2559 lines
187 KiB
Plaintext
2559 lines
187 KiB
Plaintext
{
|
||
"cells": [
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 3,
|
||
"id": "79a7758178bafdd3",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:46:06.987506Z",
|
||
"start_time": "2025-04-03T12:46:06.259551Z"
|
||
},
|
||
"jupyter": {
|
||
"source_hidden": true
|
||
}
|
||
},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"The autoreload extension is already loaded. To reload it, use:\n",
|
||
" %reload_ext autoreload\n",
|
||
"/mnt/d/PyProject/NewStock/main/train\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"%load_ext autoreload\n",
|
||
"%autoreload 2\n",
|
||
"\n",
|
||
"import gc\n",
|
||
"import os\n",
|
||
"import sys\n",
|
||
"sys.path.append('../../')\n",
|
||
"print(os.getcwd())\n",
|
||
"import pandas as pd\n",
|
||
"from main.factor.factor import get_rolling_factor, get_simple_factor\n",
|
||
"from main.utils.factor import read_industry_data\n",
|
||
"from main.utils.factor_processor import calculate_score\n",
|
||
"from main.utils.utils import read_and_merge_h5_data, merge_with_industry_data\n",
|
||
"\n",
|
||
"import warnings\n",
|
||
"\n",
|
||
"warnings.filterwarnings(\"ignore\")\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 4,
|
||
"id": "4a481c60",
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"# 设置使用核心\n",
|
||
"import os\n",
|
||
"os.environ[\"MODIN_CPUS\"] = \"4\"\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 6,
|
||
"id": "a79cafb06a7e0e43",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:47:00.212859Z",
|
||
"start_time": "2025-04-03T12:46:06.998047Z"
|
||
},
|
||
"scrolled": true
|
||
},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"daily data\n",
|
||
"daily basic\n",
|
||
"inner merge on ['ts_code', 'trade_date']\n",
|
||
"stk limit\n",
|
||
"left merge on ['ts_code', 'trade_date']\n",
|
||
"money flow\n",
|
||
"left merge on ['ts_code', 'trade_date']\n",
|
||
"cyq perf\n",
|
||
"left merge on ['ts_code', 'trade_date']\n",
|
||
"<class 'pandas.core.frame.DataFrame'>\n",
|
||
"RangeIndex: 8665405 entries, 0 to 8665404\n",
|
||
"Data columns (total 33 columns):\n",
|
||
" # Column Dtype \n",
|
||
"--- ------ ----- \n",
|
||
" 0 ts_code object \n",
|
||
" 1 trade_date datetime64[ns]\n",
|
||
" 2 open float64 \n",
|
||
" 3 close float64 \n",
|
||
" 4 high float64 \n",
|
||
" 5 low float64 \n",
|
||
" 6 vol float64 \n",
|
||
" 7 amount float64 \n",
|
||
" 8 pct_chg float64 \n",
|
||
" 9 turnover_rate float64 \n",
|
||
" 10 pe_ttm float64 \n",
|
||
" 11 circ_mv float64 \n",
|
||
" 12 total_mv float64 \n",
|
||
" 13 volume_ratio float64 \n",
|
||
" 14 is_st bool \n",
|
||
" 15 up_limit float64 \n",
|
||
" 16 down_limit float64 \n",
|
||
" 17 buy_sm_vol float64 \n",
|
||
" 18 sell_sm_vol float64 \n",
|
||
" 19 buy_lg_vol float64 \n",
|
||
" 20 sell_lg_vol float64 \n",
|
||
" 21 buy_elg_vol float64 \n",
|
||
" 22 sell_elg_vol float64 \n",
|
||
" 23 net_mf_vol float64 \n",
|
||
" 24 his_low float64 \n",
|
||
" 25 his_high float64 \n",
|
||
" 26 cost_5pct float64 \n",
|
||
" 27 cost_15pct float64 \n",
|
||
" 28 cost_50pct float64 \n",
|
||
" 29 cost_85pct float64 \n",
|
||
" 30 cost_95pct float64 \n",
|
||
" 31 weight_avg float64 \n",
|
||
" 32 winner_rate float64 \n",
|
||
"dtypes: bool(1), datetime64[ns](1), float64(30), object(1)\n",
|
||
"memory usage: 2.1+ GB\n",
|
||
"None\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"from main.utils.utils import read_and_merge_h5_data\n",
|
||
"\n",
|
||
"print('daily data')\n",
|
||
"df = read_and_merge_h5_data('../../data/daily_data.h5', key='daily_data',\n",
|
||
" columns=['ts_code', 'trade_date', 'open', 'close', 'high', 'low', 'vol', 'amount', 'pct_chg'],\n",
|
||
" df=None)\n",
|
||
"\n",
|
||
"print('daily basic')\n",
|
||
"df = read_and_merge_h5_data('../../data/daily_basic.h5', key='daily_basic',\n",
|
||
" columns=['ts_code', 'trade_date', 'turnover_rate', 'pe_ttm', 'circ_mv', 'total_mv', 'volume_ratio',\n",
|
||
" 'is_st'], df=df, join='inner')\n",
|
||
"\n",
|
||
"print('stk limit')\n",
|
||
"df = read_and_merge_h5_data('../../data/stk_limit.h5', key='stk_limit',\n",
|
||
" columns=['ts_code', 'trade_date', 'pre_close', 'up_limit', 'down_limit'],\n",
|
||
" df=df)\n",
|
||
"print('money flow')\n",
|
||
"df = read_and_merge_h5_data('../../data/money_flow.h5', key='money_flow',\n",
|
||
" columns=['ts_code', 'trade_date', 'buy_sm_vol', 'sell_sm_vol', 'buy_lg_vol', 'sell_lg_vol',\n",
|
||
" 'buy_elg_vol', 'sell_elg_vol', 'net_mf_vol'],\n",
|
||
" df=df)\n",
|
||
"print('cyq perf')\n",
|
||
"df = read_and_merge_h5_data('../../data/cyq_perf.h5', key='cyq_perf',\n",
|
||
" columns=['ts_code', 'trade_date', 'his_low', 'his_high', 'cost_5pct', 'cost_15pct',\n",
|
||
" 'cost_50pct',\n",
|
||
" 'cost_85pct', 'cost_95pct', 'weight_avg', 'winner_rate'],\n",
|
||
" df=df)\n",
|
||
"print(df.info())"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 7,
|
||
"id": "cac01788dac10678",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:47:10.527104Z",
|
||
"start_time": "2025-04-03T12:47:00.488715Z"
|
||
}
|
||
},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"industry\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"print('industry')\n",
|
||
"industry_df = read_and_merge_h5_data('../../data/industry_data.h5', key='industry_data',\n",
|
||
" columns=['ts_code', 'l2_code', 'in_date'],\n",
|
||
" df=None, on=['ts_code'], join='left')\n",
|
||
"\n",
|
||
"\n",
|
||
"def merge_with_industry_data(df, industry_df):\n",
|
||
" # 确保日期字段是 datetime 类型\n",
|
||
" df['trade_date'] = pd.to_datetime(df['trade_date'])\n",
|
||
" industry_df['in_date'] = pd.to_datetime(industry_df['in_date'])\n",
|
||
"\n",
|
||
" # 对 industry_df 按 ts_code 和 in_date 排序\n",
|
||
" industry_df_sorted = industry_df.sort_values(['in_date', 'ts_code'])\n",
|
||
"\n",
|
||
" # 对原始 df 按 ts_code 和 trade_date 排序\n",
|
||
" df_sorted = df.sort_values(['trade_date', 'ts_code'])\n",
|
||
"\n",
|
||
" # 使用 merge_asof 进行向后合并\n",
|
||
" merged = pd.merge_asof(\n",
|
||
" df_sorted,\n",
|
||
" industry_df_sorted,\n",
|
||
" by='ts_code', # 按 ts_code 分组\n",
|
||
" left_on='trade_date',\n",
|
||
" right_on='in_date',\n",
|
||
" direction='backward'\n",
|
||
" )\n",
|
||
"\n",
|
||
" # 获取每个 ts_code 的最早 in_date 记录\n",
|
||
" min_in_date_per_ts = (industry_df_sorted\n",
|
||
" .groupby('ts_code')\n",
|
||
" .first()\n",
|
||
" .reset_index()[['ts_code', 'l2_code']])\n",
|
||
"\n",
|
||
" # 填充未匹配到的记录（trade_date 早于所有 in_date 的情况）\n",
|
||
" merged['l2_code'] = merged['l2_code'].fillna(\n",
|
||
" merged['ts_code'].map(min_in_date_per_ts.set_index('ts_code')['l2_code'])\n",
|
||
" )\n",
|
||
"\n",
|
||
" # 保留需要的列并重置索引\n",
|
||
" result = merged.reset_index(drop=True)\n",
|
||
" return result\n",
|
||
"\n",
|
||
"\n",
|
||
"# 使用示例\n",
|
||
"df = merge_with_industry_data(df, industry_df)\n",
|
||
"# print(mdf[mdf['ts_code'] == '600751.SH'][['ts_code', 'trade_date', 'l2_code']])"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 8,
|
||
"id": "c4e9e1d31da6dba6",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:47:10.719252Z",
|
||
"start_time": "2025-04-03T12:47:10.541247Z"
|
||
},
|
||
"jupyter": {
|
||
"source_hidden": true
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"from main.factor.factor import *\n",
|
||
"\n",
|
||
"def calculate_indicators(df):\n",
|
||
" \"\"\"\n",
|
||
" 计算四个指标：当日涨跌幅、5日移动平均、RSI、MACD。\n",
|
||
" \"\"\"\n",
|
||
" df = df.sort_values('trade_date')\n",
|
||
" df['daily_return'] = (df['close'] - df['pre_close']) / df['pre_close'] * 100\n",
|
||
" # df['5_day_ma'] = df['close'].rolling(window=5).mean()\n",
|
||
" delta = df['close'].diff()\n",
|
||
" gain = delta.where(delta > 0, 0)\n",
|
||
" loss = -delta.where(delta < 0, 0)\n",
|
||
" avg_gain = gain.rolling(window=14).mean()\n",
|
||
" avg_loss = loss.rolling(window=14).mean()\n",
|
||
" rs = avg_gain / avg_loss\n",
|
||
" df['RSI'] = 100 - (100 / (1 + rs))\n",
|
||
"\n",
|
||
" # 计算MACD\n",
|
||
" ema12 = df['close'].ewm(span=12, adjust=False).mean()\n",
|
||
" ema26 = df['close'].ewm(span=26, adjust=False).mean()\n",
|
||
" df['MACD'] = ema12 - ema26\n",
|
||
" df['Signal_line'] = df['MACD'].ewm(span=9, adjust=False).mean()\n",
|
||
" df['MACD_hist'] = df['MACD'] - df['Signal_line']\n",
|
||
"\n",
|
||
" # 4. 情绪因子1：市场上涨比例（Up Ratio）\n",
|
||
" df['up_ratio'] = df['daily_return'].apply(lambda x: 1 if x > 0 else 0)\n",
|
||
" df['up_ratio_20d'] = df['up_ratio'].rolling(window=20).mean() # 过去20天上涨比例\n",
|
||
"\n",
|
||
" # 5. 情绪因子2：成交量变化率（Volume Change Rate）\n",
|
||
" df['volume_mean'] = df['vol'].rolling(window=20).mean() # 过去20天的平均成交量\n",
|
||
" df['volume_change_rate'] = (df['vol'] - df['volume_mean']) / df['volume_mean'] * 100 # 成交量变化率\n",
|
||
"\n",
|
||
" # 6. 情绪因子3：波动率（Volatility）\n",
|
||
" df['volatility'] = df['daily_return'].rolling(window=20).std() # 过去20天的日收益率标准差\n",
|
||
"\n",
|
||
" # 7. 情绪因子4：成交额变化率（Amount Change Rate）\n",
|
||
" df['amount_mean'] = df['amount'].rolling(window=20).mean() # 过去20天的平均成交额\n",
|
||
" df['amount_change_rate'] = (df['amount'] - df['amount_mean']) / df['amount_mean'] * 100 # 成交额变化率\n",
|
||
"\n",
|
||
" # df = sentiment_panic_greed_index(df)\n",
|
||
" # df = sentiment_market_breadth_proxy(df)\n",
|
||
" # df = sentiment_reversal_indicator(df)\n",
|
||
"\n",
|
||
" return df\n",
|
||
"\n",
|
||
"\n",
|
||
"def generate_index_indicators(h5_filename):\n",
|
||
" df = pd.read_hdf(h5_filename, key='index_data')\n",
|
||
" df['trade_date'] = pd.to_datetime(df['trade_date'], format='%Y%m%d')\n",
|
||
" df = df.sort_values('trade_date')\n",
|
||
"\n",
|
||
" # 计算每个ts_code的相关指标\n",
|
||
" df_indicators = []\n",
|
||
" for ts_code in df['ts_code'].unique():\n",
|
||
" df_index = df[df['ts_code'] == ts_code].copy()\n",
|
||
" df_index = calculate_indicators(df_index)\n",
|
||
" df_indicators.append(df_index)\n",
|
||
"\n",
|
||
" # 合并所有指数的结果\n",
|
||
" df_all_indicators = pd.concat(df_indicators, ignore_index=True)\n",
|
||
"\n",
|
||
" # 保留trade_date列，并将同一天的数据按ts_code合并成一行\n",
|
||
" df_final = df_all_indicators.pivot_table(\n",
|
||
" index='trade_date',\n",
|
||
" columns='ts_code',\n",
|
||
" values=['daily_return', \n",
|
||
" 'RSI', 'MACD', 'Signal_line', 'MACD_hist', \n",
|
||
" # 'sentiment_panic_greed_index',\n",
|
||
" 'up_ratio_20d', 'volume_change_rate', 'volatility',\n",
|
||
" 'amount_change_rate', 'amount_mean'],\n",
|
||
" aggfunc='last'\n",
|
||
" )\n",
|
||
"\n",
|
||
" df_final.columns = [f\"{col[1]}_{col[0]}\" for col in df_final.columns]\n",
|
||
" df_final = df_final.reset_index()\n",
|
||
"\n",
|
||
" return df_final\n",
|
||
"\n",
|
||
"\n",
|
||
"# 使用函数\n",
|
||
"h5_filename = '../../data/index_data.h5'\n",
|
||
"index_data = generate_index_indicators(h5_filename)\n",
|
||
"index_data = index_data.dropna()\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 9,
|
||
"id": "a735bc02ceb4d872",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:47:10.821169Z",
|
||
"start_time": "2025-04-03T12:47:10.751831Z"
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"import talib\n",
|
||
"import numpy as np"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 10,
|
||
"id": "53f86ddc0677a6d7",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:47:15.944254Z",
|
||
"start_time": "2025-04-03T12:47:10.826179Z"
|
||
},
|
||
"jupyter": {
|
||
"source_hidden": true
|
||
},
|
||
"scrolled": true
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"from main.utils.factor import get_act_factor\n",
|
||
"\n",
|
||
"\n",
|
||
"def read_industry_data(h5_filename):\n",
|
||
" # 读取 H5 文件中所有的行业数据\n",
|
||
" industry_data = pd.read_hdf(h5_filename, key='sw_daily', columns=[\n",
|
||
" 'ts_code', 'trade_date', 'open', 'close', 'high', 'low', 'pe', 'pb', 'vol'\n",
|
||
" ]) # 假设 H5 文件的键是 'industry_data'\n",
|
||
" industry_data = industry_data.sort_values(by=['ts_code', 'trade_date'])\n",
|
||
" industry_data = industry_data.reindex()\n",
|
||
" industry_data['trade_date'] = pd.to_datetime(industry_data['trade_date'], format='%Y%m%d')\n",
|
||
"\n",
|
||
" grouped = industry_data.groupby('ts_code', group_keys=False)\n",
|
||
" industry_data['obv'] = grouped.apply(\n",
|
||
" lambda x: pd.Series(talib.OBV(x['close'].values, x['vol'].values), index=x.index)\n",
|
||
" )\n",
|
||
" industry_data['return_5'] = grouped['close'].apply(lambda x: x / x.shift(5) - 1)\n",
|
||
" industry_data['return_20'] = grouped['close'].apply(lambda x: x / x.shift(20) - 1)\n",
|
||
"\n",
|
||
" industry_data = get_act_factor(industry_data, cat=False)\n",
|
||
" industry_data = industry_data.sort_values(by=['trade_date', 'ts_code'])\n",
|
||
"\n",
|
||
" # # 计算每天每个 ts_code 的因子和当天所有 ts_code 的中位数的偏差\n",
|
||
" # factor_columns = ['obv', 'return_5', 'return_20', 'act_factor1', 'act_factor2', 'act_factor3', 'act_factor4'] # 因子列\n",
|
||
" # \n",
|
||
" # for factor in factor_columns:\n",
|
||
" # if factor in industry_data.columns:\n",
|
||
