尧图网页设计 
Vibe-Trading ML 策略技能实战基于 scikit-learn 的 Walk-Forward 机器学习预测策略【免费下载链接】Vibe-TradingVibe-Trading: Your Personal Trading Agent项目地址: https://gitcode.com/GitHub_Trending/vi/Vibe-TradingVibe-Trading 将机器学习策略封装为ml-strategy技能SKILL.md为 Agent 提供一套基于 scikit-learn 的可直接复制运行的完整策略管线从 OHLCV 数据校验、多因子特征工程到未来 N 日收益方向标签构建、walk-forward 滚动训练与信号生成。读完本文你将掌握这套防数据泄漏data leakage、带输出契约no NaN / 数值裁剪的SignalEngine的每一行实现理解特征、模型、参数与信号约定的设计意图并能在任何 OHLCV 数据集上落地自己的机器学习预测策略。技能定位Vibe-Trading 策略类技能之一Vibe-Trading 的内置技能库按类别组织ml-strategy属于Strategy策略类别与strategy-generate、cross-market-strategy、technical-basic、candlestick、ichimoku、elliott-wave、smc、multi-factor等同列见 README_zh.md。该类别共 19 个技能ml-strategy的职责非常聚焦用机器学习模型预测未来收益方向并生成交易信号且明确声明适用于任何 OHLCV 数据Suitable for any OHLCV data。技能文件采用统一的 frontmatter 元数据格式name/description/category由技能加载器读取。从 技能加载器实现 可以看到Skill数据类解析 SKILL.md 的 frontmatter 得到名称、描述与类别正文则通过load_skill工具按需注入 Agent 上下文。这种渐进式披露设计意味着ml-strategy的完整管线代码平时不会全部塞进系统提示而是在 Agent 需要时被精确加载。ml-strategy还深度接入 Vibe-Trading 的 Swarm 多智能体编排ml_quant_lab预设ml_quant_lab.yaml中Feature Engineer 与 Data Scientist 两个角色的skills字段都声明了ml-strategy并通过load_skill(ml-strategy)获取特征工程最佳实践与金融 ML 设计标准随后由 Backtest Engineer 对产物做严格样本外OOS验证。这说明该技能不仅面向单个 Agent也是多智能体量化实验室的标准参考手册。信号逻辑五步流水线ml-strategy的信号生成遵循严格有序的五步流程每一步都在为避免未来函数泄漏、保证输出可被下游直接消费服务校验输入Validate input检查 OHLCV 列是否齐全、最小行数是否达标、NaN 占比是否过高——不合格的标的直接跳过绝不进入训练流程特征工程Feature engineering从原始 OHLCV 构建动量、波动率、RSI、均线比、量比等多维因子所有特征统一做消毒处理inf替换、除零防护标签构建Label construction未来 N 日收益 0 记为正类1 0 记为负类0Walk-forward 训练采用扩展窗口expanding或滑动窗口sliding只用历史数据训练逐日向前滚动预测信号生成将predict_proba[:, 1]映射到[-1.0, 1.0]或使用predict得到{-1, 0, 1}离散信号输出保证干净无 NaN、数值已裁剪。其中第 3 步与第 4 步的组合是整套管线防泄漏的关键标签基于未来收益构造而训练样本只允许取标签在预测时刻已可观测的历史切片——这一点在测试中被专门验证见下文防泄漏的正确性验证。完整 SignalEngine 示例推荐的全管线实现这是该技能推荐的标准全流程实现复制即可运行安全性已内建。它是后续特征工程、模型选型、参数调优讨论的基准代码。import numpy as np import pandas as pd from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier from sklearn.linear_model import LogisticRegression from sklearn.preprocessing import StandardScaler def validate_data(df: pd.DataFrame, min_rows: int 300) - bool: Check that OHLCV data meets minimum quality for ML training. Args: df: DataFrame with DatetimeIndex. min_rows: Minimum number of rows required. Returns: True if data is usable. required {open, high, low, close, volume} if not required.issubset(df.columns): return False if len(df) min_rows: return False if df[close].isnull().mean() 0.2: return False return True def build_features(df: pd.DataFrame) - pd.DataFrame: Build a machine-learning feature matrix from OHLCV data. All features are guarded against division-by-zero and sanitized (inf replaced with NaN) so downstream code never sees inf values. Args: df: DataFrame containing open, high, low, close, and volume columns. Returns: DataFrame with feature columns prefixed by f_. c df[close] v df[volume] ret c.pct_change(fill_methodNone) features pd.DataFrame(indexdf.index) features[f_ret_5d] c.pct_change(5, fill_methodNone) features[f_ret_20d] c.pct_change(20, fill_methodNone) features[f_vol_20d] ret.rolling(20).std() features[f_ma_ratio] c / c.rolling(20).mean() features[f_volume_ratio] v / v.rolling(20).mean() # RSI(14) — guard: loss0 in zero-volatility periods produces inf delta c.diff() gain delta.clip(lower0).rolling(14).mean() loss (-delta.clip(upper0)).rolling(14).mean() rs gain / loss.replace(0, np.nan) features[f_rsi_14] 