger.info(f"[{pair}] 无有效卖出条件")
# 数据完整性检查
if len(dataframe) > 0:
nan_columns = [col for col in dataframe.columns if dataframe[col].isna().any()]
if nan_columns:
logger.warning(f"[{pair}] 卖出检查 - 发现NaN值的列: {nan_columns}")
for col in nan_columns:
nan_count = dataframe[col].isna().sum()
if nan_count > 0:
logger.warning(f"[{pair}] 卖出检查 - 列 {col} 有 {nan_count} 个NaN值，已清理")
dataframe[col] = dataframe[col].fillna(0)
# 验证数据长度一致性
expected_length = len(dataframe)
actual_length = len(dataframe.dropna())
logger.info(f"[{pair}] 卖出检查 - 数据行数: {expected_length}, 有效行数: {actual_length}")
# 记录exit_long信号
exit_long_count = dataframe['exit_long'].sum() if 'exit_long' in dataframe.columns else 0
logger.info(f"[{pair}] exit_long 信号总数：{exit_long_count}")
return dataframe
def buy_space(self):