" # # 计算每天每个 ts_code 的因子值与当天所有 ts_code 的中位数的偏差\n",
|
||
" # industry_data[f'{factor}_deviation'] = industry_data.groupby('trade_date')[factor].transform(\n",
|
||
" # lambda x: x - x.mean())\n",
|
||
"\n",
|
||
" industry_data['return_5_percentile'] = industry_data.groupby('trade_date')['return_5'].transform(\n",
|
||
" lambda x: x.rank(pct=True))\n",
|
||
" industry_data['return_20_percentile'] = industry_data.groupby('trade_date')['return_20'].transform(\n",
|
||
" lambda x: x.rank(pct=True))\n",
|
||
"\n",
|
||
" # cs_rank_intraday_range(industry_data)\n",
|
||
" # cs_rank_close_pos_in_range(industry_data)\n",
|
||
"\n",
|
||
" industry_data = industry_data.drop(columns=['open', 'close', 'high', 'low', 'pe', 'pb', 'vol'])\n",
|
||
"\n",
|
||
" industry_data = industry_data.rename(\n",
|
||
" columns={col: f'industry_{col}' for col in industry_data.columns if col not in ['ts_code', 'trade_date']})\n",
|
||
"\n",
|
||
" industry_data = industry_data.rename(columns={'ts_code': 'cat_l2_code'})\n",
|
||
" return industry_data\n",
|
||
"\n",
|
||
"\n",
|
||
"industry_df = read_industry_data('../../data/sw_daily.h5')\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 11,
|
||
"id": "dbe2fd8021b9417f",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:47:15.969344Z",
|
||
"start_time": "2025-04-03T12:47:15.963327Z"
|
||
}
|
||
},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"['ts_code', 'open', 'close', 'high', 'low', 'amount', 'circ_mv', 'total_mv', 'is_st', 'up_limit', 'down_limit', 'buy_sm_vol', 'sell_sm_vol', 'buy_lg_vol', 'sell_lg_vol', 'buy_elg_vol', 'sell_elg_vol', 'net_mf_vol', 'his_low', 'his_high', 'cost_5pct', 'cost_15pct', 'cost_50pct', 'cost_85pct', 'cost_95pct', 'weight_avg', 'in_date']\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"origin_columns = df.columns.tolist()\n",
|
||
"origin_columns = [col for col in origin_columns if\n",
|
||
" col not in ['turnover_rate', 'pe_ttm', 'volume_ratio', 'vol', 'pct_chg', 'l2_code', 'winner_rate']]\n",
|
||
"origin_columns = [col for col in origin_columns if col not in index_data.columns]\n",
|
||
"origin_columns = [col for col in origin_columns if 'cyq' not in col]\n",
|
||
"print(origin_columns)"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 12,
|
||
"id": "85c3e3d0235ffffa",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T12:47:16.089879Z",
|
||
"start_time": "2025-04-03T12:47:15.990101Z"
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"fina_indicator_df = read_and_merge_h5_data('../../data/fina_indicator.h5', key='fina_indicator',\n",
|
||
" columns=['ts_code', 'ann_date', 'undist_profit_ps', 'ocfps', 'bps', 'roa', 'roe'],\n",
|
||
" df=None)\n",
|
||
"cashflow_df = read_and_merge_h5_data('../../data/cashflow.h5', key='cashflow',\n",
|
||
" columns=['ts_code', 'ann_date', 'n_cashflow_act'],\n",
|
||
" df=None)\n",
|
||
"balancesheet_df = read_and_merge_h5_data('../../data/balancesheet.h5', key='balancesheet',\n",
|
||
" columns=['ts_code', 'ann_date', 'money_cap', 'total_liab'],\n",
|
||
" df=None)\n",
|
||
"top_list_df = read_and_merge_h5_data('../../data/top_list.h5', key='top_list',\n",
|
||
" columns=['ts_code', 'trade_date', 'reason'],\n",
|
||
" df=None)\n",
|
||
"\n",
|
||
"top_list_df = top_list_df.sort_values(by='trade_date', ascending=False).drop_duplicates(subset=['ts_code', 'trade_date'], keep='first').sort_values(by='trade_date')\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 13,
|
||
"id": "92d84ce15a562ec6",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T13:08:01.612695Z",
|
||
"start_time": "2025-04-03T12:47:16.121802Z"
|
||
}
|
||
},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"计算因子 ts_turnover_rate_acceleration_5_20\n",
|
||
"计算因子 ts_vol_sustain_10_30\n",
|
||
"计算因子 cs_amount_outlier_10\n",
|
||
"计算因子 ts_ff_to_total_turnover_ratio\n",
|
||
"计算因子 ts_price_volume_trend_coherence_5_20\n",
|
||
"计算因子 ts_ff_turnover_rate_surge_10\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"警告: 从 financial_data_subset 中移除了 366 行，因为其 'ts_code' 或 'ann_date' 列存在空值。\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"警告: 从 financial_data_subset 中移除了 366 行，因为其 'ts_code' 或 'ann_date' 列存在空值。\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"警告: 从 financial_data_subset 中移除了 366 行，因为其 'ts_code' 或 'ann_date' 列存在空值。\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"警告: 从 financial_data_subset 中移除了 366 行，因为其 'ts_code' 或 'ann_date' 列存在空值。\n",
|
||
"开始计算因子: AR, BR (原地修改)...\n",
|
||
"因子 AR, BR 计算成功。\n",
|
||
"因子 AR, BR 计算流程结束。\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"使用 'ann_date' 作为财务数据生效日期。\n",
|
||
"警告: 从 financial_data_subset 中移除了 366 行，因为其 'ts_code' 或 'ann_date' 列存在空值。\n",
|
||
"计算 BBI...\n",
|
||
"--- 计算日级别偏离度 (使用 pct_chg) ---\n",
|
||
"--- 计算日级别动量基准 (使用 pct_chg) ---\n",
|
||
"日级别动量基准计算完成 (使用 pct_chg)。\n",
|
||
"日级别偏离度计算完成 (使用 pct_chg)。\n",
|
||
"--- 计算日级别行业偏离度 (使用 pct_chg 和行业基准) ---\n",
|
||
"--- 计算日级别行业动量基准 (使用 pct_chg 和 cat_l2_code) ---\n",
|
||
"错误: 计算日级别行业动量基准需要以下列: ['pct_chg', 'cat_l2_code', 'trade_date', 'ts_code']。\n",
|
||
"错误: 计算日级别行业偏离度需要以下列: ['pct_chg', 'daily_industry_positive_benchmark', 'daily_industry_negative_benchmark']。请先运行 daily_industry_momentum_benchmark(df)。\n",
|
||
"Index(['ts_code', 'trade_date', 'open', 'close', 'high', 'low', 'vol',\n",
|
||
" 'amount', 'pct_chg', 'turnover_rate', 'pe_ttm', 'circ_mv', 'total_mv',\n",
|
||
" 'volume_ratio', 'is_st', 'up_limit', 'down_limit', 'buy_sm_vol',\n",
|
||
" 'sell_sm_vol', 'buy_lg_vol', 'sell_lg_vol', 'buy_elg_vol',\n",
|
||
" 'sell_elg_vol', 'net_mf_vol', 'his_low', 'his_high', 'cost_5pct',\n",
|
||
" 'cost_15pct', 'cost_50pct', 'cost_85pct', 'cost_95pct', 'weight_avg',\n",
|
||
" 'winner_rate', 'l2_code', 'ts_turnover_rate_acceleration_5_20',\n",
|
||
" 'ts_vol_sustain_10_30', 'cs_amount_outlier_10',\n",
|
||
" 'ts_ff_to_total_turnover_ratio', 'ts_price_volume_trend_coherence_5_20',\n",
|
||
" 'ts_ff_turnover_rate_surge_10', 'undist_profit_ps', 'ocfps', 'roa',\n",
|
||
" 'roe', 'AR', 'BR', 'AR_BR', 'log_circ_mv', 'cashflow_to_ev_factor',\n",
|
||
" 'book_to_price_ratio', 'turnover_rate_mean_5', 'variance_20',\n",
|
||
" 'bbi_ratio_factor', 'daily_deviation', 'lg_elg_net_buy_vol',\n",
|
||
" 'flow_lg_elg_intensity', 'sm_net_buy_vol', 'flow_divergence_diff',\n",
|
||
" 'flow_divergence_ratio', 'total_buy_vol', 'lg_elg_buy_prop',\n",
|
||
" 'flow_struct_buy_change', 'lg_elg_net_buy_vol_change',\n",
|
||
" 'flow_lg_elg_accel', 'chip_concentration_range', 'chip_skewness',\n",
|
||
" 'floating_chip_proxy', 'cost_support_15pct_change',\n",
|
||
" 'cat_winner_price_zone', 'flow_chip_consistency',\n",
|
||
" 'profit_taking_vs_absorb', '_is_positive', '_is_negative',\n",
|
||
" 'cat_is_positive', '_pos_returns', '_neg_returns', '_pos_returns_sq',\n",
|
||
" '_neg_returns_sq', 'upside_vol', 'downside_vol', 'vol_ratio',\n",
|
||
" 'return_skew', 'return_kurtosis', 'volume_change_rate',\n",
|
||
" 'cat_volume_breakout', 'turnover_deviation', 'cat_turnover_spike',\n",
|
||
" 'avg_volume_ratio', 'cat_volume_ratio_breakout', 'vol_spike',\n",
|
||
" 'vol_std_5', 'atr_14', 'atr_6', 'obv'],\n",
|
||
" dtype='object')\n",
|
||
"Calculating lg_flow_mom_corr_20_60...\n",
|
||
"Finished lg_flow_mom_corr_20_60.\n",
|
||
"Calculating lg_flow_accel...\n",
|
||
"Finished lg_flow_accel.\n",
|
||
"Calculating profit_pressure...\n",
|
||
"Finished profit_pressure.\n",
|
||
"Calculating underwater_resistance...\n",
|
||
"Finished underwater_resistance.\n",
|
||
"Calculating cost_conc_std_20...\n",
|
||
"Finished cost_conc_std_20.\n",
|
||
"Calculating profit_decay_20...\n",
|
||
"Finished profit_decay_20.\n",
|
||
"Calculating vol_amp_loss_20...\n",
|
||
"Finished vol_amp_loss_20.\n",
|
||
"Calculating vol_drop_profit_cnt_5...\n",
|
||
"Finished vol_drop_profit_cnt_5.\n",
|
||
"Calculating lg_flow_vol_interact_20...\n",
|
||
"Finished lg_flow_vol_interact_20.\n",
|
||
"Calculating cost_break_confirm_cnt_5...\n",
|
||
"Finished cost_break_confirm_cnt_5.\n",
|
||
"Calculating atr_norm_channel_pos_14...\n",
|
||
"Finished atr_norm_channel_pos_14.\n",
|
||
"Calculating turnover_diff_skew_20...\n",
|
||
"Finished turnover_diff_skew_20.\n",
|
||
"Calculating lg_sm_flow_diverge_20...\n",
|
||
"Finished lg_sm_flow_diverge_20.\n",
|
||
"Calculating pullback_strong_20_20...\n",
|
||
"Finished pullback_strong_20_20.\n",
|
||
"Calculating vol_wgt_hist_pos_20...\n",
|
||
"Finished vol_wgt_hist_pos_20.\n",
|
||
"Calculating vol_adj_roc_20...\n",
|
||
"Finished vol_adj_roc_20.\n",
|
||
"Calculating cs_rank_net_lg_flow_val...\n",
|
||
"Finished cs_rank_net_lg_flow_val.\n",
|
||
"Calculating cs_rank_flow_divergence...\n",
|
||
"Finished cs_rank_flow_divergence.\n",
|
||
"Calculating cs_rank_ind_adj_lg_flow...\n",
|
||
"Finished cs_rank_ind_adj_lg_flow.\n",
|
||
"Calculating cs_rank_elg_buy_ratio...\n",
|
||
"Finished cs_rank_elg_buy_ratio.\n",
|
||
"Calculating cs_rank_rel_profit_margin...\n",
|
||
"Finished cs_rank_rel_profit_margin.\n",
|
||
"Calculating cs_rank_cost_breadth...\n",
|
||
"Finished cs_rank_cost_breadth.\n",
|
||
"Calculating cs_rank_dist_to_upper_cost...\n",
|
||
"Finished cs_rank_dist_to_upper_cost.\n",
|
||
"Calculating cs_rank_winner_rate...\n",
|
||
"Finished cs_rank_winner_rate.\n",
|
||
"Calculating cs_rank_intraday_range...\n",
|
||
"Finished cs_rank_intraday_range.\n",
|
||
"Calculating cs_rank_close_pos_in_range...\n",
|
||
"Finished cs_rank_close_pos_in_range.\n",
|
||
"Calculating cs_rank_opening_gap...\n",
|
||
"Error calculating cs_rank_opening_gap: Missing 'pre_close' column. Assigning NaN.\n",
|
||
"Calculating cs_rank_pos_in_hist_range...\n",
|
||
"Finished cs_rank_pos_in_hist_range.\n",
|
||
"Calculating cs_rank_vol_x_profit_margin...\n",
|
||
"Finished cs_rank_vol_x_profit_margin.\n",
|
||
"Calculating cs_rank_lg_flow_price_concordance...\n",
|
||
"Finished cs_rank_lg_flow_price_concordance.\n",
|
||
"Calculating cs_rank_turnover_per_winner...\n",
|
||
"Finished cs_rank_turnover_per_winner.\n",
|
||
"Calculating cs_rank_ind_cap_neutral_pe (Placeholder - requires statsmodels)...\n",
|
||
"Finished cs_rank_ind_cap_neutral_pe (Placeholder).\n",
|
||
"Calculating cs_rank_volume_ratio...\n",
|
||
"Finished cs_rank_volume_ratio.\n",
|
||
"Calculating cs_rank_elg_buy_sell_sm_ratio...\n",
|
||
"Finished cs_rank_elg_buy_sell_sm_ratio.\n",
|
||
"Calculating cs_rank_cost_dist_vol_ratio...\n",
|
||
"Finished cs_rank_cost_dist_vol_ratio.\n",
|
||
"Calculating cs_rank_size...\n",
|
||
"Finished cs_rank_size.\n",
|
||
"<class 'pandas.core.frame.DataFrame'>\n",
|
||
"RangeIndex: 4539678 entries, 0 to 4539677\n",
|
||
"Columns: 187 entries, ts_code to cs_rank_size\n",
|
||
"dtypes: bool(10), datetime64[ns](1), float64(171), int64(3), object(2)\n",
|
||
"memory usage: 6.0+ GB\n",
|
||
"None\n",
|
||
"['ts_code', 'trade_date', 'open', 'close', 'high', 'low', 'vol', 'amount', 'pct_chg', 'turnover_rate', 'pe_ttm', 'circ_mv', 'total_mv', 'volume_ratio', 'is_st', 'up_limit', 'down_limit', 'buy_sm_vol', 'sell_sm_vol', 'buy_lg_vol', 'sell_lg_vol', 'buy_elg_vol', 'sell_elg_vol', 'net_mf_vol', 'his_low', 'his_high', 'cost_5pct', 'cost_15pct', 'cost_50pct', 'cost_85pct', 'cost_95pct', 'weight_avg', 'winner_rate', 'cat_l2_code', 'ts_turnover_rate_acceleration_5_20', 'ts_vol_sustain_10_30', 'cs_amount_outlier_10', 'ts_ff_to_total_turnover_ratio', 'ts_price_volume_trend_coherence_5_20', 'ts_ff_turnover_rate_surge_10', 'undist_profit_ps', 'ocfps', 'roa', 'roe', 'AR', 'BR', 'AR_BR', 'log_circ_mv', 'cashflow_to_ev_factor', 'book_to_price_ratio', 'turnover_rate_mean_5', 'variance_20', 'bbi_ratio_factor', 'daily_deviation', 'lg_elg_net_buy_vol', 'flow_lg_elg_intensity', 'sm_net_buy_vol', 'flow_divergence_diff', 'flow_divergence_ratio', 'total_buy_vol', 'lg_elg_buy_prop', 'flow_struct_buy_change', 'lg_elg_net_buy_vol_change', 'flow_lg_elg_accel', 'chip_concentration_range', 'chip_skewness', 'floating_chip_proxy', 'cost_support_15pct_change', 'cat_winner_price_zone', 'flow_chip_consistency', 'profit_taking_vs_absorb', 'cat_is_positive', 'upside_vol', 'downside_vol', 'vol_ratio', 'return_skew', 'return_kurtosis', 'volume_change_rate', 'cat_volume_breakout', 'turnover_deviation', 'cat_turnover_spike', 'avg_volume_ratio', 'cat_volume_ratio_breakout', 'vol_spike', 'vol_std_5', 'atr_14', 'atr_6', 'obv', 'maobv_6', 'rsi_3', 'return_5', 'return_20', 'std_return_5', 'std_return_90', 'std_return_90_2', 'act_factor1', 'act_factor2', 'act_factor3', 'act_factor4', 'rank_act_factor1', 'rank_act_factor2', 'rank_act_factor3', 'cov', 'delta_cov', 'alpha_22_improved', 'alpha_003', 'alpha_007', 'alpha_013', 'vol_break', 'weight_roc5', 'price_cost_divergence', 'smallcap_concentration', 'cost_stability', 'high_cost_break_days', 'liquidity_risk', 'turnover_std', 'mv_volatility', 'volume_growth', 'mv_growth', 'momentum_factor', 'resonance_factor', 'log_close', 'cat_vol_spike', 'up', 'down', 'obv_maobv_6', 'std_return_5_over_std_return_90', 'std_return_90_minus_std_return_90_2', 'cat_af2', 'cat_af3', 'cat_af4', 'act_factor5', 'act_factor6', 'active_buy_volume_large', 'active_buy_volume_big', 'active_buy_volume_small', 'buy_lg_vol_minus_sell_lg_vol', 'buy_elg_vol_minus_sell_elg_vol', 'ctrl_strength', 'low_cost_dev', 'asymmetry', 'lock_factor', 'cat_vol_break', 'cost_atr_adj', 'cat_golden_resonance', 'mv_turnover_ratio', 'mv_adjusted_volume', 'mv_weighted_turnover', 'nonlinear_mv_volume', 'mv_volume_ratio', 'mv_momentum', 'lg_flow_mom_corr_20_60', 'lg_flow_accel', 'profit_pressure', 'underwater_resistance', 'cost_conc_std_20', 'profit_decay_20', 'vol_amp_loss_20', 'vol_drop_profit_cnt_5', 'lg_flow_vol_interact_20', 'cost_break_confirm_cnt_5', 'atr_norm_channel_pos_14', 'turnover_diff_skew_20', 'lg_sm_flow_diverge_20', 'pullback_strong_20_20', 'vol_wgt_hist_pos_20', 'vol_adj_roc_20', 'cs_rank_net_lg_flow_val', 'cs_rank_flow_divergence', 'cs_rank_ind_adj_lg_flow', 'cs_rank_elg_buy_ratio', 'cs_rank_rel_profit_margin', 'cs_rank_cost_breadth', 'cs_rank_dist_to_upper_cost', 'cs_rank_winner_rate', 'cs_rank_intraday_range', 'cs_rank_close_pos_in_range', 'cs_rank_opening_gap', 'cs_rank_pos_in_hist_range', 'cs_rank_vol_x_profit_margin', 'cs_rank_lg_flow_price_concordance', 'cs_rank_turnover_per_winner', 'cs_rank_ind_cap_neutral_pe', 'cs_rank_volume_ratio', 'cs_rank_elg_buy_sell_sm_ratio', 'cs_rank_cost_dist_vol_ratio', 'cs_rank_size']\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"\n",
