From c3d1b157e9e5978a1b5f4f5a83da25bb4b4d15c0 Mon Sep 17 00:00:00 2001 From: liaozhaorun <1300336796@qq.com> Date: Sun, 29 Mar 2026 01:34:58 +0800 Subject: [PATCH] =?UTF-8?q?feat(factors):=20=E6=96=B0=E5=A2=9E=E7=AD=B9?= =?UTF-8?q?=E7=A0=81=E9=9B=86=E4=B8=AD=E5=BA=A6=E7=9B=B8=E5=85=B3=E5=9B=A0?= =?UTF-8?q?=E5=AD=90=E5=B9=B6=E4=BC=98=E5=8C=96=E8=AE=AD=E7=BB=83=E6=A1=86?= =?UTF-8?q?=E6=9E=B6?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - 添加 19 个筹码分布和胜率相关因子(包括chip_dispersion、winner_rate等系列) - LightGBM模型添加早停和训练指标记录功能 - 统一Label配置到common.py模块 - 新增list_factors.py因子列表脚本 --- src/experiment/common.py | 83 +++++----- src/experiment/learn_to_rank.py | 173 +++------------------ src/experiment/regression.py | 102 ++++++++---- src/scripts/list_factors.py | 81 ++++++++++ src/scripts/register_factors.py | 118 +++++++++++++- src/training/components/models/lightgbm.py | 44 +++++- src/training/result_analyzer.py | 2 +- src/training/tasks/rank_task.py | 8 +- src/training/tasks/regression_task.py | 8 +- 9 files changed, 373 insertions(+), 246 deletions(-) create mode 100644 src/scripts/list_factors.py diff --git a/src/experiment/common.py b/src/experiment/common.py index b53fd66..7826c28 100644 --- a/src/experiment/common.py +++ b/src/experiment/common.py @@ -250,67 +250,58 @@ SELECTED_FACTORS = [ "GTJA_alpha188", "GTJA_alpha189", "GTJA_alpha191", + "chip_dispersion_90", + "chip_dispersion_70", + "cost_skewness", + "dispersion_change_20", + "price_to_avg_cost", + "price_to_median_cost", + "mean_median_dev", + "trap_pressure", + "bottom_profit", + "history_position", + "winner_rate_surge_5", + "winner_rate_cs_rank", + "winner_rate_dev_20", + "winner_rate_volatility", + "smart_money_accumulation", + "winner_vol_corr_20", + "cost_base_momentum", + "bottom_cost_stability", + "pivot_reversion", + "chip_transition", ] # 因子定义字典（完整因子库，用于存放尚未注册到metadata的因子） FACTOR_DEFINITIONS = {"cs_rank_circ_mv": "cs_rank(circ_mv)"} -# 需要排除的因子列表（这些因子不会被计算和使用） -# 用于临时屏蔽效果不好的因子，无需从 SELECTED_FACTORS 中删除 -# EXCLUDED_FACTORS: List[str] = [ -# # "GTJA_alpha005", -# # "GTJA_alpha028", -# # "GTJA_alpha023", -# # "GTJA_alpha002", -# # "GTJA_alpha010", -# # "GTJA_alpha011", -# # "GTJA_alpha044", -# # "GTJA_alpha036", -# # "GTJA_alpha027", -# # "GTJA_alpha109", -# # "GTJA_alpha104", -# # "GTJA_alpha103", -# # "GTJA_alpha085", -# # "GTJA_alpha111", -# # "GTJA_alpha092", -# # "GTJA_alpha067", -# # "GTJA_alpha060", -# # "GTJA_alpha062", -# # "GTJA_alpha063", -# # "GTJA_alpha079", -# # "GTJA_alpha073", -# # "GTJA_alpha087", -# # "GTJA_alpha117", -# # "GTJA_alpha113", -# # "GTJA_alpha138", -# # "GTJA_alpha121", -# # "GTJA_alpha124", -# # "GTJA_alpha133", -# # "GTJA_alpha131", -# # "GTJA_alpha118", -# # "GTJA_alpha164", -# # "GTJA_alpha162", -# # "GTJA_alpha157", -# # "GTJA_alpha171", -# # "GTJA_alpha177", -# # "GTJA_alpha180", -# # "GTJA_alpha188", -# # "GTJA_alpha191", -# ] + +# ============================================================================= +# Label 配置（统一绑定 label_name 和 label_dsl） +# ============================================================================= +# Label 名称 +LABEL_NAME = "future_return_5" + +# Label DSL 公式 +LABEL_DSL = "(ts_delay(close, -5) / ts_delay(open, -1)) - 1" + +# Label 配置字典（绑定 name 和 dsl） +LABEL_FACTOR = {LABEL_NAME: LABEL_DSL} def get_label_factor(label_name: str) -> dict: """获取Label因子定义字典。 + 警告: 此函数已废弃，请直接使用 LABEL_FACTOR 常量。 + label_name 参数将被忽略，始终返回预定义的 LABEL_FACTOR。 + Args: - label_name: label因子名称 + label_name: label因子名称（已废弃，仅保留参数保持向后兼容） Returns: Label因子定义字典 """ - return { - label_name: "(ts_delay(close, -5) / ts_delay(open, -1)) - 1", - } + return LABEL_FACTOR # ============================================================================= diff --git a/src/experiment/learn_to_rank.py b/src/experiment/learn_to_rank.py index 25a1416..0f51caa 100644 --- a/src/experiment/learn_to_rank.py +++ b/src/experiment/learn_to_rank.py @@ -21,7 +21,8 @@ from src.training.components.filters import STFilter from src.experiment.common import ( SELECTED_FACTORS, FACTOR_DEFINITIONS, - get_label_factor, + LABEL_NAME, + LABEL_FACTOR, TRAIN_START, TRAIN_END, VAL_START, @@ -44,171 +45,39 @@ TRAINING_TYPE = "rank" # %% md # ## 2. 训练特定配置 # %% -# Label 配置 -LABEL_NAME = "future_return_5" -LABEL_FACTOR = get_label_factor(LABEL_NAME) +# Label 配置（从 common.py 统一导入） +# LABEL_NAME 和 LABEL_FACTOR 已在 common.py 中绑定，只需从 common 导入 # 分位数配置 N_QUANTILES = 20 # 排除的因子列表 EXCLUDED_FACTORS = [ - "volatility_5", - "volume_ratio_5_20", - "capital_retention_20", - "volatility_squeeze_5_60", - "drawdown_from_high_60", - "ma_ratio_5_20", - "bias_10", - "high_low_ratio", - "bbi_ratio", - "volatility_20", - "std_return_20", - "sharpe_ratio_20", - "ma_5", - "max_ret_20", - "CP", - "net_profit_yoy", - "debt_to_equity", - "EP_rank", - "turnover_rank", - "return_5_rank", - "ebit_rank", - "BP", - "EP", - "amihud_illiq_20", - "profit_margin", - "return_5", - "return_20", - "kaufman_ER_20", - "GTJA_alpha043", - "GTJA_alpha042", - "GTJA_alpha041", - "GTJA_alpha040", - "GTJA_alpha039", - "GTJA_alpha037", - "GTJA_alpha036", - "GTJA_alpha035", - "GTJA_alpha033", - "GTJA_alpha032", - "GTJA_alpha031", - "GTJA_alpha028", - "GTJA_alpha026", - "GTJA_alpha027", - "GTJA_alpha023", - "GTJA_alpha024", - "GTJA_alpha009", - "GTJA_alpha011", - "GTJA_alpha022", - "GTJA_alpha020", - "GTJA_alpha018", - "GTJA_alpha019", - "GTJA_alpha014", - "GTJA_alpha013", "GTJA_alpha010", - "GTJA_alpha001", - "GTJA_alpha003", - "GTJA_alpha002", - "GTJA_alpha004", "GTJA_alpha005", - "GTJA_alpha006", - "GTJA_alpha008", - "turnover_deviation", - "turnover_cv_20", - "roa", - "GTJA_alpha073", - "GTJA_alpha078", - "GTJA_alpha077", - "GTJA_alpha076", - "GTJA_alpha067", - "GTJA_alpha085", - "GTJA_alpha084", - "GTJA_alpha087", - "GTJA_alpha088", - "GTJA_alpha090", - "GTJA_alpha083", - "GTJA_alpha079", - "GTJA_alpha080", - "GTJA_alpha094", - "GTJA_alpha092", - "GTJA_alpha089", - "GTJA_alpha095", - "GTJA_alpha064", - "GTJA_alpha065", - "GTJA_alpha066", - "GTJA_alpha063", - "GTJA_alpha060", - "GTJA_alpha058", - "GTJA_alpha057", - "GTJA_alpha056", - "GTJA_alpha046", + "GTJA_alpha002", + "GTJA_alpha