100 - (100 / (1 rs)) # Bollinger Band position — guard: bb_upper bb_lower when std0 ma20 c.rolling(20).mean() std20 c.rolling(20).std() bb_upper ma20 2 * std20 bb_lower ma20 - 2 * std20 bb_range (bb_upper - bb_lower).replace(0, np.nan) features[f_bb_position] (c - bb_lower) / bb_range # Intraday features features[f_high_low_ratio] (df[high] - df[low]) / c features[f_close_open_ratio] (c - df[open]) / df[open] features[f_skew_20d] ret.rolling(20).skew() # Sanitize: replace all inf with NaN (NaN handled by walk-forward) features features.replace([np.inf, -np.inf], np.nan) return features def walk_forward_predict( features: pd.DataFrame, labels: pd.Series, min_train_size: int 252, retrain_freq: int 20, model_type: str random_forest, window_type: str expanding, sliding_size: int 504, prediction_horizon: int 5, ) - pd.Series: Walk-forward training and prediction to avoid future data leakage. Args: features: Feature matrix aligned with labels by row index. labels: Binary labels (0/1), representing the direction of future N-day returns. min_train_size: Minimum training-set size in trading days. retrain_freq: Retrain the model every N days. model_type: One of random_forest / gradient_boosting / ridge. window_type: expanding uses all history; sliding uses a fixed lookback. sliding_size: Lookback window size when window_type is sliding. prediction_horizon: Number of bars each target label looks ahead. Returns: Predicted signal series with range [-1.0, 1.0], no NaN values. predictions pd.Series(0.0, indexfeatures.index) model None scaler None if prediction_horizon 1: raise ValueError(prediction_horizon must be 1) for i in range(min_train_size, len(features)): # Retrain every retrain_freq days if model is None or (i - min_train_size) % retrain_freq 0: # A label at row t is observable only once t horizon i. train_stop max(0, i - prediction_horizon 1) start ( max(0, train_stop - sliding_size) if window_type sliding else 0 ) X_train features.iloc[start:train_stop].values y_train labels.iloc[start:train_stop].values # Drop rows with NaN valid ~(np.isnan(X_train).any(axis1) | np.isnan(y_train)) X_train X_train[valid] y_train y_train[valid] if len(X_train) 50: continue # Standardization: fit only on training set scaler StandardScaler() X_train scaler.fit_transform(X_train) # Build the model if model_type random_forest: model RandomForestClassifier( n_estimators100, max_depth5, random_state42, ) elif model_type gradient_boosting: model GradientBoostingClassifier( n_estimators100, max_depth3, learning_rate0.05, random_state42, ) elif model_type ridge: model LogisticRegression(penaltyl2, C1.0, random_state42) else: raise ValueError(fUnsupported model_type: {model_type}) model.fit(X_train, y_train) # Predict today X_today features.iloc[i : i 1].values if np.isnan(X_today).any(): predictions.iloc[i] 0.0 continue X_today scaler.transform(X_today) if hasattr(model, predict_proba): prob model.predict_proba(X_today)[0, 1] predictions.iloc[i] prob * 2 - 1 # [0,1] - [-1,1] else: predictions.iloc[i] float(model.predict(X_today)[0]) # Output contract: no NaN, clipped to [-1, 1] predictions predictions.fillna(0.0).clip(-1.0, 1.0) return predictions class SignalEngine: Complete ML strategy with built-in data validation and safety. def generate(self, data_map: dict) - dict: Generate signals for each symbol. Args: data_map: code - OHLCV DataFrame. Returns: code - signal Series in [-1.0, 1.0]. signals {} for code, df in data_map.items(): if not validate_data(df): print(f[WARN] {code}: data quality insufficient, skipping) continue features build_features(df) prediction_horizon 5 future_returns ( df[close].shift(-prediction_horizon) / df[close] - 1 ) labels (future_returns 0).astype(float).where(future_returns.notna()) signal walk_forward_predict( features, labels, prediction_horizonprediction_horizon, ) signals[code] signal return signals各函数职责拆解validate_data(df, min_rows300)数据质量的守门员。要求列集合必须包含open/high/low/close/volume全部五列样本量低于min_rows默认 300 个交易日直接拒绝收盘价缺失率超过 20% 同样拒绝。这一层保证进入训练的数据具备最低统计意义避免用几根 K 线训练出的模型自欺欺人。build_features(df)特征工厂。产出 10 个以f_前缀命名的默认因子详见下节特征表并在返回前用replace([np.inf, -np.inf], np.nan)统一消毒——文档注释明确指出这一设计保证下游代码永远看不到 inf 值而 NaN 交由 walk-forward 的样本过滤处理。walk_forward_predict(...)防泄漏训练与预测的核心。逐日滚动每retrain_freq天重训一次每次重训时用train_stop max(0, i - prediction_horizon 1)把训练集截止点前移一个预测视界确保被训练的标签在预测时刻 i确实已经可观测标准化器StandardScaler只在训练集上fit杜绝标准化泄漏单日特征含 NaN 时该日输出 0.0 中性信号最终统一fillna(0.0).clip(-1.0, 1.0)。SignalEngine.generate(data_map)面向多标的的入口。接收code - OHLCV DataFrame的字典逐标的执行校验→特征→标签→walk-forward 预测返回code - signal Series。标签构建使用了shift(-prediction_horizon)的前移技巧where(future_returns.notna())把视界尚未走完的尾部标签保持为 NaN——这些行不会进入训练。防泄漏的正确性验证测试如何锁定行为技能文档不是空谈Vibe-Trading 用真实测试固定了 walk-forward 的防泄漏语义。仓库中的 test_ml_strategy_skill.py 通过 AST 解析直接从 SKILL.md 提取 Python 代码块执行这意味着文档代码块本身就是被测试的产物。三个测试分别验证test_walk_forward_purges_labels_not_observable_at_prediction_time构造 70 行数据、min_train_size60、retrain_freq100只重训一次、prediction_horizon5断言第一次fit收到的训练集最后一个样本行号是 55——即60 - 5 1 56个样本中的最后一行索引 55证明标签在预测时刻不可观测的尾部样本被精确剔除test_one_bar_horizon_preserves_existing_training_window同样的数据在prediction_horizon1时训练集末行是 59即完整 60 行都保留——视界为 1 时无需剔除语义自洽test_signal_engine_preserves_unavailable_future_labels_as_nan用 10 行 close 数据驱动SignalEngine.generate断言前 5 个标签为1.0、后 5 个标签为NaN同时确认prediction_horizon被正确传为 5——锁定未来标签不可用时保持 NaN的行为。这组测试是理解该技能正确性契约的最佳入口防泄漏不是注释里的愿望而是被 CI 固定下来的行为。特征工程参考默认因子表build_features()是自定义扩展点下表列出全部默认特征可按需增删。Feature NameFormulaMeaningret_5dclose.pct_change(5, fill_methodNone)Past 5-day return (short-term momentum)ret_20dclose.pct_change(20, fill_methodNone)Past 20-day return (medium-term momentum)vol_20dreturns.rolling(20).std()20-day volatilityrsi_14See RSI formula in codeRelative Strength Index (division-by-zero guarded)ma_ratioclose / close.rolling(20).mean()Degree of deviation from the 20-day moving averagevolume_ratiovolume / volume.rolling(20).mean()Volume ratio (current volume vs 20-day average)bb_position(close - bb_lower) / (bb_upper - bb_lower)Bollinger Band position (zero-bandwidth guarded)high_low_ratio(high - low) / closeIntraday range ratioclose_open_ratio(close - open) / openIntraday returnskew_20dreturns.rolling(20).skew()Return skewness这 10 个因子覆盖了四类信息维度动量ret_5d、ret_20d、波动vol_20d、bb_position、skew_20d、量能volume_ratio与日内结构high_low_ratio、close_open_ratio。其中rsi_14与bb_position是教科书式的易除零陷阱零波动时段loss0会让 RSI 产生 inf、std0会让布林带宽为 0代码分别用.replace(0, np.nan)规避——这正体现了该技能安全性内建的工程取向。值得注意的是pct_change与rolling均未指定fill_method之外的参数而是显式传入fill_methodNone避免默认前向填充把停牌/缺失日期伪装成真实行情——这是金融时间序列工程中容易忽视、却直接影响特征质量的细节。若你的数据包含复权除权跳空建议先对齐数据口径参考仓库对价格口径的既有约束见 README_zh.md 关于复权/未复权价混用告警的描述再做特征计算。