|
||
"import numpy as np\n",
|
||
"from main.factor.factor import *\n",
|
||
"\n",
|
||
"def filter_data(df):\n",
|
||
" # df = df.groupby('trade_date').apply(lambda x: x.nlargest(1000, 'act_factor1'))\n",
|
||
" df = df[~df['is_st']]\n",
|
||
" df = df[~df['ts_code'].str.endswith('BJ')]\n",
|
||
" df = df[~df['ts_code'].str.startswith('30')]\n",
|
||
" df = df[~df['ts_code'].str.startswith('68')]\n",
|
||
" df = df[~df['ts_code'].str.startswith('8')]\n",
|
||
" df = df[df['trade_date'] >= '2019-01-01']\n",
|
||
" if 'in_date' in df.columns:\n",
|
||
" df = df.drop(columns=['in_date'])\n",
|
||
" df = df.reset_index(drop=True)\n",
|
||
" return df\n",
|
||
"\n",
|
||
"gc.collect()\n",
|
||
"\n",
|
||
"df = filter_data(df)\n",
|
||
"df = df.sort_values(by=['ts_code', 'trade_date'])\n",
|
||
"\n",
|
||
"# df = price_minus_deduction_price(df, n=120)\n",
|
||
"# df = price_deduction_price_diff_ratio_to_sma(df, n=120)\n",
|
||
"# df = cat_price_vs_sma_vs_deduction_price(df, n=120)\n",
|
||
"# df = cat_reason(df, top_list_df)\n",
|
||
"# df = cat_is_on_top_list(df, top_list_df)\n",
|
||
"\n",
|
||
"# df = cat_senti_mom_vol_spike(\n",
|
||
"# df,\n",
|
||
"# return_period=3,\n",
|
||
"# return_threshold=0.03, # 近3日涨幅超3%\n",
|
||
"# volume_ratio_threshold=1.3,\n",
|
||
"# current_pct_chg_min=0.0, # 当日必须收红\n",
|
||
"# current_pct_chg_max=0.05,\n",
|
||
"# ) # 当日涨幅不宜过大\n",
|
||
"\n",
|
||
"# df = cat_senti_pre_breakout(\n",
|
||
"# df,\n",
|
||
"# atr_short_N=10,\n",
|
||
"# atr_long_M=40,\n",
|
||
"# vol_atrophy_N=10,\n",
|
||
"# vol_atrophy_M=40,\n",
|
||
"# price_stab_N=5,\n",
|
||
"# price_stab_threshold=0.06,\n",
|
||
"# current_pct_chg_min_signal=0.002,\n",
|
||
"# current_pct_chg_max_signal=0.05,\n",
|
||
"# volume_ratio_signal_threshold=1.1,\n",
|
||
"# )\n",
|
||
"\n",
|
||
"df = ts_turnover_rate_acceleration_5_20(df)\n",
|
||
"df = ts_vol_sustain_10_30(df)\n",
|
||
"# df = cs_turnover_rate_relative_strength_20(df)\n",
|
||
"df = cs_amount_outlier_10(df)\n",
|
||
"df = ts_ff_to_total_turnover_ratio(df)\n",
|
||
"df = ts_price_volume_trend_coherence_5_20(df)\n",
|
||
"# df = ts_turnover_rate_trend_strength_5(df)\n",
|
||
"df = ts_ff_turnover_rate_surge_10(df)\n",
|
||
"\n",
|
||
"df = add_financial_factor(df, fina_indicator_df, factor_value_col='undist_profit_ps')\n",
|
||
"df = add_financial_factor(df, fina_indicator_df, factor_value_col='ocfps')\n",
|
||
"df = add_financial_factor(df, fina_indicator_df, factor_value_col='roa')\n",
|
||
"df = add_financial_factor(df, fina_indicator_df, factor_value_col='roe')\n",
|
||
"\n",
|
||
"calculate_arbr(df, N=26)\n",
|
||
"df['log_circ_mv'] = np.log(df['circ_mv'])\n",
|
||
"df = calculate_cashflow_to_ev_factor(df, cashflow_df, balancesheet_df)\n",
|
||
"df = caculate_book_to_price_ratio(df, fina_indicator_df)\n",
|
||
"\n",
|
||
"df = turnover_rate_n(df, n=5)\n",
|
||
"df = variance_n(df, n=20)\n",
|
||
"df = bbi_ratio_factor(df)\n",
|
||
"df = daily_deviation(df)\n",
|
||
"df = daily_industry_deviation(df)\n",
|
||
"df, _ = get_rolling_factor(df)\n",
|
||
"df, _ = get_simple_factor(df)\n",
|
||
"\n",
|
||
"df = df.rename(columns={'l1_code': 'cat_l1_code'})\n",
|
||
"df = df.rename(columns={'l2_code': 'cat_l2_code'})\n",
|
||
"\n",
|
||
"lg_flow_mom_corr(df, N=20, M=60)\n",
|
||
"lg_flow_accel(df)\n",
|
||
"profit_pressure(df)\n",
|
||
"underwater_resistance(df)\n",
|
||
"cost_conc_std(df, N=20)\n",
|
||
"profit_decay(df, N=20)\n",
|
||
"vol_amp_loss(df, N=20)\n",
|
||
"vol_drop_profit_cnt(df, N=20, M=5)\n",
|
||
"lg_flow_vol_interact(df, N=20)\n",
|
||
"cost_break_confirm_cnt(df, M=5)\n",
|
||
"atr_norm_channel_pos(df, N=14)\n",
|
||
"turnover_diff_skew(df, N=20)\n",
|
||
"lg_sm_flow_diverge(df, N=20)\n",
|
||
"pullback_strong(df, N=20, M=20)\n",
|
||
"vol_wgt_hist_pos(df, N=20)\n",
|
||
"vol_adj_roc(df, N=20)\n",
|
||
"\n",
|
||
"cs_rank_net_lg_flow_val(df)\n",
|
||
"cs_rank_flow_divergence(df)\n",
|
||
"cs_rank_industry_adj_lg_flow(df) # Needs cat_l2_code\n",
|
||
"cs_rank_elg_buy_ratio(df)\n",
|
||
"cs_rank_rel_profit_margin(df)\n",
|
||
"cs_rank_cost_breadth(df)\n",
|
||
"cs_rank_dist_to_upper_cost(df)\n",
|
||
"cs_rank_winner_rate(df)\n",
|
||
"cs_rank_intraday_range(df)\n",
|
||
"cs_rank_close_pos_in_range(df)\n",
|
||
"cs_rank_opening_gap(df) # Needs pre_close\n",
|
||
"cs_rank_pos_in_hist_range(df) # Needs his_low, his_high\n",
|
||
"cs_rank_vol_x_profit_margin(df)\n",
|
||
"cs_rank_lg_flow_price_concordance(df)\n",
|
||
"cs_rank_turnover_per_winner(df)\n",
|
||
"cs_rank_ind_cap_neutral_pe(df) # Placeholder - needs external libraries\n",
|
||
"cs_rank_volume_ratio(df) # Needs volume_ratio\n",
|
||
"cs_rank_elg_buy_sell_sm_ratio(df)\n",
|
||
"cs_rank_cost_dist_vol_ratio(df) # Needs volume_ratio\n",
|
||
"cs_rank_size(df) # Needs circ_mv\n",
|
||
"\n",
|
||
"\n",
|
||
"# df = df.merge(index_data, on='trade_date', how='left')\n",
|
||
"\n",
|
||
"print(df.info())\n",
|
||
"print(df.columns.tolist())"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 14,
|
||
"id": "b87b938028afa206",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T13:08:03.658725Z",
|
||
"start_time": "2025-04-03T13:08:02.469611Z"
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"from scipy.stats import ks_2samp, wasserstein_distance\n",
|
||
"\n",
|
||
"\n",
|
||
"def remove_shifted_features(train_data, test_data, feature_columns, ks_threshold=0.05, wasserstein_threshold=0.1,\n",
|
||
" importance_threshold=0.05):\n",
|
||
" dropped_features = []\n",
|
||
"\n",
|
||
" # **统计数据漂移**\n",
|
||
" numeric_columns = train_data.select_dtypes(include=['float64', 'int64']).columns\n",
|
||
" numeric_columns = [col for col in numeric_columns if col in feature_columns]\n",
|
||
" for feature in numeric_columns:\n",
|
||
" ks_stat, p_value = ks_2samp(train_data[feature], test_data[feature])\n",
|
||
" wasserstein_dist = wasserstein_distance(train_data[feature], test_data[feature])\n",
|
||
"\n",
|
||
" if p_value < ks_threshold or wasserstein_dist > wasserstein_threshold:\n",
|
||
" dropped_features.append(feature)\n",
|
||
"\n",
|
||
" print(f\"检测到 {len(dropped_features)} 个可能漂移的特征: {dropped_features}\")\n",
|
||
"\n",
|
||
" # **应用阈值进行最终筛选**\n",
|
||
" filtered_features = [f for f in feature_columns if f not in dropped_features]\n",
|
||
"\n",
|
||
" return filtered_features, dropped_features\n",
|
||
"\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 15,
|
||
"id": "f4f16d63ad18d1bc",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T13:08:03.670700Z",
|
||
"start_time": "2025-04-03T13:08:03.665739Z"
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"import numpy as np\n",
|
||
"import statsmodels.api as sm # 用于中性化回归\n",
|
||
"from tqdm import tqdm # 可选，用于显示进度条\n",
|
||
"\n",
|
||
"# --- 常量 ---\n",
|
||
"epsilon = 1e-10 # 防止除零\n",
|
||
"\n",
|
||
"# --- 1. 中位数去极值 (MAD) ---\n",
|
||
"\n",
|
||
"def cs_mad_filter(df: pd.DataFrame,\n",
|
||
" features: list,\n",
|
||
" k: float = 3.0,\n",
|
||
" scale_factor: float = 1.4826):\n",
|
||
" \"\"\"\n",
|
||
" 对指定特征列进行截面 MAD 去极值处理 (原地修改)。\n",
|
||
"\n",
|
||
" 方法: 对每日截面数据，计算 median 和 MAD，\n",
|
||
" 将超出 [median - k * scale * MAD, median + k * scale * MAD] 范围的值\n",
|
||
" 替换为边界值 (Winsorization)。\n",
|
||
" scale_factor=1.4826 使得 MAD 约等于正态分布的标准差。\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" df (pd.DataFrame): 输入 DataFrame，需包含 'trade_date' 和 features 列。\n",
|
||
" features (list): 需要处理的特征列名列表。\n",
|
||
" k (float): MAD 的倍数，用于确定边界。默认为 3.0。\n",
|
||
" scale_factor (float): MAD 的缩放因子。默认为 1.4826。\n",
|
||
"\n",
|
||
" WARNING: 此函数会原地修改输入的 DataFrame 'df'。\n",
|
||
" \"\"\"\n",
|
||
" print(f\"开始截面 MAD 去极值处理 (k={k})...\")\n",
|
||
" if not all(col in df.columns for col in features):\n",
|
||
" missing = [col for col in features if col not in df.columns]\n",
|
||
" print(f\"错误: DataFrame 中缺少以下特征列: {missing}。跳过去极值处理。\")\n",
|
||
" return\n",
|
||
"\n",
|
||
" grouped = df.groupby('trade_date')\n",
|
||
"\n",
|
||
" for col in tqdm(features, desc=\"MAD Filtering\"):\n",
|
||
" try:\n",
|
||
" # 计算截面中位数\n",
|
||
" median = grouped[col].transform('median')\n",
|
||
" # 计算截面 MAD (Median Absolute Deviation from Median)\n",
|
||
" mad = (df[col] - median).abs().groupby(df['trade_date']).transform('median')\n",
|
||
"\n",
|
||
" # 计算上下边界\n",
|
||
" lower_bound = median - k * scale_factor * mad\n",
|
||
" upper_bound = median + k * scale_factor * mad\n",
|
||
"\n",
|
||
" # 原地应用 clip\n",
|
||
" df[col] = np.clip(df[col], lower_bound, upper_bound)\n",
|
||
"\n",
|
||
" except KeyError:\n",
|
||
" print(f\"警告: 列 '{col}' 可能不存在或在分组中出错，跳过此列的 MAD 处理。\")\n",
|
||
" except Exception as e:\n",
|
||
" print(f\"警告: 处理列 '{col}' 时发生错误: {e}，跳过此列的 MAD 处理。\")\n",
|
||
"\n",
|
||
" print(\"截面 MAD 去极值处理完成。\")\n",
|
||
"\n",
|
||
"\n",
|
||
"# --- 2. 行业市值中性化 ---\n",
|
||
"\n",
|
||
"from tqdm import tqdm\n",
|
||
"\n",
|
||
"def cs_neutralize_market_cap_numpy(df: pd.DataFrame,\n",
|
||
" features: list,\n",
|
||
" market_cap_col: str = 'circ_mv'):\n",
|
||
" \"\"\"\n",
|
||
" 对 DataFrame 中的指定特征进行截面市值中性化 (NumPy 优化)。\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" df (pd.DataFrame): 包含数据的 DataFrame，需要有 'trade_date' 和 market_cap_col 列。\n",
|
||
" features (list): 需要进行市值中性化的特征列名列表。\n",
|
||
" market_cap_col (str): 包含市值数据的列名，默认为 'circ_mv'。\n",
|
||
" \"\"\"\n",
|
||
" print(\"开始截面市值中性化 (NumPy 优化)...\")\n",
|
||
" required_cols = features + ['trade_date', market_cap_col]\n",
|
||
" if not all(col in df.columns for col in required_cols):\n",
|
||
" missing = [col for col in required_cols if col not in df.columns]\n",
|
||
" print(f\"错误: DataFrame 中缺少必需列: {missing}。无法进行中性化。\")\n",
|
||
" return\n",
|
||
"\n",
|
||
" df_copy = df\n",
|
||
" log_cap_col = '_log_market_cap'\n",
|
||
" df_copy[log_cap_col] = np.log1p(df_copy[market_cap_col])\n",
|
||
"\n",
|
||
" # 创建一个 DataFrame 来存储所有日期的残差结果\n",
|
||
" residuals_container = pd.DataFrame(index=df_copy.index, columns=features, dtype=float)\n",
|
||
"\n",
|
||
" for date, group_df in tqdm(df_copy.groupby('trade_date'), desc=\"Neutralizing by Date (NumPy)\"):\n",
|
||
" # 准备 X 矩阵 (自变量)：常数项和对数市值\n",
|
||
" X_daily = np.concatenate([np.ones((len(group_df), 1)), group_df[[log_cap_col]].values], axis=1)\n",
|
||
"\n",
|
||
" for feature_col in features:\n",
|
||
" Y_daily = group_df[feature_col].values\n",
|
||
"\n",
|
||
" # 处理 NaN：只对有效数据对进行回归\n",
|
||
" valid_mask_y = ~np.isnan(Y_daily)\n",
|
||
" valid_mask_x = ~np.isnan(X_daily).any(axis=1)\n",
|
||
" valid_mask = valid_mask_y & valid_mask_x\n",
|
||
"\n",
|
||
" current_feature_indices = group_df.index[valid_mask]\n",
|
||
"\n",
|
||
" if np.sum(valid_mask) < X_daily.shape[1] + 1:\n",
|
||
" # 有效数据不足，此特征在此日期保持 NaN\n",
|
||
" continue\n",
|
||
"\n",
|
||
" Y_valid = Y_daily[valid_mask]\n",
|
||
" X_valid = X_daily[valid_mask, :]\n",
|
||
"\n",
|
||
" try:\n",
|
||
" # 使用 np.linalg.lstsq 进行 OLS 计算\n",
|
||
" beta, sum_sq_resid, rank, s = np.linalg.lstsq(X_valid, Y_valid, rcond=None)\n",
|
||
"\n",
|
||
" # 计算预测值 Y_hat = X_valid @ beta\n",
|
||
" Y_hat_valid = X_valid @ beta\n",
|
||
"\n",
|
||
" # 计算残差 residuals = Y_valid - Y_hat_valid\n",
|
||
" residuals_valid = Y_valid - Y_hat_valid\n",
|
||
"\n",
|
||
" # 将计算得到的残差放回 residuals_container\n",
|
||
" residuals_container.loc[current_feature_indices, feature_col] = residuals_valid\n",
|
||
"\n",
|
||
" except np.linalg.LinAlgError:\n",
|
||
" pass\n",
|
||
" except Exception as e:\n",
|
||
" pass\n",
|
||
"\n",
|
||
" # 将所有计算得到的残差更新回原始的 df (原地修改)\n",
|
||
" for feature_col in features:\n",
|
||
" df[feature_col] = residuals_container[feature_col]\n",