027", + "GTJA_alpha051", "GTJA_alpha044", - "GTJA_alpha049", - "GTJA_alpha050", - "GTJA_alpha110", - "GTJA_alpha107", - "GTJA_alpha104", - "GTJA_alpha106", - "GTJA_alpha103", - "GTJA_alpha100", - "GTJA_alpha101", - "GTJA_alpha102", - "GTJA_alpha098", - "GTJA_alpha097", - "GTJA_alpha096", - "GTJA_alpha099", - "GTJA_alpha117", - "GTJA_alpha118", - "GTJA_alpha114", - "GTJA_alpha111", - "GTJA_alpha129", - "GTJA_alpha130", - "GTJA_alpha132", + "GTJA_alpha041", "GTJA_alpha131", - "GTJA_alpha134", - "GTJA_alpha135", - "GTJA_alpha136", - "GTJA_alpha112", - "GTJA_alpha120", - "GTJA_alpha119", - "GTJA_alpha122", - "GTJA_alpha124", - "GTJA_alpha126", + "GTJA_alpha103", + "GTJA_alpha087", + "GTJA_alpha093", + "GTJA_alpha092", + "GTJA_alpha073", "GTJA_alpha127", - "GTJA_alpha128", - "GTJA_alpha115", - "GTJA_alpha153", - "GTJA_alpha152", - "GTJA_alpha151", - "GTJA_alpha150", - "GTJA_alpha148", - "GTJA_alpha142", - "GTJA_alpha141", - "GTJA_alpha139", - "GTJA_alpha133", - "GTJA_alpha161", - "GTJA_alpha164", + "GTJA_alpha117", + "GTJA_alpha124", "GTJA_alpha162", - "GTJA_alpha157", - "GTJA_alpha156", - "GTJA_alpha160", - "GTJA_alpha155", - "GTJA_alpha170", - "GTJA_alpha169", - "GTJA_alpha168", - "GTJA_alpha166", - "GTJA_alpha163", - "GTJA_alpha176", - "GTJA_alpha175", - "GTJA_alpha174", - "GTJA_alpha178", "GTJA_alpha177", - "GTJA_alpha185", - "GTJA_alpha180", - "GTJA_alpha187", "GTJA_alpha188", - "GTJA_alpha189", - "GTJA_alpha191", + "smart_money_accumulation", + "GTJA_alpha014", + "GTJA_alpha056", + "GTJA_alpha085", + "GTJA_alpha154", + "GTJA_alpha141", ] # LambdaRank 模型参数配置 diff --git a/src/experiment/regression.py b/src/experiment/regression.py index 225eebf..3fabbac 100644 --- a/src/experiment/regression.py +++ b/src/experiment/regression.py @@ -15,13 +15,15 @@ from src.training import ( NullFiller, Winsorizer, StandardScaler, + CrossSectionalStandardScaler, ) from src.training.core.trainer_v2 import Trainer from src.training.components.filters import STFilter from src.experiment.common import ( SELECTED_FACTORS, FACTOR_DEFINITIONS, - get_label_factor, + LABEL_NAME, + LABEL_FACTOR, TRAIN_START, TRAIN_END, VAL_START, @@ -44,58 +46,93 @@ TRAINING_TYPE = "regression" # %% md # ## 2. 训练特定配置 # %% -# Label 配置 -LABEL_NAME = "future_return_5" -LABEL_FACTOR = get_label_factor(LABEL_NAME) +# Label 配置（从 common.py 统一导入） +# LABEL_NAME 和 LABEL_FACTOR 已在 common.py 中绑定，只需从 common 导入 # 排除的因子列表 EXCLUDED_FACTORS = [ + "GTJA_alpha062", + "GTJA_alpha060", + "GTJA_alpha058", + "GTJA_alpha056", + "GTJA_alpha053", + "GTJA_alpha040", + "GTJA_alpha043", + "GTJA_alpha027", + "CP", + "max_ret_20", + "debt_to_equity", + "close_vwap_deviation", + "EP", + "BP", + "EP_rank", + "GTJA_alpha044", + "GTJA_alpha036", "GTJA_alpha010", "GTJA_alpha005", - "GTJA_alpha036", - "GTJA_alpha027", - "GTJA_alpha044", + "GTJA_alpha001", + "GTJA_alpha002", + "GTJA_alpha007", + "GTJA_alpha016", "GTJA_alpha073", - "GTJA_alpha104", - "GTJA_alpha103", - "GTJA_alpha105", - "GTJA_alpha092", - "GTJA_alpha087", - "GTJA_alpha085", - "GTJA_alpha062", - "GTJA_alpha124