模型选型指南ModelAdvantagesDisadvantagesApplicable ScenarioRandomForestClassifierHard to overfit, robust to hyperparameters, can output feature importanceWeaker at capturing trend-style featuresDefault first-choice model, medium data sizeGradientBoostingClassifierHigh accuracy, captures complex nonlinear relationshipsEasy to overfit, slow to train, requires careful tuningSufficient data and tuning experienceRidge / LogisticRegressionFast training, interpretable, difficult to overfitCaptures only linear relationshipsFast baseline, few features, small dataset代码中的默认实现细节与之对应随机森林默认n_estimators100, max_depth5浅树抗过拟合梯度提升默认max_depth3, learning_rate0.05小步长配合浅树进一步抑制过拟合ridge实际使用带 L2 惩罚的LogisticRegression(penaltyl2, C1.0)即岭式线性分类器适合作为快速基线。三者统一random_state42保证结果可复现。参数说明ParameterDefaultDescriptionmodel_typerandom_forestModel type:random_forest/gradient_boosting/ridgemin_train_size252Minimum training-set size (starting length of the expanding window)retrain_freq20Retraining frequency (every N trading days)prediction_horizon5Prediction horizon (future N-day return)n_estimators100Number of trees for tree-based modelsmax_depth5Maximum tree depth (prevents overfitting)threshold0.0Signal filtering threshold (abs(signal) thresholdis set to 0)window_typeexpandingTraining window:expanding(all history) orsliding(fixed lookback)sliding_size504Lookback size for sliding window (2 years of trading days)几个关键参数的设计意图min_train_size252对应约一年的交易日是扩展窗口的起点长度——太短则统计意义不足太长则冷启动成本高retrain_freq20约等于每月重训一次在模型时效性与训练成本之间取平衡prediction_horizon5定义标签视界即预测未来 5 日收益方向代码在入口处校验prediction_horizon 1非法值直接抛ValueErrorwindow_type决定训练窗口形态expanding用全部历史样本更多但可能混入过时 regimesliding只用最近sliding_size根 K 线504≈ 两年交易日更贴近 regime 变化频繁的市场threshold0.0是信号过滤阈值abs(signal) threshold的信号被置 0可用于在回测/实盘前剔除弱信号代码示例中该参数未显式使用属于文档声明的可调契约。常见陷阱代码已解决 vs 仍需人工判断文档明确区分了两类问题。代码已内置防护的有数据泄漏标签可观测性剔除、标准化泄漏scaler 只在训练集拟合、inf/NaN 传播消毒 样本过滤、重训频率retrain_freq。仍需你判断的陷阱有三类过拟合Overfitting树过深max_depth 10、特征过多、训练集过小都会导致。建议保持max_depth3~5特征数 15类别不平衡Class imbalance牛市环境涨跌比可能达 7:3模型会偏向预测多数类。必要时使用class_weightbalanced或 SMOTE 重采样前视偏差的非泄漏形态Look-ahead bias用今日收盘计算特征、又预测今日信号这不算未来泄漏但实践中应确保特征只用 T-1 及更早数据。对应到仓库的 swarm 预设Feature Engineer 的角色提示也强调所有特征必须严格 point-in-time 对齐用 t−1 信息预测 t 期收益见 ml_quant_lab.yaml并把相关性 0.85 的特征剔除、按 1%/99% 分位数去极值作为特征工程标准。依赖与运行环境pip install scikit-learn pandas numpy技能依赖仅三个库scikit-learn模型与标准化、pandas数据与特征、numpy数值操作。在 Vibe-Trading 项目中你可以在 Agent 会话中通过load_skill(ml-strategy)加载本文档或直接运行 CLIvibe-trading run -p Backtest a BTC-USDT 20/50 moving-average strategy for 2024 and summarize return and drawdown示例见 README_zh.md将本管线的信号接入回测。数据获取可借助仓库内置的数据加载器get_market_data工具或 MCP 同源 loader 注册表它们输出归一化 OHLCV与本技能的输入约定天然兼容。信号约定连续强度信号predict_proba[:, 1]正类概率经prob * 2 - 1映射到[-1.0, 1.0]离散信号使用predict()得到{-1, 0, 1}空头、中性、多头语义正值 看涨方向负值 看跌方向绝对值 置信强度输出契约保证无 NaN、无 inf数值裁剪到[-1.0, 1.0]——任何下游回测引擎、信号过滤、仓位构建都可以无条件消费该输出。这一契约与 Vibe-Trading 整体的严格 JSON/非有限值处理风格一致仓库在多处强调 NaN 泄漏进输出的危害参见 README_zh.md 中关于 worker 输出 NaN 泄漏进非严格 JSON 的修复记录也是该技能能安全嵌入 Agent 工具链的基石。参考路径速览技能文档本体agent/src/skills/ml-strategy/SKILL.md正确性测试防泄漏语义锁定agent/tests/test_ml_strategy_skill.pySwarm 多智能体应用预设agent/src/swarm/presets/ml_quant_lab.yaml技能加载机制agent/src/agent/skills.py技能库分类说明README_zh.md【免费下载链接】Vibe-TradingVibe-Trading: Your Personal Trading Agent项目地址: https://gitcode.com/GitHub_Trending/vi/Vibe-Trading创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考 Vibe-Trading ML 策略技能实战：基于 scikit-learn 的 Walk-Forward 机器学习预测策略