|
||
"\n",
|
||
" # 清理临时列\n",
|
||
" df.drop(columns=[log_cap_col], inplace=True, errors='ignore')\n",
|
||
" print(\"截面市值中性化完成 (NumPy 优化)。\")\n",
|
||
"\n",
|
||
"# --- 3. Z-Score 标准化 ---\n",
|
||
"\n",
|
||
"def cs_zscore_standardize(df: pd.DataFrame, features: list, epsilon: float = 1e-10):\n",
|
||
" \"\"\"\n",
|
||
" 对指定特征列进行截面 Z-Score 标准化 (原地修改)。\n",
|
||
" 方法: Z = (value - cross_sectional_mean) / (cross_sectional_std + epsilon)\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" df (pd.DataFrame): 输入 DataFrame，需包含 'trade_date' 和 features 列。\n",
|
||
" features (list): 需要处理的特征列名列表。\n",
|
||
" epsilon (float): 防止除以零的小常数。\n",
|
||
"\n",
|
||
" WARNING: 此函数会原地修改输入的 DataFrame 'df'。\n",
|
||
" \"\"\"\n",
|
||
" print(\"开始截面 Z-Score 标准化...\")\n",
|
||
" if not all(col in df.columns for col in features):\n",
|
||
" missing = [col for col in features if col not in df.columns]\n",
|
||
" print(f\"错误: DataFrame 中缺少以下特征列: {missing}。跳过标准化处理。\")\n",
|
||
" return\n",
|
||
"\n",
|
||
" grouped = df.groupby('trade_date')\n",
|
||
"\n",
|
||
" for col in tqdm(features, desc=\"Standardizing\"):\n",
|
||
" try:\n",
|
||
" # 使用 transform 计算截面均值和标准差\n",
|
||
" mean = grouped[col].transform('mean')\n",
|
||
" std = grouped[col].transform('std')\n",
|
||
"\n",
|
||
" # 计算 Z-Score 并原地赋值\n",
|
||
" df[col] = (df[col] - mean) / (std + epsilon)\n",
|
||
"\n",
|
||
" except KeyError:\n",
|
||
" print(f\"警告: 列 '{col}' 可能不存在或在分组中出错，跳过此列的标准化处理。\")\n",
|
||
" except Exception as e:\n",
|
||
" print(f\"警告: 处理列 '{col}' 时发生错误: {e}，跳过此列的标准化处理。\")\n",
|
||
"\n",
|
||
" print(\"截面 Z-Score 标准化完成。\")\n",
|
||
"\n",
|
||
"def fill_nan_with_daily_median(df: pd.DataFrame, feature_columns: list[str]) -> pd.DataFrame:\n",
|
||
" \"\"\"\n",
|
||
" 对指定特征列进行每日截面中位数填充缺失值 (NaN)。\n",
|
||
"\n",
|
||
" 参数:\n",
|
||
" df (pd.DataFrame): 包含多日数据的DataFrame，需要包含 'trade_date' 和 feature_columns 中的列。\n",
|
||
" feature_columns (list[str]): 需要进行缺失值填充的特征列名称列表。\n",
|
||
"\n",
|
||
" 返回:\n",
|
||
" pd.DataFrame: 包含缺失值填充后特征列的DataFrame。在输入DataFrame的副本上操作。\n",
|
||
" \"\"\"\n",
|
||
" processed_df = df.copy() # 在副本上操作，保留原始数据\n",
|
||
"\n",
|
||
" # 确保 trade_date 是 datetime 类型以便正确分组\n",
|
||
" processed_df['trade_date'] = pd.to_datetime(processed_df['trade_date'])\n",
|
||
"\n",
|
||
" def _fill_daily_nan(group):\n",
|
||
" # group 是某一个交易日的 DataFrame\n",
|
||
"\n",
|
||
" # 遍历指定的特征列\n",
|
||
" for feature_col in feature_columns:\n",
|
||
" # 检查列是否存在于当前分组中\n",
|
||
" if feature_col in group.columns:\n",
|
||
" # 计算当日该特征的中位数\n",
|
||
" median_val = group[feature_col].median()\n",
|
||
"\n",
|
||
" # 使用当日中位数填充该特征列的 NaN 值\n",
|
||
" # inplace=True 会直接修改 group DataFrame\n",
|
||
" group[feature_col].fillna(median_val, inplace=True)\n",
|
||
" # else:\n",
|
||
" # print(f\"Warning: Feature column '{feature_col}' not found in daily group for {group['trade_date'].iloc[0]}. Skipping.\")\n",
|
||
"\n",
|
||
" return group\n",
|
||
"\n",
|
||
" # 按交易日期分组，并应用每日填充函数\n",
|
||
" # group_keys=False 避免将分组键添加到结果索引中\n",
|
||
" filled_df = processed_df.groupby('trade_date', group_keys=False).apply(_fill_daily_nan)\n",
|
||
"\n",
|
||
" return filled_df"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 16,
|
||
"id": "40e6b68a91b30c79",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T13:08:04.694262Z",
|
||
"start_time": "2025-04-03T13:08:03.694904Z"
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"def remove_outliers_label_percentile(label: pd.Series, lower_percentile: float = 0.01, upper_percentile: float = 0.99,\n",
|
||
" log=True):\n",
|
||
" if not (0 <= lower_percentile < upper_percentile <= 1):\n",
|
||
" raise ValueError(\"Percentile values must satisfy 0 <= lower_percentile < upper_percentile <= 1.\")\n",
|
||
"\n",
|
||
" # Calculate lower and upper bounds based on percentiles\n",
|
||
" lower_bound = label.quantile(lower_percentile)\n",
|
||
" upper_bound = label.quantile(upper_percentile)\n",
|
||
"\n",
|
||
" # Filter out values outside the bounds\n",
|
||
" filtered_label = label[(label >= lower_bound) & (label <= upper_bound)]\n",
|
||
"\n",
|
||
" # Print the number of removed outliers\n",
|
||
" if log:\n",
|
||
" print(f\"Removed {len(label) - len(filtered_label)} outliers.\")\n",
|
||
" return filtered_label\n",
|
||
"\n",
|
||
"\n",
|
||
"def calculate_risk_adjusted_target(df, days=5):\n",
|
||
" df = df.sort_values(by=['ts_code', 'trade_date'])\n",
|
||
"\n",
|
||
" df['future_close'] = df.groupby('ts_code')['close'].shift(-days)\n",
|
||
" df['future_open'] = df.groupby('ts_code')['open'].shift(-1)\n",
|
||
" df['future_return'] = (df['future_close'] - df['future_open']) / df['future_open']\n",
|
||
"\n",
|
||
" df['future_volatility'] = df.groupby('ts_code')['future_return'].rolling(days, min_periods=1).std().reset_index(\n",
|
||
" level=0, drop=True)\n",
|
||
" sharpe_ratio = df['future_return'] * df['future_volatility']\n",
|
||
" sharpe_ratio.replace([np.inf, -np.inf], np.nan, inplace=True)\n",
|
||
"\n",
|
||
" return sharpe_ratio\n",
|
||
"\n",
|
||
"\n",
|
||
"def calculate_score(df, days=5, lambda_param=1.0):\n",
|
||
" def calculate_max_drawdown(prices):\n",
|
||
" peak = prices.iloc[0] # 初始化峰值\n",
|
||
" max_drawdown = 0 # 初始化最大回撤\n",
|
||
"\n",
|
||
" for price in prices:\n",
|
||
" if price > peak:\n",
|
||
" peak = price # 更新峰值\n",
|
||
" else:\n",
|
||
" drawdown = (peak - price) / peak # 计算当前回撤\n",
|
||
" max_drawdown = max(max_drawdown, drawdown) # 更新最大回撤\n",
|
||
"\n",
|
||
" return max_drawdown\n",
|
||
"\n",
|
||
" def compute_stock_score(stock_df):\n",
|
||
" stock_df = stock_df.sort_values(by=['trade_date'])\n",
|
||
" future_return = stock_df['future_return']\n",
|
||
" # 使用已有的 pct_chg 字段计算波动率\n",
|
||
" volatility = stock_df['pct_chg'].rolling(days).std().shift(-days)\n",
|
||
" max_drawdown = stock_df['close'].rolling(days).apply(calculate_max_drawdown, raw=False).shift(-days)\n",
|
||
" score = future_return - lambda_param * max_drawdown\n",
|
||
" return score\n",
|
||
"\n",
|
||
" # # 确保 DataFrame 按照股票代码和交易日期排序\n",
|
||
" # df = df.sort_values(by=['ts_code', 'trade_date'])\n",
|
||
"\n",
|
||
" # 对每个股票分别计算 score\n",
|
||
" df['score'] = df.groupby('ts_code').apply(compute_stock_score).reset_index(level=0, drop=True)\n",
|
||
"\n",
|
||
" return df['score']\n",
|
||
"\n",
|
||
"\n",
|
||
"def remove_highly_correlated_features(df, feature_columns, threshold=0.9):\n",
|
||
" numeric_features = df[feature_columns].select_dtypes(include=[np.number]).columns.tolist()\n",
|
||
" if not numeric_features:\n",
|
||
" raise ValueError(\"No numeric features found in the provided data.\")\n",
|
||
"\n",
|
||
" corr_matrix = df[numeric_features].corr().abs()\n",
|
||
" upper = corr_matrix.where(np.triu(np.ones(corr_matrix.shape), k=1).astype(bool))\n",
|
||
" to_drop = [column for column in upper.columns if any(upper[column] > threshold)]\n",
|
||
" remaining_features = [col for col in feature_columns if col not in to_drop\n",
|
||
" or 'act' in col or 'af' in col]\n",
|
||
" return remaining_features\n",
|
||
"\n",
|
||
"\n",
|
||
"def cross_sectional_standardization(df, features):\n",
|
||
" df_sorted = df.sort_values(by='trade_date') # 按时间排序\n",
|
||
" df_standardized = df_sorted.copy()\n",
|
||
"\n",
|
||
" for date in df_sorted['trade_date'].unique():\n",
|
||
" # 获取当前时间点的数据\n",
|
||
" current_data = df_standardized[df_standardized['trade_date'] == date]\n",
|
||
"\n",
|
||
" # 只对指定特征进行标准化\n",
|
||
" scaler = StandardScaler()\n",
|
||
" standardized_values = scaler.fit_transform(current_data[features])\n",
|
||
"\n",
|
||
" # 将标准化结果重新赋值回去\n",
|
||
" df_standardized.loc[df_standardized['trade_date'] == date, features] = standardized_values\n",
|
||
"\n",
|
||
" return df_standardized\n",
|
||
"\n",
|
||
"\n",
|
||
"def neutralize_manual_revised(df: pd.DataFrame, features: list, industry_col: str, mkt_cap_col: str) -> pd.DataFrame:\n",
|
||
" \"\"\"\n",
|
||
" 手动实现简单回归以提升速度，通过构建 Series 确保索引对齐。\n",
|
||
" 对特征在行业内部进行市值中性化。\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" df: 输入的 DataFrame，包含特征、行业分类和市值列。\n",
|
||
" features: 需要进行中性化的特征列名列表。\n",
|
||
" industry_col: 行业分类列的列名。\n",
|
||
" mkt_cap_col: 市值列的列名。\n",
|
||
"\n",
|
||
" Returns:\n",
|
||
" 中性化后的 DataFrame。\n",
|
||
" \"\"\"\n",
|
||
"\n",
|
||
" df[mkt_cap_col] = pd.to_numeric(df[mkt_cap_col], errors='coerce')\n",
|
||
" df_cleaned = df.dropna(subset=[mkt_cap_col]).copy()\n",
|
||
" df_cleaned = df_cleaned[df_cleaned[mkt_cap_col] > 0].copy()\n",
|
||
"\n",
|
||
" if df_cleaned.empty:\n",
|
||
" print(\"警告: 清理市值异常值后 DataFrame 为空。\")\n",
|
||
" return df # 返回原始或空df，取决于清理前的状态\n",
|
||
"\n",
|
||
" processed_df = df\n",
|
||
"\n",
|
||
" for col in features:\n",
|
||
" if col not in df_cleaned.columns:\n",
|
||
" print(f\"警告: 特征列 '{col}' 不存在于清理后的 DataFrame 中，已跳过。\")\n",
|
||
" # 对于原始 df 中该列不存在的，在结果 df 中也保持原样（可能全是NaN）\n",
|
||
" processed_df[col] = df[col] if col in df.columns else np.nan\n",
|
||
" continue\n",
|
||
"\n",
|
||
" # 跳过对控制变量本身进行中性化\n",
|
||
" if col == mkt_cap_col or col == industry_col:\n",
|
||
" print(f\"警告: 特征列 '{col}' 是控制变量或内部使用的列，跳过中性化。\")\n",
|
||
" # 在结果 df 中也保持原样\n",
|
||
" processed_df[col] = df[col] if col in df.columns else np.nan\n",
|
||
" continue\n",
|
||
"\n",
|
||
" residual_series = pd.Series(index=df_cleaned.index, dtype=float)\n",
|
||
"\n",
|
||
" # 在分组前处理特征列的 NaN，只对有因子值的行进行回归计算\n",
|
||
" df_subset_factor = df_cleaned.dropna(subset=[col]).copy()\n",
|
||
"\n",
|
||
" if not df_subset_factor.empty:\n",
|
||
" for industry, group in df_subset_factor.groupby(industry_col):\n",
|
||
" x = group[mkt_cap_col] # 市值对数\n",
|
||
" y = group[col] # 因子值\n",
|
||
"\n",
|
||
" # 确保有足够的数据点 (>1) 且市值对数有方差 (>0) 进行回归计算\n",
|
||
" # 检查 np.var > 一个很小的正数，避免浮点数误差导致的零方差判断问题\n",
|
||
" if len(group) > 1 and np.var(x) > 1e-9:\n",
|
||
" try:\n",
|
||
" beta = np.cov(y, x)[0, 1] / np.var(x)\n",
|
||
" alpha = np.mean(y) - beta * np.mean(x)\n",
|
||
"\n",
|
||
" # 计算残差\n",
|
||
" resid = y - (alpha + beta * x)\n",
|
||
"\n",
|
||
" # 将计算出的残差存储到 residual_series 中，通过索引自动对齐\n",
|
||
" residual_series.loc[resid.index] = resid\n",
|
||
"\n",
|
||
" except Exception as e:\n",
|
||
" # 捕获可能的计算异常，例如np.cov或np.var因为极端数据报错\n",
|
||
" print(f\"警告: 在行业 {industry} 计算回归时发生错误: {e}。该组残差将设为原始值或 NaN。\")\n",
|
||
" # 此时该组的残差会保持 residual_series 初始化时的 NaN 或后续处理\n",
|
||
" # 也可以选择保留原始值：residual_series.loc[group.index] = group[col]\n",
|
||
"\n",
|
||
" else:\n",
|
||
" residual_series.loc[group.index] = group[col] # 保留原始因子值\n",
|
||
" processed_df.loc[residual_series.index, col] = residual_series\n",
|
||
"\n",
|
||
"\n",
|
||
" else:\n",
|
||
" processed_df[col] = np.nan # 或 df[col] if col in df.columns else np.nan\n",
|
||
"\n",
|
||
" return processed_df\n",
|
||
"\n",
|
||
"\n",
|
||
"import gc\n",
|
||
"\n",
|
||
"gc.collect()\n",
|
||
"\n",
|
||
"\n",
|
||
"def mad_filter(df, features, n=3):\n",
|
||
" for col in features:\n",
|
||
" median = df[col].median()\n",
|
||
" mad = np.median(np.abs(df[col] - median))\n",
|
||
" upper = median + n * mad\n",
|
||
" lower = median - n * mad\n",
|
||
" df[col] = np.clip(df[col], lower, upper) # 截断极值\n",
|
||
" return df\n",
|
||
"\n",
|
||
"\n",
|
||
"def percentile_filter(df, features, lower_percentile=0.01, upper_percentile=0.99):\n",
|
||
" for col in features:\n",
|
||
" # 按日期分组计算上下百分位数\n",
|
||
" lower_bound = df.groupby('trade_date')[col].transform(\n",
|
||
" lambda x: x.quantile(lower_percentile)\n",
|
||
" )\n",
|
||
" upper_bound = df.groupby('trade_date')[col].transform(\n",
|
||
" lambda x: x.quantile(upper_percentile)\n",
|
||
" )\n",
|
||
" # 截断超出范围的值\n",
|
||
" df[col] = np.clip(df[col], lower_bound, upper_bound)\n",
|
||
" return df\n",
|
||
"\n",
|
||
"\n",
|
||
"from scipy.stats import iqr\n",
|
||
"\n",
|
||
"\n",
|
||
"def iqr_filter(df, features):\n",
|
||
" for col in features:\n",
|
||
" df[col] = df.groupby('trade_date')[col].transform(\n",
|
||
" lambda x: (x - x.median()) / iqr(x) if iqr(x) != 0 else x\n",
|
||
" )\n",
|
||
" return df\n",
|
||
"\n",
|
||
"\n",
|
||
"def quantile_filter(df, features, lower_quantile=0.01, upper_quantile=0.99, window=60):\n",
|
||
" df = df.copy()\n",
|
||
" for col in features:\n",
|
||
" # 计算 rolling 统计量，需要按日期进行 groupby\n",
|
||
" rolling_lower = df.groupby('trade_date')[col].transform(lambda x: x.rolling(window=min(len(x), window)).quantile(lower_quantile))\n",
|
||
" rolling_upper = df.groupby('trade_date')[col].transform(lambda x: x.rolling(window=min(len(x), window)).quantile(upper_quantile))\n",
|
||
"\n",
|
||
" # 对数据进行裁剪\n",
|
||
" df[col] = np.clip(df[col], rolling_lower, rolling_upper)\n",
|
||
" \n",
|
||
" return df\n",
|
||
"\n",
|
||
"def select_top_features_by_rankic(df: pd.DataFrame, feature_columns: list, n: int, target_column: str = 'future_return') -> list:\n",
|
||
" \"\"\"\n",
|
||
" 计算给定特征与目标列的 RankIC，并返回 RankIC 绝对值最高的 n 个特征。\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" df: 包含特征列和目标列的 Pandas DataFrame。\n",
|
||