", "GTJA_alpha133", "GTJA_alpha131", "GTJA_alpha117", + "GTJA_alpha124", + "GTJA_alpha120", + "GTJA_alpha119", + "GTJA_alpha103", + "GTJA_alpha099", + "GTJA_alpha105", + "GTJA_alpha104", + "GTJA_alpha090", + "GTJA_alpha085", + "GTJA_alpha083", + "GTJA_alpha084", + "GTJA_alpha087", + "GTJA_alpha092", + "GTJA_alpha074", + "GTJA_alpha089", + "GTJA_alpha173", "GTJA_alpha157", + "GTJA_alpha139", "GTJA_alpha162", + "GTJA_alpha163", "GTJA_alpha177", - "GTJA_alpha180", + "price_to_avg_cost", + "cost_skewness", "GTJA_alpha191", + "GTJA_alpha180", + "history_position", + "bottom_profit", + "smart_money_accumulation", ] # 模型参数配置 MODEL_PARAMS = { - # 基础设置 - "objective": "regression_l1", + # ==================== 基础设置 ==================== + "objective": "huber", # 【修改】相比纯 L1(MAE)，huber 对异常值鲁棒且在极小误差处平滑，更适合收益率预测 "metric": "mae", - # 树结构约束 - "max_depth": 5, - "num_leaves": 24, - "min_data_in_leaf": 100, - # 学习参数 - "learning_rate": 0.01, - "n_estimators": 1500, - # 随机采样 - "subsample": 0.8, + # ==================== 树结构约束 ==================== + "max_depth": 5, # 【修改】适当加深，允许捕捉一定的高阶交叉 + "num_leaves": 31, # 【修改】限制为 31（2的5次方-1），确保树是不对称生长的，防止过拟合 + "min_data_in_leaf": 512, # 【大幅增加】从256加到1000。训练集有97万条，极大地限制叶子节点样本量能有效抵抗股市噪音 + # ==================== 学习参数 ==================== + "learning_rate": 0.02, # 【修改】稍微调大一点，帮助模型跳出初始的局部最优（避免十几轮就早停） + "n_estimators": 2000, + # ==================== 随机采样与降维 ==================== + "subsample": 0.85, "subsample_freq": 1, - "colsample_bytree": 0.8, - # 正则化 - "reg_alpha": 0.5, - "reg_lambda": 1.0, - # 杂项 + "colsample_bytree": 0.4, # 【大幅降低】从0.8降到0.4。强制打压 GTJA_alpha127 的霸权，逼迫模型去学习其他因子的信息 + "extra_trees": True, # 【新增且极度推荐】极度随机树模式。在分裂点选择时增加随机性，是量化比赛中防过拟合的神器 + # ==================== 正则化 ==================== + "reg_alpha": 1.0, # 【修改】L1正则增加，强行把一些无用特征的权重压到0 + "reg_lambda": 5.0, # 【修改】L2正则大幅增加（从1到5），惩罚过大的叶子节点输出权重 + "max_bin": 127, # 【新增】默认255，降低到127相当于对连续特征做了一次粗颗粒度的分箱，也是极好的正则化手段 + # ==================== 杂项 ==================== "verbose": -1, "random_state": 42, + "n_jobs": -1, } # 日期范围配置 @@ -143,6 +180,7 @@ def main(): (NullFiller, {"strategy": "mean"}), (Winsorizer, {"lower": 0.01, "upper": 0.99}), (StandardScaler, {}), + # (CrossSectionalStandardScaler, {}), ], filters=[STFilter(data_router=engine.router)], stock_pool_filter_func=stock_pool_filter, diff --git a/src/scripts/list_factors.py b/src/scripts/list_factors.py new file mode 100644 index 0000000..62fcffe --- /dev/null +++ b/src/scripts/list_factors.py @@ -0,0 +1,81 @@ +"""列出所有已入库的因子。 + +以 Python 列表格式输出所有已注册因子的名称，方便复制使用。 +保持 factors.jsonl 中的原始顺序（按 factor_id）。 + +使用方法: + uv run python -m src.scripts.list_factors +""" + +import json +import re +from pathlib import Path + +from src.config.settings import get_settings + + +def extract_factor_id_number(factor_id: str) -> int: + """从 factor_id 中提取数字部分用于排序。 + + Args: + factor_id: 如 "F_001" + + Returns: + 数字部分，如 1 + """ + match = re.match(r"F_(\d+)", factor_id) + if match: + return int(match.group(1)) + return 0 + + +def list_factors(): + """读取 factors.jsonl 并按 factor_id 顺序打印因子名称列表。""" + settings = get_settings() + factors_path = settings.data_path_resolved / "factors.jsonl" + + if not factors_path.exists(): + print(f"[错误] 因子文件不存在: {factors_path}") + return + + # 读取所有因子并按 factor_id 排序 + factors = [] + try: + with open(factors_path, "r", encoding="utf-8") as f: + for line in f: + line = line.strip() + if not line: + continue + try: + data = json.loads(line) + factor_id = data.get("factor_id", "") + name = data.get("name") + if name and factor_id: + factors.append((factor_id, name)) + except json.JSONDecodeError: + continue + except Exception as e: + print(f"[错误] 读取因子文件失败: {e}") + return + + if not factors: + print("[信息] 没有找到任何因子") + return + + # 按 factor_id 数字排序（保持入库顺序） + factors.sort(key=lambda x: extract_factor_id_number(x[0])) + + # 以 Python 列表格式输出 + print("[") + for i, (factor_id, name) in enumerate(factors): + if i == len(factors) - 1: + print(f' "{name}"') + else: + print(f' "{name}",') + print("]") + + print(f"\n[统计] 共计 {len(factors)} 个因子") + + +if __name__ == "__main__": + list_factors() diff --git a/src/scripts/register_factors.py b/src/scripts/register_factors.py index 122570a..76108e2 100644 --- a/src/scripts/register_factors.py +++ b/src/scripts/register_factors.py @@ -26,24 +26,126 @@ from typing import Any, Dict, List, Optional from src.factors.metadata import FactorManager from src.factors.metadata.exceptions import DuplicateFactorError, ValidationError +from src.config.settings import get_settings # ============================================================================ # 用户配置区域 - 在这里添加要注册的因子 # ============================================================================ -FACTORS: List[Dict[str, Any]] = [ - # 示例因子，请根据实际需要修改或添加 +FACTORS: List[Dict[str, Any]] =[ + # ==================== 第一类：筹码集中度与离散度因子 ==================== { - "name": "turnover_volatility_ratio", - "desc": "5日价格动量，收盘价相对于5日前收盘价的涨跌幅进行截面排名", - "dsl": "cs_rank(close / ts_delay(close, 5) - 1)", - "category": "momentum", + "name": "chip_dispersion_90", + "desc": "90%筹码离散度：衡量市场90%持仓筹码的宽度，值越小表示筹码越高度集中（单峰密集），往往是洗盘结束的前兆", + "dsl": "(cost_95pct - cost_5pct) / (cost_95pct + cost_5pct)", + }, + { + "name": "chip_dispersion_70", + "desc": "70%核心筹码离散度：剔除极端的底部死筹和高位套牢盘，反映中间70%主流资金的成本集中度", + "dsl": "(cost_85pct - cost_15pct) / (cost_85pct + cost_15pct)", + }, + { + "name": "cost_skewness", + "desc": "筹码偏度：反映筹码分布的不对称性。大于1说明上方套牢盘拖尾严重，小于1说明下方获利盘雄厚", + "dsl": "(cost_95pct - cost_50pct) / (cost_50pct - cost_5pct)", + }, + { + "name": "dispersion_change_20", + "desc": "筹码集中度近期变化率：过去20天筹码宽度的变化比例，持续下降说明主力正在暗中吸筹", + "dsl": "ts_pct_change((cost_95pct - cost_5pct) / cost_50pct, 20)", + }, + + # ==================== 第二类：筹码相对位置与压力/支撑因子 ==================== + { + "name": "price_to_avg_cost", + "desc": "整体浮盈比例：当前价格相对加权平均成本的溢价率。高溢价有均值回归压力，负溢价代表超跌", + "dsl": "(close - weight_avg) / weight_avg", + }, + { + "name": "price_to_median_cost", + "desc": "中位数成本偏离度：价格相对于50%分位点（绝对半数人持仓价）的偏离，向上突破通常是右侧买点", + "dsl": "(close - cost_50pct) / cost_50pct", + }, + { + "name": "mean_median_dev", + "desc": "均值中位数背离：均值显著大于中位数说明高位筹码堆积，上涨阻力大", + "dsl": "(weight_avg - cost_50pct) / cost_50pct", + }, + { + "name": "trap_pressure", + "desc": "高位套牢盘压力指数：当前价格距离上方95%高位套牢成本的距离。