Vibe-Trading ML 策略技能实战基于 scikit-learn 的 Walk-Forward 机器学习预测策略【免费下载链接】Vibe-TradingVibe-Trading: Your Personal Trading Agent项目地址: https://gitcode.com/GitHub_Trending/vi/Vibe-TradingVibe-Trading 将机器学习策略封装为ml-strategy技能SKILL.md为 Agent 提供一套基于 scikit-learn 的可直接复制运行的完整策略管线从 OHLCV 数据校验、多因子特征工程到未来 N 日收益方向标签构建、walk-forward 滚动训练与信号生成。读完本文你将掌握这套防数据泄漏data leakage、带输出契约no NaN / 数值裁剪的SignalEngine的每一行实现理解特征、模型、参数与信号约定的设计意图并能在任何 OHLCV 数据集上落地自己的机器学习预测策略。技能定位Vibe-Trading 策略类技能之一Vibe-Trading 的内置技能库按类别组织ml-strategy属于Strategy策略类别与strategy-generate、cross-market-strategy、technical-basic、candlestick、ichimoku、elliott-wave、smc、multi-factor等同列见 README_zh.md。该类别共 19 个技能ml-strategy的职责非常聚焦用机器学习模型预测未来收益方向并生成交易信号且明确声明适用于任何 OHLCV 数据Suitable for any OHLCV data。技能文件采用统一的 frontmatter 元数据格式name/description/category由技能加载器读取。从 技能加载器实现 可以看到Skill数据类解析 SKILL.md 的 frontmatter 得到名称、描述与类别正文则通过load_skill工具按需注入 Agent 上下文。这种渐进式披露设计意味着ml-strategy的完整管线代码平时不会全部塞进系统提示而是在 Agent 需要时被精确加载。ml-strategy还深度接入 Vibe-Trading 的 Swarm 多智能体编排ml_quant_lab预设ml_quant_lab.yaml中Feature Engineer 与 Data Scientist 两个角色的skills字段都声明了ml-strategy并通过load_skill(ml-strategy)获取特征工程最佳实践与金融 ML 设计标准随后由 Backtest Engineer 对产物做严格样本外OOS验证。这说明该技能不仅面向单个 Agent也是多智能体量化实验室的标准参考手册。信号逻辑五步流水线ml-strategy的信号生成遵循严格有序的五步流程每一步都在为避免未来函数泄漏、保证输出可被下游直接消费服务校验输入Validate input检查 OHLCV 列是否齐全、最小行数是否达标、NaN 占比是否过高——不合格的标的直接跳过绝不进入训练流程特征工程Feature engineering从原始 OHLCV 构建动量、波动率、RSI、均线比、量比等多维因子所有特征统一做消毒处理inf替换、除零防护标签构建Label construction未来 N 日收益 0 记为正类1 0 记为负类0Walk-forward 训练采用扩展窗口expanding或滑动窗口sliding只用历史数据训练逐日向前滚动预测信号生成将predict_proba[:, 1]映射到[-1.0, 1.0]或使用predict得到{-1, 0, 1}离散信号输出保证干净无 NaN、数值已裁剪。其中第 3 步与第 4 步的组合是整套管线防泄漏的关键标签基于未来收益构造而训练样本只允许取标签在预测时刻已可观测的历史切片——这一点在测试中被专门验证见下文防泄漏的正确性验证。完整 SignalEngine 示例推荐的全管线实现这是该技能推荐的标准全流程实现复制即可运行安全性已内建。它是后续特征工程、模型选型、参数调优讨论的基准代码。import numpy as np import pandas as pd from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier from sklearn.linear_model import LogisticRegression from sklearn.preprocessing import StandardScaler def validate_data(df: pd.DataFrame, min_rows: int 300) - bool: Check that OHLCV data meets minimum quality for ML training. Args: df: DataFrame with DatetimeIndex. min_rows: Minimum number of rows required. Returns: True if data is usable. required {open, high, low, close, volume} if not required.issubset(df.columns): return False if len(df) min_rows: return False if df[close].isnull().mean() 0.2: return False return True def build_features(df: pd.DataFrame) - pd.DataFrame: Build a machine-learning feature matrix from OHLCV data. All features are guarded against division-by-zero and sanitized (inf replaced with NaN) so downstream code never sees inf values. Args: df: DataFrame containing open, high, low, close, and volume columns. Returns: DataFrame with feature columns prefixed by f_. c df[close] v df[volume] ret c.pct_change(fill_methodNone) features pd.DataFrame(indexdf.index) features[f_ret_5d] c.pct_change(5, fill_methodNone) features[f_ret_20d] c.pct_change(20, fill_methodNone) features[f_vol_20d] ret.rolling(20).std() features[f_ma_ratio] c / c.rolling(20).mean() features[f_volume_ratio] v / v.rolling(20).mean() # RSI(14) — guard: loss0 in zero-volatility periods produces inf delta c.diff() gain delta.clip(lower0).rolling(14).mean() loss (-delta.clip(upper0)).rolling(14).mean() rs