" feature_columns: 包含所有待评估特征列名的列表。\n",
|
||
" n: 希望选取的 RankIC 绝对值最高的特征数量。\n",
|
||
" target_column: 目标列的名称，用于计算 RankIC。默认为 'future_return'。\n",
|
||
"\n",
|
||
" Returns:\n",
|
||
" 包含 RankIC 绝对值最高的 n 个特征列名的列表。\n",
|
||
" \"\"\"\n",
|
||
" numeric_columns = df.select_dtypes(include=['float64', 'int64']).columns\n",
|
||
" numeric_columns = [col for col in numeric_columns if col in feature_columns]\n",
|
||
" if target_column not in df.columns:\n",
|
||
" raise ValueError(f\"目标列 '{target_column}' 不存在于 DataFrame 中。\")\n",
|
||
"\n",
|
||
" rankic_scores = {}\n",
|
||
" for feature in numeric_columns:\n",
|
||
" if feature not in df.columns:\n",
|
||
" print(f\"警告: 特征列 '{feature}' 不存在于 DataFrame 中，已跳过。\")\n",
|
||
" continue\n",
|
||
"\n",
|
||
" # 计算特征与目标列的 RankIC (斯皮尔曼相关系数)\n",
|
||
" # dropna() 是为了处理缺失值，确保相关性计算不失败\n",
|
||
" valid_data = df[[feature, target_column]].dropna()\n",
|
||
" if len(valid_data) > 1: # 确保有足够的数据点进行相关性计算\n",
|
||
" # 计算斯皮尔曼相关性\n",
|
||
" correlation = valid_data[feature].corr(valid_data[target_column], method='spearman')\n",
|
||
" rankic_scores[feature] = abs(correlation) # 使用绝对值来衡量相关性强度\n",
|
||
" else:\n",
|
||
" rankic_scores[feature] = 0 # 数据不足，RankIC设为0或跳过\n",
|
||
"\n",
|
||
" # 将 RankIC 分数转换为 Series 便于排序\n",
|
||
" rankic_series = pd.Series(rankic_scores)\n",
|
||
"\n",
|
||
" # 按 RankIC 绝对值降序排序，选取前 n 个特征\n",
|
||
" # handle case where n might be larger than available features\n",
|
||
" n_actual = min(n, len(rankic_series))\n",
|
||
" top_features = rankic_series.sort_values(ascending=False).head(n_actual).index.tolist()\n",
|
||
" top_features = [col for col in feature_columns if col in top_features or col not in numeric_columns]\n",
|
||
" return top_features\n",
|
||
"\n",
|
||
"def create_deviation_within_dates(df, feature_columns):\n",
|
||
" groupby_col = 'cat_l2_code' # 使用 trade_date 进行分组\n",
|
||
" new_columns = {}\n",
|
||
" ret_feature_columns = feature_columns[:]\n",
|
||
"\n",
|
||
" # 自动选择所有数值型特征\n",
|
||
" num_features = [col for col in feature_columns if 'cat' not in col and 'index' not in col]\n",
|
||
"\n",
|
||
" # num_features = ['vol', 'pct_chg', 'turnover_rate', 'volume_ratio', 'cat_vol_spike', 'obv', 'maobv_6', 'return_5', 'return_10', 'return_20', 'std_return_5', 'std_return_15', 'std_return_90', 'std_return_90_2', 'act_factor1', 'act_factor2', 'act_factor3', 'act_factor4', 'act_factor5', 'act_factor6', 'rank_act_factor1', 'rank_act_factor2', 'rank_act_factor3', 'active_buy_volume_large', 'active_buy_volume_big', 'active_buy_volume_small', 'alpha_022', 'alpha_003', 'alpha_007', 'alpha_013']\n",
|
||
" num_features = [col for col in num_features if 'cat' not in col and 'industry' not in col]\n",
|
||
" num_features = [col for col in num_features if 'limit' not in col]\n",
|
||
" num_features = [col for col in num_features if 'cyq' not in col]\n",
|
||
"\n",
|
||
" # 遍历所有数值型特征\n",
|
||
" for feature in num_features:\n",
|
||
" if feature == 'trade_date': # 不需要对 'trade_date' 计算偏差\n",
|
||
" continue\n",
|
||
"\n",
|
||
" # grouped_mean = df.groupby(['trade_date'])[feature].transform('mean')\n",
|
||
" # deviation_col_name = f'deviation_mean_{feature}'\n",
|
||
" # new_columns[deviation_col_name] = df[feature] - grouped_mean\n",
|
||
" # ret_feature_columns.append(deviation_col_name)\n",
|
||
"\n",
|
||
" grouped_mean = df.groupby(['trade_date', groupby_col])[feature].transform('mean')\n",
|
||
" deviation_col_name = f'deviation_mean_{feature}'\n",
|
||
" new_columns[deviation_col_name] = df[feature] - grouped_mean\n",
|
||
" ret_feature_columns.append(deviation_col_name)\n",
|
||
"\n",
|
||
" # 将新计算的偏差特征与原始 DataFrame 合并\n",
|
||
" df = pd.concat([df, pd.DataFrame(new_columns)], axis=1)\n",
|
||
"\n",
|
||
" # for feature in ['obv', 'return_20', 'act_factor1', 'act_factor2', 'act_factor3', 'act_factor4']:\n",
|
||
" # df[f'deviation_industry_{feature}'] = df[feature] - df[f'industry_{feature}']\n",
|
||
"\n",
|
||
" return df, ret_feature_columns\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 17,
|
||
"id": "47c12bb34062ae7a",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T14:57:50.841165Z",
|
||
"start_time": "2025-04-03T14:49:25.889057Z"
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"days = 5\n",
|
||
"validation_days = 120\n",
|
||
"\n",
|
||
"import gc\n",
|
||
"\n",
|
||
"gc.collect()\n",
|
||
"\n",
|
||
"df = df.sort_values(by=['ts_code', 'trade_date'])\n",
|
||
"df['future_return'] = df.groupby('ts_code', group_keys=False)['close'].apply(lambda x: x.shift(-days) / x - 1)\n",
|
||
"# df['future_return'] = (df.groupby('ts_code')['close'].shift(-days) - df.groupby('ts_code')['open'].shift(-1)) / \\\n",
|
||
"# df.groupby('ts_code')['open'].shift(-1)\n",
|
||
"\n",
|
||
"df['cat_up_limit'] = df['pct_chg'] > 5\n",
|
||
"df['label'] = df.groupby('ts_code')['cat_up_limit'].rolling(window=5, min_periods=1).max().groupby('ts_code').shift(-5).fillna(0).astype(int).reset_index(level=0, drop=True)\n",
|
||
"\n",
|
||
"filter_index = df['future_return'].between(df['future_return'].quantile(0.01), df['future_return'].quantile(0.99))\n",
|
||
"\n",
|
||
"# for col in [col for col in df.columns]:\n",
|
||
"# train_data[col] = train_data[col].astype('str')\n",
|
||
"# test_data[col] = test_data[col].astype('str')"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 18,
|
||
"id": "29221dde",
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"197\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"feature_columns = [col for col in df.head(10).merge(industry_df, on=['cat_l2_code', 'trade_date'], how='left').merge(index_data, on='trade_date', how='left').columns]\n",
|
||
"feature_columns = [col for col in feature_columns if col not in ['trade_date',\n",
|
||
" 'ts_code',\n",
|
||
" 'label']]\n",
|
||
"feature_columns = [col for col in feature_columns if 'future' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if 'label' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if 'score' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if 'gen' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if 'is_st' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if 'pe_ttm' not in col]\n",
|
||
"# feature_columns = [col for col in feature_columns if 'volatility' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if 'circ_mv' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if 'code' not in col]\n",
|
||
"feature_columns = [col for col in feature_columns if col not in origin_columns]\n",
|
||
"feature_columns = [col for col in feature_columns if not col.startswith('_')]\n",
|
||
"# feature_columns = [col for col in feature_columns if col not in ['ts_code', 'trade_date', 'vol_std_5', 'cov', 'delta_cov', 'alpha_22_improved', 'alpha_007', 'consecutive_up_limit', 'mv_volatility', 'volume_growth', 'mv_growth', 'arbr']]\n",
|
||
"feature_columns = [col for col in feature_columns if col not in ['intraday_lg_flow_corr_20', \n",
|
||
" 'cap_neutral_cost_metric', \n",
|
||
" 'hurst_net_mf_vol_60', \n",
|
||
" 'complex_factor_deap_1', \n",
|
||
" 'lg_buy_consolidation_20',\n",
|
||
" 'cs_rank_ind_cap_neutral_pe',\n",
|
||
" 'cs_rank_opening_gap',\n",
|
||
" 'cs_rank_ind_adj_lg_flow']]\n",
|
||
"feature_columns = [col for col in feature_columns if col not in ['roa', 'roe']]\n",
|
||
"print(len(feature_columns))"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 19,
|
||
"id": "03ee5daf",
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"# df = fill_nan_with_daily_median(df, feature_columns)\n",
|
||
"for feature_col in [col for col in feature_columns if col in df.columns]:\n",
|
||
" # median_val = df[feature_col].median()\n",
|
||
" df[feature_col].fillna(0, inplace=True)"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 20,
|
||
"id": "b76ea08a",
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
" ts_code trade_date log_circ_mv\n",
|
||
"0 000001.SZ 2019-01-02 16.574219\n",
|
||
"1 000001.SZ 2019-01-03 16.583965\n",
|
||
"2 000001.SZ 2019-01-04 16.633371\n",
|
||
"['vol', 'pct_chg', 'turnover_rate', 'volume_ratio', 'winner_rate', 'ts_turnover_rate_acceleration_5_20', 'ts_vol_sustain_10_30', 'cs_amount_outlier_10', 'ts_ff_to_total_turnover_ratio', 'ts_price_volume_trend_coherence_5_20', 'ts_ff_turnover_rate_surge_10', 'undist_profit_ps', 'ocfps', 'AR', 'BR', 'AR_BR', 'cashflow_to_ev_factor', 'book_to_price_ratio', 'turnover_rate_mean_5', 'variance_20', 'bbi_ratio_factor', 'daily_deviation', 'lg_elg_net_buy_vol', 'flow_lg_elg_intensity', 'sm_net_buy_vol', 'total_buy_vol', 'lg_elg_buy_prop', 'flow_struct_buy_change', 'lg_elg_net_buy_vol_change', 'flow_lg_elg_accel', 'chip_concentration_range', 'chip_skewness', 'floating_chip_proxy', 'cost_support_15pct_change', 'cat_winner_price_zone', 'flow_chip_consistency', 'profit_taking_vs_absorb', 'cat_is_positive', 'upside_vol', 'downside_vol', 'vol_ratio', 'return_skew', 'return_kurtosis', 'volume_change_rate', 'cat_volume_breakout', 'turnover_deviation', 'cat_turnover_spike', 'avg_volume_ratio', 'cat_volume_ratio_breakout', 'vol_spike', 'vol_std_5', 'atr_14', 'atr_6', 'obv', 'maobv_6', 'rsi_3', 'return_5', 'return_20', 'std_return_5', 'std_return_90', 'std_return_90_2', 'act_factor1', 'act_factor2', 'act_factor3', 'act_factor4', 'rank_act_factor1', 'rank_act_factor2', 'rank_act_factor3', 'cov', 'delta_cov', 'alpha_22_improved', 'alpha_003', 'alpha_007', 'alpha_013', 'vol_break', 'weight_roc5', 'smallcap_concentration', 'cost_stability', 'high_cost_break_days', 'liquidity_risk', 'turnover_std', 'mv_volatility', 'volume_growth', 'mv_growth', 'momentum_factor', 'resonance_factor', 'log_close', 'cat_vol_spike', 'up', 'down', 'obv_maobv_6', 'std_return_5_over_std_return_90', 'std_return_90_minus_std_return_90_2', 'cat_af2', 'cat_af3', 'cat_af4', 'act_factor5', 'act_factor6', 'active_buy_volume_large', 'active_buy_volume_big', 'active_buy_volume_small', 'buy_lg_vol_minus_sell_lg_vol', 'buy_elg_vol_minus_sell_elg_vol', 'ctrl_strength', 'low_cost_dev', 'asymmetry', 'lock_factor', 'cat_vol_break', 'cost_atr_adj', 'cat_golden_resonance', 'mv_turnover_ratio', 'mv_adjusted_volume', 'mv_weighted_turnover', 'nonlinear_mv_volume', 'mv_volume_ratio', 'mv_momentum', 'lg_flow_mom_corr_20_60', 'lg_flow_accel', 'profit_pressure', 'underwater_resistance', 'cost_conc_std_20', 'profit_decay_20', 'vol_amp_loss_20', 'vol_drop_profit_cnt_5', 'lg_flow_vol_interact_20', 'cost_break_confirm_cnt_5', 'atr_norm_channel_pos_14', 'turnover_diff_skew_20', 'lg_sm_flow_diverge_20', 'pullback_strong_20_20', 'vol_wgt_hist_pos_20', 'vol_adj_roc_20', 'cs_rank_net_lg_flow_val', 'cs_rank_elg_buy_ratio', 'cs_rank_rel_profit_margin', 'cs_rank_cost_breadth', 'cs_rank_dist_to_upper_cost', 'cs_rank_winner_rate', 'cs_rank_intraday_range', 'cs_rank_close_pos_in_range', 'cs_rank_pos_in_hist_range', 'cs_rank_vol_x_profit_margin', 'cs_rank_lg_flow_price_concordance', 'cs_rank_turnover_per_winner', 'cs_rank_volume_ratio', 'cs_rank_elg_buy_sell_sm_ratio', 'cs_rank_cost_dist_vol_ratio', 'cs_rank_size', 'cat_up_limit', 'industry_obv', 'industry_return_5', 'industry_return_20', 'industry__ema_5', 'industry__ema_13', 'industry__ema_20', 'industry__ema_60', 'industry_act_factor1', 'industry_act_factor2', 'industry_act_factor3', 'industry_act_factor4', 'industry_act_factor5', 'industry_act_factor6', 'industry_rank_act_factor1', 'industry_rank_act_factor2', 'industry_rank_act_factor3', 'industry_return_5_percentile', 'industry_return_20_percentile', '000852.SH_MACD', '000905.SH_MACD', '399006.SZ_MACD', '000852.SH_MACD_hist', '000905.SH_MACD_hist', '399006.SZ_MACD_hist', '000852.SH_RSI', '000905.SH_RSI', '399006.SZ_RSI', '000852.SH_Signal_line', '000905.SH_Signal_line', '399006.SZ_Signal_line', '000852.SH_amount_change_rate', '000905.SH_amount_change_rate', '399006.SZ_amount_change_rate', '000852.SH_amount_mean', '000905.SH_amount_mean', '399006.SZ_amount_mean', '000852.SH_daily_return', '000905.SH_daily_return', '399006.SZ_daily_return', '000852.SH_up_ratio_20d', '000905.SH_up_ratio_20d', '399006.SZ_up_ratio_20d', '000852.SH_volatility', '000905.SH_volatility', '399006.SZ_volatility', '000852.SH_volume_change_rate', '000905.SH_volume_change_rate', '399006.SZ_volume_change_rate']\n",
|
||
"去除极值\n",
|
||
"开始截面 MAD 去极值处理 (k=3.0)...\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"MAD Filtering: 100%|██████████| 137/137 [00:16<00:00, 8.35it/s]\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"截面 MAD 去极值处理完成。\n",
|
||
"开始截面 MAD 去极值处理 (k=3.0)...\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"MAD Filtering: 100%|██████████| 137/137 [00:11<00:00, 12.05it/s]\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"截面 MAD 去极值处理完成。\n",
|
||
"开始截面 MAD 去极值处理 (k=3.0)...\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"MAD Filtering: 0it [00:00, ?it/s]\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"截面 MAD 去极值处理完成。\n",
|
||
"开始截面 MAD 去极值处理 (k=3.0)...\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"MAD Filtering: 0it [00:00, ?it/s]\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"截面 MAD 去极值处理完成。\n",