距离越大，反弹的真空期阻力越小", + "dsl": "(cost_95pct - close) / close", + }, + { + "name": "bottom_profit", + "desc": "底部支撑底仓利润率：当前价格距离底部5%筹码的利润空间。暴跌时大于0说明底仓极度稳定", + "dsl": "(close - cost_5pct) / cost_5pct", + }, + { + "name": "history_position", + "desc": "历史区间分位点：当前价格在个股上市以来历史最高点和最低点之间的相对位置", + "dsl": "(close - his_low) / (his_high - his_low)", + }, + + # ==================== 第三类：胜率相关的动量与反转因子 ==================== + { + "name": "winner_rate_surge_5", + "desc": "获利盘短期爆发力：胜率在过去5天内的变化值，急剧上升是极强的动量做多信号", + "dsl": "ts_delta(winner_rate, 5)", + }, + { + "name": "winner_rate_cs_rank", + "desc": "获利盘高位反转信号：全市场胜率截面排名，极端高胜率往往面临多头踩踏的获利了结压力（反转因子）", + "dsl": "cs_rank(winner_rate)", + }, + { + "name": "winner_rate_dev_20", + "desc": "获利盘均线偏离：当前胜率相对过去20天平均胜率的偏离程度，捕捉筹码情绪的边际超买/超卖", + "dsl": "winner_rate - ts_mean(winner_rate, 20)", + }, + { + "name": "winner_rate_volatility", + "desc": "获利盘波动率：过去20天胜率的波动率。波动率低且胜率高说明单边上涨极度稳健", + "dsl": "ts_std(winner_rate, 20)", + }, + { + "name": "smart_money_accumulation", + "desc": "潜在主力吸筹隐蔽指标：胜率的60日时序分位数减去价格的时序分位数。值越大说明‘价平而获利盘增’，底部吸筹明显", + "dsl": "ts_rank(winner_rate, 60) - ts_rank(close, 60)", + }, + + # ==================== 第四类：量价与筹码交乘因子 ==================== + { + "name": "winner_vol_corr_20", + "desc": "放量突破筹码密集区：胜率与成交量的20日时序相关性，正相关说明增量资金在主动解套上方筹码", + "dsl": "ts_corr(winner_rate, vol, 20)", + }, + { + "name": "cost_base_momentum", + "desc": "成本重心上移换手率：过去20天加权平均成本的变化幅度，快速上移说明高位换手极其充分", + "dsl": "ts_pct_change(weight_avg, 20)", + }, + { + "name": "bottom_cost_stability", + "desc": "底部坚如磐石因子：底部5%成本的60天波动率相对于中位数的比值，波动越小说明死筹越稳固", + "dsl": "ts_std(cost_5pct, 60) / cost_50pct", + }, + { + "name": "pivot_reversion", + "desc": "盈亏分界线乖离修复：价格偏离50%分位点除以近20日价格标准差，用于寻找超跌后的均值回归买点", + "dsl": "(close - cost_50pct) / ts_std(close, 20)", + }, + { + "name": "chip_transition", + "desc": "强弱筹码切换度：上方厚度与下方厚度差值的20日变化量。由正变负说明筹码彻底完成了自上而下的转移（洗盘结束）", + "dsl": "ts_delta((cost_85pct - cost_50pct) - (cost_50pct - cost_15pct), 20)", }, ] -# 因子存储路径（默认使用实验目录） -OUTPUT_PATH = Path(__file__).parent.parent / "experiment" / "data" / "factors.jsonl" +# 因子存储路径（使用项目根路径下的 data 目录） +settings = get_settings() +OUTPUT_PATH = settings.data_path_resolved / "factors.jsonl" # ============================================================================ diff --git a/src/training/components/models/lightgbm.py b/src/training/components/models/lightgbm.py index 34ba654..1aae1c2 100644 --- a/src/training/components/models/lightgbm.py +++ b/src/training/components/models/lightgbm.py @@ -48,6 +48,7 @@ class LightGBMModel(BaseModel): self.params = dict(params) if params is not None else {} self.model = None self.feature_names_: Optional[list] = None + self.evals_result_: Optional[dict] = None def fit( self, @@ -90,14 +91,23 @@ class LightGBMModel(BaseModel): y_val_np = y_val.to_numpy() valid_sets = lgb.Dataset(X_val_np, label=y_val_np, reference=train_data) - # 从 params 中提取 num_boost_round，默认 100 - num_boost_round = self.params.pop("n_estimators", 100) + # 从 params 中提取训练控制参数 + params_copy = dict(self.params) + num_boost_round = params_copy.pop("n_estimators", 100) + early_stopping_round = params_copy.pop("early_stopping_round", 50) + + self.evals_result_ = {} + callbacks = [ + lgb.early_stopping(stopping_rounds=early_stopping_round), + lgb.record_evaluation(self.evals_result_), + ] self.model = lgb.train( - self.params, + params_copy, train_data, num_boost_round=num_boost_round, valid_sets=[valid_sets] if valid_sets else None, + callbacks=callbacks, ) return self @@ -121,6 +131,34 @@ class LightGBMModel(BaseModel): result = self.model.predict(X_np) return np.asarray(result) + def get_evals_result(self) -> Optional[dict]: + """获取训练评估结果 + + Returns: + 评估结果字典，如果模型尚未训练返回 None + """ + return self.evals_result_ + + def get_best_iteration(self) -> Optional[int]: + """获取最佳迭代轮数（考虑早停） + + Returns: + 最佳迭代轮数，如果模型未训练返回 None + """ + if self.model is None: + return None + return self.model.best_iteration + + def get_best_score(self) -> Optional[dict]: + """获取最佳评分 + + Returns: + 最佳评分字典，如果模型未训练返回 None + """ + if self.model is None: + return None + return self.model.best_score + def feature_importance(self) -> Optional[pd.Series]: """返回特征重要性 diff --git a/src/training/result_analyzer.py b/src/training/result_analyzer.py index 29b5587..e589883 100644 --- a/src/training/result_analyzer.py +++ b/src/training/result_analyzer.py @@ -84,7 +84,7 @@ class ResultAnalyzer: print("\n" + "-" * 80) print(f"[警告] 贡献为0的特征（共 {len(zero_importance_features)} 个）:") for i, feature in enumerate(zero_importance_features, 1): - print(f" {i}. {feature}") + print(f"'{feature}',") # 统计摘要 print("\n" + "=" * 80) diff --git a/src/training/tasks/rank_task.py b/src/training/tasks/rank_task.py index 4a31fdc..0afdc26 100644 --- a/src/training/tasks/rank_task.py +++ b/src/training/tasks/rank_task.py @@ -189,10 +189,14 @@ class RankTask(BaseTask): def plot_training_metrics(self) -> None: """绘制训练指标曲线（NDCG）""" - if self.model and hasattr(self.model, "model") and self.model.model: + if self.model and hasattr(self.model, "get_evals_result"): try: import lightgbm as lgb - lgb.plot_metric(self.model.model) + evals_result = self.model.get_evals_result() + if evals_result: + lgb.plot_metric(evals_result) + else: + print("[警告] 没有训练指标数据可供绘制") except Exception as e: print(f"[警告] 无法绘制训练曲线: {e}") diff --git a/src/training/tasks/regression_task.py b/src/training/tasks/regression_task.py index a3ceb79..9ae6ad7 100644 --- a/src/training/tasks/regression_task.py +++ b/src/training/tasks/regression_task.py @@ -77,10 +77,14 @@ class RegressionTask(BaseTask): def plot_training_metrics(self) -> None: """绘制训练指标曲线""" - if self.model and hasattr(self.model, "model") and self.model.model: + if self.model and hasattr(self.model, "get_evals_result"): try: import lightgbm as lgb - lgb.plot_metric(self.model.model) + evals_result = self.model.get_evals_result() + if evals_result: + lgb.plot_metric(evals_result) + else: + print("[警告] 没有训练指标数据可供绘制") except Exception as e: print(f"[警告] 无法绘制训练曲线: {e}")