gain / loss.replace(0, np.nan) features[f_rsi_14] 100 - (100 / (1 rs)) # Bollinger Band position — guard: bb_upper bb_lower when std0 ma20 c.rolling(20).mean() std20 c.rolling(20).std() bb_upper ma20 2 * std20 bb_lower ma20 - 2 * std20 bb_range (bb_upper - bb_lower).replace(0, np.nan) features[f_bb_position] (c - bb_lower) / bb_range # Intraday features features[f_high_low_ratio] (df[high] - df[low]) / c features[f_close_open_ratio] (c - df[open]) / df[open] features[f_skew_20d] ret.rolling(20).skew() # Sanitize: replace all inf with NaN (NaN handled by walk-forward) features features.replace([np.inf, -np.inf], np.nan) return features def walk_forward_predict( features: pd.DataFrame, labels: pd.Series, min_train_size: int 252, retrain_freq: int 20, model_type: str random_forest, window_type: str expanding, sliding_size: int 504, prediction_horizon: int 5, ) - pd.Series: Walk-forward training and prediction to avoid future data leakage. Args: features: Feature matrix aligned with labels by row index. labels: Binary labels (0/1), representing the direction of future N-day returns. min_train_size: Minimum training-set size in trading days. retrain_freq: Retrain the model every N days. model_type: One of random_forest / gradient_boosting / ridge. window_type: expanding uses all history; sliding uses a fixed lookback. sliding_size: Lookback window size when window_type is sliding. prediction_horizon: Number of bars each target label looks ahead. Returns: Predicted signal series with range [-1.0, 1.0], no NaN values. predictions pd.Series(0.0, indexfeatures.index) model None scaler None if prediction_horizon 1: raise ValueError(prediction_horizon must be 1) for i in range(min_train_size, len(features)): # Retrain every retrain_freq days if model is None or (i - min_train_size) % retrain_freq 0: # A label at row t is observable only once t horizon i. train_stop max(0, i - prediction_horizon 1) start ( max(0, train_stop - sliding_size) if window_type sliding else 0 ) X_train features.iloc[start:train_stop].values y_train labels.iloc[start:train_stop].values # Drop rows with NaN valid ~(np.isnan(X_train).any(axis1) | np.isnan(y_train)) X_train X_train[valid] y_train y_train[valid] if len(X_train) 50: continue # Standardization: fit only on training set scaler StandardScaler() X_train scaler.fit_transform(X_train) # Build the model if model_type random_forest: model RandomForestClassifier( n_estimators100, max_depth5, random_state42, ) elif model_type gradient_boosting: model GradientBoostingClassifier( n_estimators100, max_depth3, learning_rate0.05, random_state42, ) elif model_type ridge: model LogisticRegression(penaltyl2, C1.0, random_state42) else: raise ValueError(fUnsupported model_type: {model_type}) model.fit(X_train, y_train) # Predict today X_today features.iloc[i : i 1].values if np.isnan(X_today).any(): predictions.iloc[i] 0.0 continue X_today scaler.transform(X_today) if hasattr(model, predict_proba): prob model.predict_proba(X_today)[0, 1] predictions.iloc[i] prob * 2 - 1 # [0,1] - [-1,1] else: predictions.iloc[i] float(model.predict(X_today)[0]) # Output contract: no NaN, clipped to [-1, 1] predictions predictions.fillna(0.0).clip(-1.0, 1.0) return predictions class SignalEngine: Complete ML strategy with built-in data validation and safety. def generate(self, data_map: dict) - dict: Generate signals for each symbol. Args: data_map: code - OHLCV DataFrame. Returns: code - signal Series in [-1.0, 1.0]. signals {} for code, df in data_map.items(): if not validate_data(df): print(f[WARN] {code}: data quality insufficient, skipping) continue features build_features(df) prediction_horizon 5 future_returns ( df[close].shift(-prediction_horizon) / df[close] - 1 ) labels (future_returns 0).astype(float).where(future_returns.notna()) signal walk_forward_predict( features, labels, prediction_horizonprediction_horizon, ) signals[code] signal return signals各函数职责拆解validate_data(df, min_rows300)数据质量的守门员。要求列集合必须包含open/high/low/close/volume全部五列样本量低于min_rows默认 300 个交易日直接拒绝收盘价缺失率超过 20% 同样拒绝。这一层保证进入训练的数据具备最低统计意义避免用几根 K 线训练出的模型自欺欺人。build_features(df)特征工厂。产出 10 个以f_前缀命名的默认因子详见下节特征表并在返回前用replace([np.inf, -np.inf], np.nan)统一消毒——文档注释明确指出这一设计保证下游代码永远看不到 inf 值而 NaN 交由 walk-forward 的样本过滤处理。walk_forward_predict(...)防泄漏训练与预测的核心。逐日滚动每retrain_freq天重训一次每次重训时用train_stop max(0, i - prediction_horizon 1)把训练集截止点前移一个预测视界确保被训练的标签在预测时刻 i确实已经可观测标准化器StandardScaler只在训练集上fit杜绝标准化泄漏单日特征含 NaN 时该日输出 0.0 中性信号最终统一fillna(0.0).clip(-1.0, 1.0)。SignalEngine.generate(data_map)面向多标的的入口。接收code - OHLCV DataFrame的字典逐标的执行校验→特征→标签→walk-forward 预测返回code - signal Series。标签构建使用了shift(-prediction_horizon)的前移技巧where(future_returns.notna())把视界尚未走完的尾部标签保持为 NaN——这些行不会进入训练。防泄漏的正确性验证测试如何锁定行为技能文档不是空谈Vibe-Trading 用真实测试固定了 walk-forward 的防泄漏语义。仓库中的 test_ml_strategy_skill.py 通过 AST 解析直接从 SKILL.md 提取 Python 代码块执行这意味着文档代码块本身就是被测试的产物。三个测试分别验证test_walk_forward_purges_labels_not_observable_at_prediction_time构造 70 行数据、min_train_size60、retrain_freq100只重训一次、prediction_horizon5断言第一次fit收到的训练集最后一个样本行号是 55——即60 - 5 1 56个样本中的最后一行索引 55证明标签在预测时刻不可观测的尾部样本被精确剔除test_one_bar_horizon_preserves_existing_training_window同样的数据在prediction_horizon1时训练集末行是 59即完整 60 行都保留——视界为 1 时无需剔除语义自洽test_signal_engine_preserves_unavailable_future_labels_as_nan用 10 行 close 数据驱动SignalEngine.generate断言前 5 个标签为1.0、后 5 个标签为NaN同时确认prediction_horizon被正确传为 5——锁定未来标签不可用时保持 NaN的行为。这组测试是理解该技能正确性契约的最佳入口防泄漏不是注释里的愿望而是被 CI 固定下来的行为。特征工程参考默认因子表build_features()是自定义扩展点下表列出全部默认特征可按需增删。Feature NameFormulaMeaningret_5dclose.pct_change(5, fill_methodNone)Past 5-day return (short-term momentum)ret_20dclose.pct_change(20, fill_methodNone)Past 20-day return (medium-term momentum)vol_20dreturns.rolling(20).std()20-day volatilityrsi_14See RSI formula in codeRelative Strength Index (division-by-zero guarded)ma_ratioclose / close.rolling(20).mean()Degree of deviation from the 20-day moving averagevolume_ratiovolume / volume.rolling(20).mean()Volume ratio (current volume vs 20-day average)bb_position(close - bb_lower) / (bb_upper - bb_lower)Bollinger Band position (zero-bandwidth guarded)high_low_ratio(high - low) / closeIntraday range ratioclose_open_ratio(close - open) / openIntraday returnskew_20dreturns.rolling(20).skew()Return skewness这 10 个因子覆盖了四类信息维度动量ret_5d、ret_20d、波动vol_20d、bb_position、skew_20d、量能volume_ratio与日内结构high_low_ratio、close_open_ratio。其中rsi_14与bb_position是教科书式的易除零陷阱零波动时段loss0会让 RSI 产生 inf、std0会让布林带宽为 0代码分别用.replace(0, np.nan)规避——这正体现了该技能安全性内建的工程取向。值得注意的是pct_change与rolling均未指定fill_method之外的参数而是显式传入fill_methodNone避免默认前向填充把停牌/缺失日期伪装成真实行情——这是金融时间序列工程中容易忽视、却直接影响特征质量的细节。若你的数据包含复权除权跳空建议先对齐数据口径参考仓库对价格口径的既有约束见 README_zh.md 关于复权/未复权价混用告警的描述再做特征计算。