|
||
"feature_columns: ['vol', 'pct_chg', 'turnover_rate', 'volume_ratio', 'winner_rate', 'ts_turnover_rate_acceleration_5_20', 'ts_vol_sustain_10_30', 'cs_amount_outlier_10', 'ts_ff_to_total_turnover_ratio', 'ts_price_volume_trend_coherence_5_20', 'ts_ff_turnover_rate_surge_10', 'undist_profit_ps', 'ocfps', 'AR', 'BR', 'AR_BR', 'cashflow_to_ev_factor', 'book_to_price_ratio', 'turnover_rate_mean_5', 'variance_20', 'bbi_ratio_factor', 'daily_deviation', 'lg_elg_net_buy_vol', 'flow_lg_elg_intensity', 'sm_net_buy_vol', 'total_buy_vol', 'lg_elg_buy_prop', 'flow_struct_buy_change', 'lg_elg_net_buy_vol_change', 'flow_lg_elg_accel', 'chip_concentration_range', 'chip_skewness', 'floating_chip_proxy', 'cost_support_15pct_change', 'cat_winner_price_zone', 'flow_chip_consistency', 'profit_taking_vs_absorb', 'cat_is_positive', 'upside_vol', 'downside_vol', 'vol_ratio', 'return_skew', 'return_kurtosis', 'volume_change_rate', 'cat_volume_breakout', 'turnover_deviation', 'cat_turnover_spike', 'avg_volume_ratio', 'cat_volume_ratio_breakout', 'vol_spike', 'vol_std_5', 'atr_14', 'atr_6', 'obv', 'maobv_6', 'rsi_3', 'return_5', 'return_20', 'std_return_5', 'std_return_90', 'std_return_90_2', 'act_factor1', 'act_factor2', 'act_factor3', 'act_factor4', 'rank_act_factor1', 'rank_act_factor2', 'rank_act_factor3', 'cov', 'delta_cov', 'alpha_22_improved', 'alpha_003', 'alpha_007', 'alpha_013', 'vol_break', 'weight_roc5', 'smallcap_concentration', 'cost_stability', 'high_cost_break_days', 'liquidity_risk', 'turnover_std', 'mv_volatility', 'volume_growth', 'mv_growth', 'momentum_factor', 'resonance_factor', 'log_close', 'cat_vol_spike', 'up', 'down', 'obv_maobv_6', 'std_return_5_over_std_return_90', 'std_return_90_minus_std_return_90_2', 'cat_af2', 'cat_af3', 'cat_af4', 'act_factor5', 'act_factor6', 'active_buy_volume_large', 'active_buy_volume_big', 'active_buy_volume_small', 'buy_lg_vol_minus_sell_lg_vol', 'buy_elg_vol_minus_sell_elg_vol', 'ctrl_strength', 'low_cost_dev', 'asymmetry', 'lock_factor', 'cat_vol_break', 'cost_atr_adj', 'cat_golden_resonance', 'mv_turnover_ratio', 'mv_adjusted_volume', 'mv_weighted_turnover', 'nonlinear_mv_volume', 'mv_volume_ratio', 'mv_momentum', 'lg_flow_mom_corr_20_60', 'lg_flow_accel', 'profit_pressure', 'underwater_resistance', 'cost_conc_std_20', 'profit_decay_20', 'vol_amp_loss_20', 'vol_drop_profit_cnt_5', 'lg_flow_vol_interact_20', 'cost_break_confirm_cnt_5', 'atr_norm_channel_pos_14', 'turnover_diff_skew_20', 'lg_sm_flow_diverge_20', 'pullback_strong_20_20', 'vol_wgt_hist_pos_20', 'vol_adj_roc_20', 'cs_rank_net_lg_flow_val', 'cs_rank_elg_buy_ratio', 'cs_rank_rel_profit_margin', 'cs_rank_cost_breadth', 'cs_rank_dist_to_upper_cost', 'cs_rank_winner_rate', 'cs_rank_intraday_range', 'cs_rank_close_pos_in_range', 'cs_rank_pos_in_hist_range', 'cs_rank_vol_x_profit_margin', 'cs_rank_lg_flow_price_concordance', 'cs_rank_turnover_per_winner', 'cs_rank_volume_ratio', 'cs_rank_elg_buy_sell_sm_ratio', 'cs_rank_cost_dist_vol_ratio', 'cs_rank_size', 'cat_up_limit', 'industry_obv', 'industry_return_5', 'industry_return_20', 'industry__ema_5', 'industry__ema_13', 'industry__ema_20', 'industry__ema_60', 'industry_act_factor1', 'industry_act_factor2', 'industry_act_factor3', 'industry_act_factor4', 'industry_act_factor5', 'industry_act_factor6', 'industry_rank_act_factor1', 'industry_rank_act_factor2', 'industry_rank_act_factor3', 'industry_return_5_percentile', 'industry_return_20_percentile', '000852.SH_MACD', '000905.SH_MACD', '399006.SZ_MACD', '000852.SH_MACD_hist', '000905.SH_MACD_hist', '399006.SZ_MACD_hist', '000852.SH_RSI', '000905.SH_RSI', '399006.SZ_RSI', '000852.SH_Signal_line', '000905.SH_Signal_line', '399006.SZ_Signal_line', '000852.SH_amount_change_rate', '000905.SH_amount_change_rate', '399006.SZ_amount_change_rate', '000852.SH_amount_mean', '000905.SH_amount_mean', '399006.SZ_amount_mean', '000852.SH_daily_return', '000905.SH_daily_return', '399006.SZ_daily_return', '000852.SH_up_ratio_20d', '000905.SH_up_ratio_20d', '399006.SZ_up_ratio_20d', '000852.SH_volatility', '000905.SH_volatility', '399006.SZ_volatility', '000852.SH_volume_change_rate', '000905.SH_volume_change_rate', '399006.SZ_volume_change_rate']\n",
|
||
"df最小日期: 2019-01-02\n",
|
||
"df最大日期: 2025-05-23\n",
|
||
"2057539\n",
|
||
"train_data最小日期: 2020-01-02\n",
|
||
"train_data最大日期: 2022-12-30\n",
|
||
"1766694\n",
|
||
"test_data最小日期: 2023-01-03\n",
|
||
"test_data最大日期: 2025-05-23\n",
|
||
" ts_code trade_date log_circ_mv\n",
|
||
"0 000001.SZ 2019-01-02 16.574219\n",
|
||
"1 000001.SZ 2019-01-03 16.583965\n",
|
||
"2 000001.SZ 2019-01-04 16.633371\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"split_date = '2023-01-01'\n",
|
||
"train_data = df[filter_index & (df['trade_date'] <= split_date) & (df['trade_date'] >= '2020-01-01')]\n",
|
||
"test_data = df[(df['trade_date'] >= split_date)]\n",
|
||
"\n",
|
||
"print(df[['ts_code', 'trade_date', 'log_circ_mv']].head(3))\n",
|
||
"\n",
|
||
"industry_df = industry_df.sort_values(by=['trade_date'])\n",
|
||
"index_data = index_data.sort_values(by=['trade_date'])\n",
|
||
"\n",
|
||
"# train_data = train_data.merge(industry_df, on=['cat_l2_code', 'trade_date'], how='left')\n",
|
||
"# train_data = train_data.merge(index_data, on='trade_date', how='left')\n",
|
||
"# test_data = test_data.merge(industry_df, on=['cat_l2_code', 'trade_date'], how='left')\n",
|
||
"# test_data = test_data.merge(index_data, on='trade_date', how='left')\n",
|
||
"\n",
|
||
"train_data, test_data = train_data.replace([np.inf, -np.inf], np.nan), test_data.replace([np.inf, -np.inf], np.nan)\n",
|
||
"\n",
|
||
"# feature_columns_new = feature_columns[:]\n",
|
||
"# train_data, _ = create_deviation_within_dates(train_data, [col for col in feature_columns if col in train_data.columns])\n",
|
||
"# test_data, _ = create_deviation_within_dates(test_data, [col for col in feature_columns if col in train_data.columns])\n",
|
||
"\n",
|
||
"# feature_columns = [\n",
|
||
"# 'undist_profit_ps', \n",
|
||
"# 'AR_BR',\n",
|
||
"# 'pe_ttm',\n",
|
||
"# 'alpha_22_improved', \n",
|
||
"# 'alpha_003', \n",
|
||
"# 'alpha_007', \n",
|
||
"# 'alpha_013', \n",
|
||
"# 'cat_up_limit', \n",
|
||
"# 'cat_down_limit', \n",
|
||
"# 'up_limit_count_10d', \n",
|
||
"# 'down_limit_count_10d', \n",
|
||
"# 'consecutive_up_limit', \n",
|
||
"# 'vol_break', \n",
|
||
"# 'weight_roc5', \n",
|
||
"# 'price_cost_divergence', \n",
|
||
"# 'smallcap_concentration', \n",
|
||
"# 'cost_stability', \n",
|
||
"# 'high_cost_break_days', \n",
|
||
"# 'liquidity_risk', \n",
|
||
"# 'turnover_std', \n",
|
||
"# 'mv_volatility', \n",
|
||
"# 'volume_growth', \n",
|
||
"# 'mv_growth', \n",
|
||
"# 'lg_flow_mom_corr_20_60', \n",
|
||
"# 'lg_flow_accel', \n",
|
||
"# 'profit_pressure', \n",
|
||
"# 'underwater_resistance', \n",
|
||
"# 'cost_conc_std_20', \n",
|
||
"# 'profit_decay_20', \n",
|
||
"# 'vol_amp_loss_20', \n",
|
||
"# 'vol_drop_profit_cnt_5', \n",
|
||
"# 'lg_flow_vol_interact_20', \n",
|
||
"# 'cost_break_confirm_cnt_5', \n",
|
||
"# 'atr_norm_channel_pos_14', \n",
|
||
"# 'turnover_diff_skew_20', \n",
|
||
"# 'lg_sm_flow_diverge_20', \n",
|
||
"# 'pullback_strong_20_20', \n",
|
||
"# 'vol_wgt_hist_pos_20', \n",
|
||
"# 'vol_adj_roc_20',\n",
|
||
"# 'cashflow_to_ev_factor',\n",
|
||
"# 'ocfps',\n",
|
||
"# 'book_to_price_ratio',\n",
|
||
"# 'turnover_rate_mean_5',\n",
|
||
"# 'variance_20',\n",
|
||
"# 'bbi_ratio_factor'\n",
|
||
"# ]\n",
|
||
"# feature_columns = [col for col in feature_columns if col in train_data.columns]\n",
|
||
"# feature_columns = [col for col in feature_columns if not col.startswith('_')]\n",
|
||
"\n",
|
||
"numeric_columns = df.select_dtypes(include=['float64', 'int64']).columns\n",
|
||
"numeric_columns = [col for col in numeric_columns if col in feature_columns]\n",
|
||
"# feature_columns = select_top_features_by_rankic(df, numeric_columns, n=10)\n",
|
||
"print(feature_columns)\n",
|
||
"\n",
|
||
"# train_data = fill_nan_with_daily_median(train_data, feature_columns)\n",
|
||
"# test_data = fill_nan_with_daily_median(test_data, feature_columns)\n",
|
||
"\n",
|
||
"train_data = train_data.dropna(subset=[col for col in feature_columns if col in train_data.columns])\n",
|
||
"train_data = train_data.dropna(subset=['label'])\n",
|
||
"train_data = train_data.reset_index(drop=True)\n",
|
||
"# print(test_data.tail())\n",
|
||
"test_data = test_data.dropna(subset=[col for col in feature_columns if col in train_data.columns])\n",
|
||
"# test_data = test_data.dropna(subset=['label'])\n",
|
||
"test_data = test_data.reset_index(drop=True)\n",
|
||
"\n",
|
||
"transform_feature_columns = feature_columns\n",
|
||
"transform_feature_columns = [col for col in transform_feature_columns if col in feature_columns and not col.startswith('cat') and col in train_data.columns]\n",
|
||
"# transform_feature_columns.remove('undist_profit_ps')\n",
|
||
"print('去除极值')\n",
|
||
"cs_mad_filter(train_data, transform_feature_columns)\n",
|
||
"# print('中性化')\n",
|
||
"# cs_neutralize_market_cap_numpy(train_data, transform_feature_columns)\n",
|
||
"# print('标准化')\n",
|
||
"# cs_zscore_standardize(train_data, transform_feature_columns)\n",
|
||
"\n",
|
||
"cs_mad_filter(test_data, transform_feature_columns)\n",
|
||
"# cs_neutralize_market_cap_numpy(test_data, transform_feature_columns)\n",
|
||
"# cs_zscore_standardize(test_data, transform_feature_columns)\n",
|
||
"\n",
|
||
"mad_filter_feature_columns = [col for col in feature_columns if col not in transform_feature_columns and not col.startswith('cat') and col in train_data.columns]\n",
|
||
"cs_mad_filter(train_data, mad_filter_feature_columns)\n",
|
||
"cs_mad_filter(test_data, mad_filter_feature_columns)\n",
|
||
"\n",
|
||
"\n",
|
||
"print(f'feature_columns: {feature_columns}')\n",
|
||
"\n",
|
||
"\n",
|
||
"print(f\"df最小日期: {df['trade_date'].min().strftime('%Y-%m-%d')}\")\n",
|
||
"print(f\"df最大日期: {df['trade_date'].max().strftime('%Y-%m-%d')}\")\n",
|
||
"print(len(train_data))\n",
|
||
"print(f\"train_data最小日期: {train_data['trade_date'].min().strftime('%Y-%m-%d')}\")\n",
|
||
"print(f\"train_data最大日期: {train_data['trade_date'].max().strftime('%Y-%m-%d')}\")\n",
|
||
"print(len(test_data))\n",
|
||
"print(f\"test_data最小日期: {test_data['trade_date'].min().strftime('%Y-%m-%d')}\")\n",
|
||
"print(f\"test_data最大日期: {test_data['trade_date'].max().strftime('%Y-%m-%d')}\")\n",
|
||
"\n",
|
||
"cat_columns = [col for col in feature_columns if col.startswith('cat')]\n",
|
||
"for col in cat_columns:\n",
|
||
" train_data[col] = train_data[col].astype('category')\n",
|
||
" test_data[col] = test_data[col].astype('category')\n",
|
||
"\n",
|
||
"print(df[['ts_code', 'trade_date', 'log_circ_mv']].head(3))\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 23,
|
||
"id": "3ff2d1c5",
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"from sklearn.preprocessing import StandardScaler\n",
|
||
"from sklearn.linear_model import LogisticRegression\n",
|
||
"import matplotlib.pyplot as plt # 保持 matplotlib 导入，尽管LightGBM的绘图功能已移除\n",
|
||
"from sklearn.decomposition import PCA\n",
|
||
"import datetime # 用于日期计算\n",
|
||
"from catboost import CatBoostClassifier\n",
|
||
"from catboost import Pool\n",
|
||
"import lightgbm as lgb\n",
|
||
"\n",
|
||
"def train_model(train_data_df, feature_columns,\n",
|
||
" print_info=True, # 调整参数名，更通用\n",
|
||
" validation_days=180, use_pca=False, split_date=None,\n",
|
||
" target_column='label', type='light'): # 增加目标列参数\n",
|
||
"\n",
|
||
" print('train data size: ', len(train_data_df))\n",
|
||
" print(train_data_df[['ts_code', 'trade_date', 'log_circ_mv']])\n",
|
||
" # 确保数据按时间排序\n",
|
||
" train_data_df = train_data_df.sort_values(by='trade_date')\n",
|
||
"\n",
|
||
" # 识别数值型特征列\n",
|
||
" numeric_feature_columns = train_data_df[feature_columns].select_dtypes(include=['float64', 'int64']).columns.tolist()\n",
|
||
"\n",
|
||
" # 去除标签为空的样本\n",
|
||
" initial_len = len(train_data_df)\n",
|
||
" train_data_df = train_data_df.dropna(subset=[target_column])\n",
|
||
"\n",
|
||
" if print_info:\n",
|
||
" print(f'原始样本数: {initial_len}, 去除标签为空后样本数: {len(train_data_df)}')\n",
|
||
"\n",
|
||
" # 提取特征和标签，只取数值型特征用于线性回归\n",
|
||
" \n",
|
||
" if split_date is None:\n",
|
||
" all_dates = train_data_df['trade_date'].unique() # 获取所有唯一的 trade_date\n",
|
||
" split_date = all_dates[-validation_days] # 划分点为倒数第 validation_days 天\n",
|
||
" train_data_split = train_data_df[train_data_df['trade_date'] < split_date] # 训练集\n",
|
||
" val_data_split = train_data_df[train_data_df['trade_date'] >= split_date] # 验证集\n",
|
||
" \n",
|
||
" X_train = train_data_split[feature_columns]\n",
|
||
" y_train = train_data_split[target_column]\n",
|
||
" \n",
|
||
" X_val = val_data_split[feature_columns]\n",
|
||
" y_val = val_data_split['label']\n",
|
||
"\n",
|
||
"\n",
|
||
" # # 标准化数值特征 (使用 StandardScaler 对训练集fit并transform, 对验证集只transform)\n",
|
||
" scaler = StandardScaler()\n",
|
||
" # X_train = scaler.fit_transform(X_train)\n",
|
||
"\n",
|
||
" # 训练线性回归模型\n",
|
||
" # model = LogisticRegression(random_state=42)\n",
|
||
" \n",
|
||
" # # 使用处理后的特征和样本权重进行训练\n",
|
||
" # model.fit(X_train, y_train)\n",
|
||
"\n",
|
||
"\n",
|
||
" if type == 'cat':\n",
|
||
" params = {\n",
|
||
" 'loss_function': 'Logloss', # 适用于二分类\n",
|
||
" 'eval_metric': 'Logloss', # 评估指标\n",
|
||
" 'iterations': 1500,\n",
|
||
" 'learning_rate': 0.01,\n",
|
||
" 'depth': 10, # 控制模型复杂度\n",
|
||
" 'l2_leaf_reg': 50, # L2 正则化\n",
|
||
" 'verbose': 100,\n",
|
||
" 'early_stopping_rounds': 300,\n",
|
||
" # 'od_type': 'Iter', # Overfitting detector type\n",
|
||
" # 'od_wait': 300, # Number of iterations to wait after the bes\n",
|
||
" 'one_hot_max_size': 50,\n",
|
||
" 'class_weights': [0.6, 1.2],\n",
|
||
" 'task_type': 'GPU',\n",
|
||
" 'has_time': True,\n",
|
||
" 'random_seed': 7\n",
|
||
" }\n",
|
||
" cat_features = [i for i, col in enumerate(feature_columns) if col.startswith('cat')]\n",
|
||
" train_pool = Pool(data=X_train, label=y_train, cat_features=cat_features)\n",
|
||
" val_pool = Pool(data=X_val, label=y_val, cat_features=cat_features)\n",
|
||
"\n",
|
||
"\n",
|
||
" model = CatBoostClassifier(**params)\n",
|
||
" model.fit(train_pool,\n",
|
||
" eval_set=val_pool, \n",
|
||
" plot=True, \n",
|
||
" use_best_model=True\n",
|
||
" )\n",
|
||
" elif type == 'light':\n",
|
||
" params = {\n",
|
||
" 'objective': 'binary',\n",
|
||
" 'metric': 'average_precision',\n",
|
||
" 'learning_rate': 0.01,\n",
|
||
" 'is_unbalance': True,\n",
|
||
" 'num_leaves': 2048,\n",
|
||
" 'min_data_in_leaf': 1024,\n",
|
||
" 'max_depth': 32,\n",
|
||
" 'max_bin': 1024,\n",
|
||
" 'feature_fraction': 0.5,\n",
|
||
" 'bagging_fraction': 0.5,\n",
|
||
" 'bagging_freq': 1,\n",
|
||
" 'lambda_l1': 50,\n",
|
||
" 'lambda_l2': 50,\n",
|
||
" 'verbosity': -1,\n",
|
||
" 'num_threads' : 8\n",
|
||
" }\n",
|
||
" categorical_feature = [col for col in feature_columns if 'cat' in col]\n",
|
||
" train_dataset = lgb.Dataset(\n",
|
||
" X_train, label=y_train,\n",
|
||
" categorical_feature=categorical_feature\n",
|
||
" )\n",
|
||
" val_dataset = lgb.Dataset(\n",
|
||
" X_val, label=y_val,\n",
|
||
" categorical_feature=categorical_feature\n",
|
||
" )\n",
|
||
"\n",
|
||
" evals = {}\n",
|
||
" callbacks = [lgb.log_evaluation(period=1000),\n",
|
||
" lgb.callback.record_evaluation(evals),\n",
|
||
" lgb.early_stopping(100, first_metric_only=True)\n",
|
||
" ]\n",
|
||
" # 训练模型\n",
|
||
" model = lgb.train(\n",
|
||
" params, train_dataset, num_boost_round=1000,\n",
|
||
" valid_sets=[train_dataset, val_dataset], valid_names=['train', 'valid'],\n",
|
||
" callbacks=callbacks\n",
|
||
" )\n",
|
||
"\n",
|
||
" # 打印特征重要性（如果需要）\n",
|
||
" if True:\n",
|
||
" lgb.plot_metric(evals)\n",
|
||
" lgb.plot_importance(model, importance_type='split', max_num_features=20)\n",
|
||
" plt.show()\n",
|
||
"\n",
|
||
"\n",
|
||
" return model, scaler, None # 返回训练好的模型、scaler 和 pca 对象"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 24,
|
||
"id": "c6eb5cd4-e714-420a-ac48-39af3e11ee81",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T15:03:18.426481Z",
|
||
"start_time": "2025-04-03T15:02:19.926352Z"
|
||
}
|
||
},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"train data size: 36400\n",
|
||
" ts_code trade_date log_circ_mv\n",
|
||
"0 600306.SH 2020-01-02 11.552040\n",
|
||
"1 603269.SH 2020-01-02 11.324801\n",
|
||
"2 002633.SZ 2020-01-02 11.759023\n",
|
||
"3 603991.SH 2020-01-02 11.181150\n",
|
||
"4 000691.SZ 2020-01-02 11.677910\n",
|
||
"... ... ... ...\n",
|
||
"36395 600615.SH 2022-12-30 12.027909\n",
|
||
"36396 603829.SH 2022-12-30 12.034572\n",
|
||
"36397 603037.SH 2022-12-30 12.035767\n",
|
||
"36398 002767.SZ 2022-12-30 11.896427\n",
|
||
"36399 600561.SH 2022-12-30 11.858571\n",
|
||
"\n",
|
||
"[36400 rows x 3 columns]\n",
|
||
"原始样本数: 36400, 去除标签为空后样本数: 36400\n"
|
||
]
|
||
},
|
||
{
|
||
"data": {
|
||
"application/vnd.jupyter.widget-view+json": {
|
||
"model_id": "f2be3dc13c844232880f85edebee22c9",
|
||
"version_major": 2,
|
||
"version_minor": 0
|
||
},
|
||
"text/plain": [
|
||
"MetricVisualizer(layout=Layout(align_self='stretch', height='500px'))"
|
||
]
|
||
},
|
||
"metadata": {},
|
||
"output_type": "display_data"
|
||
},
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"0:\tlearn: 0.6889525\ttest: 0.6893155\tbest: 0.6893155 (0)\ttotal: 30.4ms\tremaining: 45.6s\n",
|
||
"100:\tlearn: 0.5065717\ttest: 0.5493420\tbest: 0.5493420 (100)\ttotal: 12.6s\tremaining: 2m 54s\n",
|
||
"200:\tlearn: 0.4728006\ttest: 0.5277064\tbest: 0.5277064 (200)\ttotal: 25.2s\tremaining: 2m 42s\n",
|
||
"300:\tlearn: 0.4572238\ttest: 0.5226125\tbest: 0.5225487 (298)\ttotal: 40s\tremaining: 2m 39s\n",
|
||
"400:\tlearn: 0.4479278\ttest: 0.5208197\tbest: 0.5208197 (400)\ttotal: 54.5s\tremaining: 2m 29s\n",
|
||
"500:\tlearn: 0.4413356\ttest: 0.5201745\tbest: 0.5201414 (499)\ttotal: 1m 8s\tremaining: 2m 16s\n",
|
||
"600:\tlearn: 0.4356760\ttest: 0.5205238\tbest: 0.5201414 (499)\ttotal: 1m 22s\tremaining: 2m 2s\n",
|
||
"700:\tlearn: 0.4293049\ttest: 0.5203200\tbest: 0.5201414 (499)\ttotal: 1m 35s\tremaining: 1m 48s\n",
|
||
"800:\tlearn: 0.4227554\ttest: 0.5200567\tbest: 0.5198929 (756)\ttotal: 1m 48s\tremaining: 1m 34s\n",
|
||
"900:\tlearn: 0.4162998\ttest: 0.5200167\tbest: 0.5198929 (756)\ttotal: 2m 3s\tremaining: 1m 21s\n",
|
||
"1000:\tlearn: 0.4097120\ttest: 0.5200234\tbest: 0.5198929 (756)\ttotal: 2m 17s\tremaining: 1m 8s\n",
|
||
"bestTest = 0.5198929335\n",
|
||
"bestIteration = 756\n",
|
||
"Shrink model to first 757 iterations.\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"\n",
|
||
"gc.collect()\n",
|
||
"\n",
|
||
"use_pca = False\n",
|
||
"type = 'cat'\n",
|
||
"# feature_contri = [2 if feat.startswith('act_factor') or 'buy' in feat or 'sell' in feat else 1 for feat in feature_columns]\n",
|
||
"# light_params['feature_contri'] = feature_contri\n",
|
||
"# print(f'feature_contri: {feature_contri}')\n",
|
||
"model, scaler, pca = train_model(train_data\n",
|
||
" .dropna(subset=['label']).groupby('trade_date', group_keys=False)\n",
|
||
" .apply(lambda x: x.nsmallest(50, 'total_mv'))\n",
|
||
" .merge(industry_df, on=['cat_l2_code', 'trade_date'], how='left')\n",
|
||
" .merge(index_data, on='trade_date', how='left'), feature_columns, type=type)\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 32,
|
||
"id": "5d1522a7538db91b",
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2025-04-03T15:04:39.656944Z",
|
||
"start_time": "2025-04-03T15:04:39.298483Z"
|
||
}
|
||
},
|
||
"outputs": [],
|
||
"source": [
|
||
"score_df = test_data.groupby('trade_date', group_keys=False).apply(lambda x: x.nsmallest(300, 'total_mv'))\n",
|
||
"# score_df = fill_nan_with_daily_median(score_df, ['pe_ttm'])\n",
|
||
"# score_df = score_df[score_df['pe_ttm'] > 0]\n",
|
||
"score_df = score_df.merge(industry_df, on=['cat_l2_code', 'trade_date'], how='left')\n",
|
||
"score_df = score_df.merge(index_data, on='trade_date', how='left')\n",
|
||
"# score_df = score_df.groupby('trade_date', group_keys=False).apply(lambda x: x.nsmallest(50, 'total_mv')).reset_index()\n",
|
||
"numeric_columns = score_df.select_dtypes(include=['float64', 'int64']).columns\n",
|
||
"numeric_columns = [col for col in feature_columns if col in numeric_columns]\n",
|
||
"# score_df.loc[:, numeric_columns] = scaler.transform(score_df[numeric_columns])\n",
|
||
"# score_df = cross_sectional_standardization(score_df, numeric_columns)\n",
|
||
"\n",
|
||
"if type == 'cat':\n",
|
||
" score_df['score'] = model.predict_proba(score_df[feature_columns])[:, 1]\n",
|
||
"elif type == 'light':\n",
|
||
" score_df['score'] = model.predict(score_df[feature_columns])\n",
|
||
"score_df['score_ranks'] = score_df.groupby('trade_date')['score'].rank(ascending=True)\n",
|
||
"\n",
|
||
"score_df = score_df.groupby('trade_date', group_keys=False).apply(\n",
|
||
" lambda x: x[x['score'] >= x['score'].quantile(0.90)] # 计算90%分位数作为阈值，筛选分数>=阈值的行\n",
|
||
").reset_index(drop=True) # drop=True 避免添加旧索引列\n",
|
||
"# save_df = score_df.groupby('trade_date', group_keys=False).apply(lambda x: x.nlargest(1, 'score')).reset_index()\n",
|
||
"save_df = score_df.groupby('trade_date', group_keys=False).apply(lambda x: x.nsmallest(2, 'total_mv')).reset_index()\n",
|
||
"save_df = save_df.sort_values(['trade_date', 'score'])\n",
|
||
"save_df[['trade_date', 'score', 'ts_code']].to_csv('predictions_test.tsv', index=False)\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 26,
|
||
"id": "09b1799e",
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"197\n",
|
||
"['vol', 'pct_chg', 'turnover_rate', 'volume_ratio', 'winner_rate', 'ts_turnover_rate_acceleration_5_20', 'ts_vol_sustain_10_30', 'cs_amount_outlier_10', 'ts_ff_to_total_turnover_ratio', 'ts_price_volume_trend_coherence_5_20', 'ts_ff_turnover_rate_surge_10', 'undist_profit_ps', 'ocfps', 'AR', 'BR', 'AR_BR', 'cashflow_to_ev_factor', 'book_to_price_ratio', 'turnover_rate_mean_5', 'variance_20', 'bbi_ratio_factor', 'daily_deviation', 'lg_elg_net_buy_vol', 'flow_lg_elg_intensity', 'sm_net_buy_vol', 'total_buy_vol', 'lg_elg_buy_prop', 'flow_struct_buy_change', 'lg_elg_net_buy_vol_change', 'flow_lg_elg_accel', 'chip_concentration_range', 'chip_skewness', 'floating_chip_proxy', 'cost_support_15pct_change', 'cat_winner_price_zone', 'flow_chip_consistency', 'profit_taking_vs_absorb', 'cat_is_positive', 'upside_vol', 'downside_vol', 'vol_ratio', 'return_skew', 'return_kurtosis', 'volume_change_rate', 'cat_volume_breakout', 'turnover_deviation', 'cat_turnover_spike', 'avg_volume_ratio', 'cat_volume_ratio_breakout', 'vol_spike', 'vol_std_5', 'atr_14', 'atr_6', 'obv', 'maobv_6', 'rsi_3', 'return_5', 'return_20', 'std_return_5', 'std_return_90', 'std_return_90_2', 'act_factor1', 'act_factor2', 'act_factor3', 'act_factor4', 'rank_act_factor1', 'rank_act_factor2', 'rank_act_factor3', 'cov', 'delta_cov', 'alpha_22_improved', 'alpha_003', 'alpha_007', 'alpha_013', 'vol_break', 'weight_roc5', 'smallcap_concentration', 'cost_stability', 'high_cost_break_days', 'liquidity_risk', 'turnover_std', 'mv_volatility', 'volume_growth', 'mv_growth', 'momentum_factor', 'resonance_factor', 'log_close', 'cat_vol_spike', 'up', 'down', 'obv_maobv_6', 'std_return_5_over_std_return_90', 'std_return_90_minus_std_return_90_2', 'cat_af2', 'cat_af3', 'cat_af4', 'act_factor5', 'act_factor6', 'active_buy_volume_large', 'active_buy_volume_big', 'active_buy_volume_small', 'buy_lg_vol_minus_sell_lg_vol', 'buy_elg_vol_minus_sell_elg_vol', 'ctrl_strength', 'low_cost_dev', 'asymmetry', 'lock_factor', 'cat_vol_break', 'cost_atr_adj', 'cat_golden_resonance', 'mv_turnover_ratio', 'mv_adjusted_volume', 'mv_weighted_turnover', 'nonlinear_mv_volume', 'mv_volume_ratio', 'mv_momentum', 'lg_flow_mom_corr_20_60', 'lg_flow_accel', 'profit_pressure', 'underwater_resistance', 'cost_conc_std_20', 'profit_decay_20', 'vol_amp_loss_20', 'vol_drop_profit_cnt_5', 'lg_flow_vol_interact_20', 'cost_break_confirm_cnt_5', 'atr_norm_channel_pos_14', 'turnover_diff_skew_20', 'lg_sm_flow_diverge_20', 'pullback_strong_20_20', 'vol_wgt_hist_pos_20', 'vol_adj_roc_20', 'cs_rank_net_lg_flow_val', 'cs_rank_elg_buy_ratio', 'cs_rank_rel_profit_margin', 'cs_rank_cost_breadth', 'cs_rank_dist_to_upper_cost', 'cs_rank_winner_rate', 'cs_rank_intraday_range', 'cs_rank_close_pos_in_range', 'cs_rank_pos_in_hist_range', 'cs_rank_vol_x_profit_margin', 'cs_rank_lg_flow_price_concordance', 'cs_rank_turnover_per_winner', 'cs_rank_volume_ratio', 'cs_rank_elg_buy_sell_sm_ratio', 'cs_rank_cost_dist_vol_ratio', 'cs_rank_size', 'cat_up_limit', 'industry_obv', 'industry_return_5', 'industry_return_20', 'industry__ema_5', 'industry__ema_13', 'industry__ema_20', 'industry__ema_60', 'industry_act_factor1', 'industry_act_factor2', 'industry_act_factor3', 'industry_act_factor4', 'industry_act_factor5', 'industry_act_factor6', 'industry_rank_act_factor1', 'industry_rank_act_factor2', 'industry_rank_act_factor3', 'industry_return_5_percentile', 'industry_return_20_percentile', '000852.SH_MACD', '000905.SH_MACD', '399006.SZ_MACD', '000852.SH_MACD_hist', '000905.SH_MACD_hist', '399006.SZ_MACD_hist', '000852.SH_RSI', '000905.SH_RSI', '399006.SZ_RSI', '000852.SH_Signal_line', '000905.SH_Signal_line', '399006.SZ_Signal_line', '000852.SH_amount_change_rate', '000905.SH_amount_change_rate', '399006.SZ_amount_change_rate', '000852.SH_amount_mean', '000905.SH_amount_mean', '399006.SZ_amount_mean', '000852.SH_daily_return', '000905.SH_daily_return', '399006.SZ_daily_return', '000852.SH_up_ratio_20d', '000905.SH_up_ratio_20d', '399006.SZ_up_ratio_20d', '000852.SH_volatility', '000905.SH_volatility', '399006.SZ_volatility', '000852.SH_volume_change_rate', '000905.SH_volume_change_rate', '399006.SZ_volume_change_rate']\n",
|
||
"[]\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"print(len(feature_columns))\n",