模型选型指南ModelAdvantagesDisadvantagesApplicable ScenarioRandomForestClassifierHard to overfit, robust to hyperparameters, can output feature importanceWeaker at capturing trend-style featuresDefault first-choice model, medium data sizeGradientBoostingClassifierHigh accuracy, captures complex nonlinear relationshipsEasy to overfit, slow to train, requires careful tuningSufficient data and tuning experienceRidge / LogisticRegressionFast training, interpretable, difficult to overfitCaptures only linear relationshipsFast baseline, few features, small dataset代码中的默认实现细节与之对应随机森林默认n_estimators100, max_depth5浅树抗过拟合梯度提升默认max_depth3, learning_rate0.05小步长配合浅树进一步抑制过拟合ridge实际使用带 L2 惩罚的LogisticRegression(penaltyl2, C1.0)即岭式线性分类器适合作为快速基线。三者统一random_state42保证结果可复现。参数说明ParameterDefaultDescriptionmodel_typerandom_forestModel type:random_forest/gradient_boosting/ridgemin_train_size252Minimum training-set size (starting length of the expanding window)retrain_freq20Retraining frequency (every N trading days)prediction_horizon5Prediction horizon (future N-day return)n_estimators100Number of trees for tree-based modelsmax_depth5Maximum tree depth (prevents overfitting)threshold0.0Signal filtering threshold (abs(signal) thresholdis set to 0)window_typeexpandingTraining window:expanding(all history) orsliding(fixed lookback)sliding_size504Lookback size for sliding window (2 years of trading days)几个关键参数的设计意图min_train_size252对应约一年的交易日是扩展窗口的起点长度——太短则统计意义不足太长则冷启动成本高retrain_freq20约等于每月重训一次在模型时效性与训练成本之间取平衡prediction_horizon5定义标签视界即预测未来 5 日收益方向代码在入口处校验prediction_horizon 1非法值直接抛ValueErrorwindow_type决定训练窗口形态expanding用全部历史样本更多但可能混入过时 regimesliding只用最近sliding_size根 K 线504≈ 两年交易日更贴近 regime 变化频繁的市场threshold0.0是信号过滤阈值abs(signal) threshold的信号被置 0可用于在回测/实盘前剔除弱信号代码示例中该参数未显式使用属于文档声明的可调契约。常见陷阱代码已解决 vs 仍需人工判断文档明确区分了两类问题。代码已内置防护的有数据泄漏标签可观测性剔除、标准化泄漏scaler 只在训练集拟合、inf/NaN 传播消毒 样本过滤、重训频率retrain_freq。仍需你判断的陷阱有三类过拟合Overfitting树过深max_depth 10、特征过多、训练集过小都会导致。建议保持max_depth3~5特征数 15类别不平衡Class imbalance牛市环境涨跌比可能达 7:3模型会偏向预测多数类。必要时使用class_weightbalanced或 SMOTE 重采样前视偏差的非泄漏形态Look-ahead bias用今日收盘计算特征、又预测今日信号这不算未来泄漏但实践中应确保特征只用 T-1 及更早数据。对应到仓库的 swarm 预设Feature Engineer 的角色提示也强调所有特征必须严格 point-in-time 对齐用 t−1 信息预测 t 期收益见 ml_quant_lab.yaml并把相关性 0.85 的特征剔除、按 1%/99% 分位数去极值作为特征工程标准。依赖与运行环境pip install scikit-learn pandas numpy技能依赖仅三个库scikit-learn模型与标准化、pandas数据与特征、numpy数值操作。在 Vibe-Trading 项目中你可以在 Agent 会话中通过load_skill(ml-strategy)加载本文档或直接运行 CLIvibe-trading run -p Backtest a BTC-USDT 20/50 moving-average strategy for 2024 and summarize return and drawdown示例见 README_zh.md将本管线的信号接入回测。数据获取可借助仓库内置的数据加载器get_market_data工具或 MCP 同源 loader 注册表它们输出归一化 OHLCV与本技能的输入约定天然兼容。信号约定连续强度信号predict_proba[:, 1]正类概率经prob * 2 - 1映射到[-1.0, 1.0]离散信号使用predict()得到{-1, 0, 1}空头、中性、多头语义正值 看涨方向负值 看跌方向绝对值 置信强度输出契约保证无 NaN、无 inf数值裁剪到[-1.0, 1.0]——任何下游回测引擎、信号过滤、仓位构建都可以无条件消费该输出。这一契约与 Vibe-Trading 整体的严格 JSON/非有限值处理风格一致仓库在多处强调 NaN 泄漏进输出的危害参见 README_zh.md 中关于 worker 输出 NaN 泄漏进非严格 JSON 的修复记录也是该技能能安全嵌入 Agent 工具链的基石。参考路径速览技能文档本体agent/src/skills/ml-strategy/SKILL.md正确性测试防泄漏语义锁定agent/tests/test_ml_strategy_skill.pySwarm 多智能体应用预设agent/src/swarm/presets/ml_quant_lab.yaml技能加载机制agent/src/agent/skills.py技能库分类说明README_zh.md【免费下载链接】Vibe-TradingVibe-Trading: Your Personal Trading Agent项目地址: https://gitcode.com/GitHub_Trending/vi/Vibe-Trading创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考