|
||
"print(feature_columns)\n",
|
||
"print([col for col in feature_columns if 'total_mv' in col])"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 27,
|
||
"id": "e53b209a",
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"5595 2057539\n",
|
||
" ts_code trade_date turnover_rate\n",
|
||
"0 000001.SZ 2023-01-03 1.1307\n",
|
||
"1 000001.SZ 2023-01-04 1.1284\n",
|
||
"2 000001.SZ 2023-01-05 0.8582\n",
|
||
"3 000001.SZ 2023-01-06 0.6162\n",
|
||
"4 000001.SZ 2023-01-09 0.5450\n",
|
||
"... ... ... ...\n",
|
||
"1766689 605599.SH 2025-05-19 0.4952\n",
|
||
"1766690 605599.SH 2025-05-20 1.6447\n",
|
||
"1766691 605599.SH 2025-05-21 1.2658\n",
|
||
"1766692 605599.SH 2025-05-22 0.7522\n",
|
||
"1766693 605599.SH 2025-05-23 0.6051\n",
|
||
"\n",
|
||
"[1766694 rows x 3 columns]\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"print(len(train_data[train_data['pct_chg'] > 7]), len(train_data))\n",
|
||
"print(test_data[['ts_code', 'trade_date', 'turnover_rate']])"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 28,
|
||
"id": "364e821a",
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, roc_curve\n",
|
||
"import matplotlib.pyplot as plt\n",
|
||
"import numpy as np\n",
|
||
"\n",
|
||
"def calculate_binary_classification_metrics(df: pd.DataFrame, score_col: str, label_col: str, future_return_col: str = None, total_mv_col: str = None, n_mv_bins: int = 10, threshold: float = 0.5):\n",
|
||
" \"\"\"\n",
|
||
" 计算二分类模型的评估指标，可选择计算 score 和 future_return 的相关性，\n",
|
||
" 并可选择计算 score 在按总市值 (total_mv) 分为 n 份后的每个分组上的预测性能（ROC AUC）。\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" df: 包含 score (预测概率或置信度), label (真实二分类标签), 可选的 future_return 和 total_mv 的 Pandas DataFrame。\n",
|
||
" score_col: 包含模型预测 score 的列名。\n",
|
||
" label_col: 包含真实二分类标签 (0 或 1) 的列名。\n",
|
||
" future_return_col: (可选) 包含未来收益率的列名，用于计算相关性。\n",
|
||
" total_mv_col: (可选) 包含总市值的列名，用于按市值分 n 份分析预测性能。\n",
|
||
" n_mv_bins: (可选) 将总市值分为多少份，默认为 5。\n",
|
||
" threshold: 将 score 转换为预测类别的阈值，默认为 0.5。\n",
|
||
"\n",
|
||
" Returns:\n",
|
||
" 一个包含以下评估指标的字典：\n",
|
||
" - accuracy: 准确率\n",
|
||
" - precision: 精确率\n",
|
||
" - recall: 召回率\n",
|
||
" - f1: F1 分数\n",
|
||
" - roc_auc: ROC AUC 值\n",
|
||
" - fpr: ROC 曲线的假正率 (False Positive Rate)\n",
|
||
" - tpr: ROC 曲线的真正率 (True Positive Rate)\n",
|
||
" - thresholds: ROC 曲线的阈值\n",
|
||
" - score_return_correlation: (如果 future_return_col 提供) score 和 future_return 的皮尔逊相关系数\n",
|
||
" - mv_roc_auc: (如果 total_mv_col 提供) 一个字典，包含按总市值分为 n 份后的每个市值分组对应的 ROC AUC 值\n",
|
||
" \"\"\"\n",
|
||
" y_true = df[label_col].values\n",
|
||
" y_score = df[score_col].values\n",
|
||
" y_pred = (y_score >= threshold).astype(int)\n",
|
||
"\n",
|
||
" metrics = {}\n",
|
||
" metrics['accuracy'] = accuracy_score(y_true, y_pred)\n",
|
||
" metrics['precision'] = precision_score(y_true, y_pred)\n",
|
||
" metrics['recall'] = recall_score(y_true, y_pred)\n",
|
||
" metrics['f1'] = f1_score(y_true, y_pred)\n",
|
||
" metrics['roc_auc'] = roc_auc_score(y_true, y_score)\n",
|
||
" metrics['fpr'], metrics['tpr'], metrics['thresholds'] = roc_curve(y_true, y_score)\n",
|
||
"\n",
|
||
" if future_return_col in df.columns:\n",
|
||
" metrics['score_return_correlation'] = df[score_col].corr(df[future_return_col])\n",
|
||
"\n",
|
||
" if total_mv_col in df.columns and n_mv_bins > 1:\n",
|
||
" metrics['mv_roc_auc'] = {}\n",
|
||
" df['mv_quantile'] = pd.cut(df[total_mv_col], bins=n_mv_bins, labels=False, duplicates='drop')\n",
|
||
" for i in range(df['mv_quantile'].nunique()):\n",
|
||
" mv_group = df[df['mv_quantile'] == i]\n",
|
||
" if len(mv_group) > 0 and len(np.unique(mv_group[label_col])) > 1 and len(np.unique(mv_group[score_col])) > 1:\n",
|
||
" roc_auc_mv = roc_auc_score(mv_group[label_col], mv_group[score_col])\n",
|
||
" lower_bound = df[total_mv_col][df['mv_quantile'] == i].min()\n",
|
||
" upper_bound = df[total_mv_col][df['mv_quantile'] == i].max()\n",
|
||
" metrics['mv_roc_auc'][f'{lower_bound:.0e}-{upper_bound:.0e}'] = roc_auc_mv\n",
|
||
" else:\n",
|
||
" lower_bound = df[total_mv_col][df['mv_quantile'] == i].min()\n",
|
||
" upper_bound = df[total_mv_col][df['mv_quantile'] == i].max()\n",
|
||
" metrics['mv_roc_auc'][f'{lower_bound:.0e}-{upper_bound:.0e}'] = np.nan\n",
|
||
" print(f'{lower_bound:.0e}-{upper_bound:.0e}')\n",
|
||
" df.drop(columns=['mv_quantile'], inplace=True)\n",
|
||
"\n",
|
||
" return metrics\n",
|
||
"\n",
|
||
"def plot_roc_curve(metrics: dict):\n",
|
||
" plt.figure(figsize=(8, 6))\n",
|
||
" plt.plot(metrics['fpr'], metrics['tpr'], color='darkorange', lw=2, label=f'ROC curve (AUC = {metrics[\"roc_auc\"]:.2f})')\n",
|
||
" plt.plot([0, 1], [0, 1], color='navy', lw=2, linestyle='--')\n",
|
||
" plt.xlabel('False Positive Rate')\n",
|
||
" plt.ylabel('True Positive Rate')\n",
|
||
" plt.title('Receiver Operating Characteristic (ROC)')\n",
|
||
" plt.legend(loc=\"lower right\")\n",
|
||
" plt.grid(True)\n",
|
||
" plt.show()\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 29,
|
||
"id": "1f6e6336",
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"6e+04-9e+04\n",
|
||
"9e+04-1e+05\n",
|
||
"1e+05-1e+05\n",
|
||
"1e+05-1e+05\n",
|
||
"1e+05-2e+05\n",
|
||
"2e+05-2e+05\n",
|
||
"2e+05-2e+05\n",
|
||
"2e+05-2e+05\n",
|
||
"2e+05-3e+05\n",
|
||
"3e+05-3e+05\n",
|
||
"二分类评估指标:\n",
|
||
"accuracy: 0.6521\n",
|
||
"precision: 0.4678\n",
|
||
"recall: 0.2293\n",
|
||
"f1: 0.3077\n",
|
||
"roc_auc: 0.6188\n",
|
||
"fpr: (array of length 7419)\n",
|
||
"tpr: (array of length 7419)\n",
|
||
"thresholds: (array of length 7419)\n",
|
||
"score_return_correlation: -0.0420\n",
|
||
"mv_roc_auc: {'6e+04-9e+04': np.float64(0.5368389780154487), '9e+04-1e+05': np.float64(0.5842281184370101), '1e+05-1e+05': np.float64(0.5813697835731819), '1e+05-2e+05': np.float64(0.5755756816649003), '2e+05-2e+05': np.float64(0.6025259170254149), '2e+05-3e+05': np.float64(0.6160037489958047), '3e+05-3e+05': np.float64(0.6159040696993592)}\n"
|
||
]
|
||
},
|
||
{
|
||
"data": {
|
||
"image/png": "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",
|
||
"text/plain": [
|
||
"<Figure size 800x600 with 1 Axes>"
|
||
]
|
||
},
|
||
"metadata": {},
|
||
"output_type": "display_data"
|
||
}
|
||
],
|
||
"source": [
|
||
"\n",
|
||
"for sd in [\n",
|
||
" # save_df1, \n",
|
||
" # save_df2, \n",
|
||
" score_df\n",
|
||
" ]:\n",
|
||
" # 计算二分类指标\n",
|
||
" evaluation_metrics = calculate_binary_classification_metrics(sd, score_col='score', label_col='label', threshold=0.6,\n",
|
||
" future_return_col='future_return', total_mv_col='total_mv')\n",
|
||
"\n",
|
||
" # 打印指标\n",
|
||
" print(\"二分类评估指标:\")\n",
|
||
" for metric, value in evaluation_metrics.items():\n",
|
||
" if isinstance(value, (float, int)):\n",
|
||
" print(f\"{metric}: {value:.4f}\")\n",
|
||
" elif isinstance(value, (list, tuple, np.ndarray)):\n",
|
||
" print(f\"{metric}: (array of length {len(value)})\")\n",
|
||
" else:\n",
|
||
" print(f\"{metric}: {value}\")\n",
|
||
"\n",
|
||
" # 绘制 ROC 曲线\n",
|
||
" plot_roc_curve(evaluation_metrics)"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 30,
|
||
"id": "7e9023cc",
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def analyze_factors(\n",
|
||
" df: pd.DataFrame,\n",
|
||
" feature_columns: list[str],\n",
|
||
" target_column: str = 'target', # 假设目标列默认为 'target'\n",
|
||
" trade_date_col: str = 'trade_date', # 假设日期列默认为 'trade_date'\n",
|
||
" mcap_col: str = 'total_mv', # 新增: 市值列名称\n",
|
||
" mcap_bins: int = 5 # 新增: 市值分位数的数量 (例如 5 表示五分位数)\n",
|
||
") -> pd.DataFrame:\n",
|
||
" \"\"\"\n",
|
||
" 分析DataFrame中指定特征列的各种指标，包括基本统计、相关性、日间IC、ICIR以及在不同市值分位数上的IC。\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" df (pd.DataFrame): 包含日期、目标列、特征列和市值列的DataFrame。\n",
|
||
" 需要包含 trade_date_col, target_column, feature_columns 和 mcap_col 中的所有列。\n",
|
||
" feature_columns (list[str]): 需要分析的特征列名称列表。\n",
|
||
" target_column (str): 目标变量列的名称。\n",
|
||
" trade_date_col (str): 交易日期列的名称。\n",
|
||
" mcap_col (str): 市值列的名称。\n",
|
||
" mcap_bins (int): 市值分位数的数量 (例如 5 表示五分位数)。\n",
|
||
"\n",
|
||
" Returns:\n",
|
||
" pd.DataFrame: 包含各个因子分析指标的汇总DataFrame。\n",
|
||
" 同时打印因子在不同市值分位数上的平均IC表格。\n",
|
||
" 如果输入数据或列有问题，可能返回空或包含NaN的DataFrame。\n",
|
||
" \"\"\"\n",
|
||
"\n",
|
||
" # --- 数据校验 ---\n",
|
||
" required_cols = [trade_date_col, target_column, mcap_col] + feature_columns\n",
|
||
" if not all(col in df.columns for col in required_cols):\n",
|
||
" missing = [col for col in required_cols if col not in df.columns]\n",
|
||
" print(f\"错误: 输入DataFrame缺少必需的列: {missing}\")\n",
|
||
" return pd.DataFrame() # 返回空DataFrame\n",
|
||
"\n",
|
||
" # 确保日期列是 datetime 类型\n",
|
||
" df = df.copy() # 在副本上操作\n",
|
||
" df[trade_date_col] = pd.to_datetime(df[trade_date_col], errors='coerce')\n",
|
||
" df.dropna(subset=[trade_date_col], inplace=True) # 移除日期转换失败的行\n",
|
||
"\n",
|
||
" # 过滤掉那些在 feature_columns, target_column, mcap_col 上有 NaN 的行，以确保后续计算是在完整数据上\n",
|
||
" # 直接在 df 副本上进行清洗\n",
|
||
" initial_rows_before_clean = len(df)\n",
|
||
" df.dropna(subset=feature_columns + [target_column, mcap_col], inplace=True)\n",
|
||
" rows_dropped_clean = initial_rows_before_clean - len(df)\n",
|
||
" if rows_dropped_clean > 0:\n",
|
||
" print(f\"警告: 移除了 {rows_dropped_clean} 行，因为其特征、目标或市值列存在空值。\")\n",
|
||
"\n",
|
||
" if df.empty:\n",
|
||
" print(\"错误: 清理缺失值后数据为空，无法进行因子分析。\")\n",
|
||
" return pd.DataFrame() # 返回空DataFrame\n",
|
||
"\n",
|
||
"\n",
|
||
" print(f\"开始分析 {len(feature_columns)} 个因子指标...\")\n",
|
||
"\n",
|
||
" # --- 1. 基本因子统计量 ---\n",
|
||
" basic_stats = df[feature_columns].describe().T\n",
|
||
"\n",
|
||
" print(\"\\n--- 基本因子统计量 ---\")\n",
|
||
" print(basic_stats)\n",
|
||
"\n",
|
||
" # --- 2. 因子与目标变量的整体相关性 ---\n",
|
||
" overall_correlation = {}\n",
|
||
" for feature in feature_columns:\n",
|
||
" # 在清理后的 df 上计算相关性\n",
|
||
" if df[[feature, target_column]].dropna().shape[0] > 1: # 确保至少有两个有效数据点\n",
|
||
" overall_correlation[feature] = {\n",
|
||
" 'Pearson_Correlation_with_Target': df[feature].corr(df[target_column], method='pearson'),\n",
|
||
" 'Spearman_Correlation_with_Target': df[feature].corr(df[target_column], method='spearman')\n",
|
||
" }\n",
|
||
" else:\n",
|
||
" overall_correlation[feature] = {\n",
|
||
" 'Pearson_Correlation_with_Target': np.nan,\n",
|
||
" 'Spearman_Correlation_with_Target': np.nan\n",
|
||
" }\n",
|
||
" overall_corr_df = pd.DataFrame.from_dict(overall_correlation, orient='index')\n",
|
||
"\n",
|
||
" print(\"\\n--- 因子与目标变量的整体相关性 ---\")\n",
|
||
" print(overall_corr_df)\n",
|
||
"\n",
|
||
" # --- 3. 因子之间的相关性矩阵 ---\n",
|
||
" # 在清理后的 df 上计算相关性\n",
|
||
" factor_correlation_matrix = df[feature_columns].corr(method='spearman') # 改回 Spearman\n",
|
||
"\n",
|
||
" print(\"\\n--- 因子之间的相关性矩阵 (Spearman) ---\") # 修正打印信息\n",
|
||
" print(factor_correlation_matrix)\n",
|
||
"\n",
|
||
" # --- 4. 日间 IC 和 ICIR ---\n",
|
||
" print(\"\\n--- 计算日间 IC (Spearman 相关性) 和 ICIR ---\")\n",
|
||
"\n",
|
||
" # 直接在清理后的 df 上计算每日 IC\n",
|
||
" if df.empty: # 理论上上面已经检查过，这里再检查一次更安全\n",
|
||
" daily_ic_series = pd.Series(dtype=float) # 空 Series\n",
|
||
" ic_stats = pd.DataFrame({\n",
|
||
" 'Mean_IC (Spearman)': np.nan, 'Std_Dev_IC': np.nan, 'ICIR': np.nan\n",
|
||
" }, index=feature_columns)\n",
|
||
" else:\n",
|
||
" daily_ic_series = df.groupby(trade_date_col).apply(\n",
|
||
" lambda day_group: {\n",
|
||
" feature: day_group[feature].corr(day_group[target_column], method='spearman')\n",
|
||
" for feature in feature_columns if day_group.shape[0] > 1 # 确保每日数据点多于1才能计算相关性\n",
|
||
" }\n",
|
||
" ).apply(pd.Series) # 将字典结果转换为 DataFrame\n",
|
||
"\n",
|
||
" # 计算 IC 的统计量\n",
|
||
" if not daily_ic_series.empty:\n",
|
||
" ic_mean = daily_ic_series.mean()\n",
|
||
" ic_std = daily_ic_series.std()\n",
|
||
" # 避免除以零\n",
|
||
" ic_ir = ic_mean / ic_std.replace(0, np.nan) # 使用 replace 0 为 NaN\n",
|
||
"\n",
|
||
" ic_stats = pd.DataFrame({\n",
|
||
" 'Mean_IC (Spearman)': ic_mean,\n",
|
||
" 'Std_Dev_IC': ic_std,\n",
|
||
" 'ICIR': ic_ir\n",
|
||
" })\n",
|
||
" print(\"\\n--- 日间 IC 和 ICIR (Spearman) ---\")\n",
|
||
" print(ic_stats)\n",
|
||
" else:\n",
|
||
" ic_stats = pd.DataFrame({\n",
|
||
" 'Mean_IC (Spearman)': np.nan, 'Std_Dev_IC': np.nan, 'ICIR': np.nan\n",
|
||
" }, index=feature_columns)\n",
|
||
"\n",
|
||
"\n",
|
||
" # --- 5. 因子在不同市值分位数上的平均 IC ---\n",
|
||
" print(f\"\\n--- 计算因子在 {mcap_bins} 个市值分位数上的平均 IC (Spearman) ---\")\n",
|
||
"\n",
|
||
" # 在清理后的 df 上计算每日市值分位数，直接添加到 df 中\n",
|
||
" # 使用 transform() 和 qcut() 在每个日期分组内计算分位数\n",
|
||
" # labels=False 返回整数 0 to mcap_bins-1\n",
|
||
" # duplicates='drop' 处理在某些日期股票数量少于 bins 导致分位数边缘重复的情况，会返回 NaN\n",
|
||
" # 添加一个临时列来存储分位数\n",
|
||
" mcap_bin_col_name = f'_mcap_bin_{mcap_bins}'\n",
|
||
" df[mcap_bin_col_name] = df.groupby(trade_date_col)[mcap_col].transform(\n",
|
||
" lambda x: pd.qcut(x, q=mcap_bins, labels=False, duplicates='drop') if len(x) >= mcap_bins else np.nan # 确保股票数量足够进行分位数划分\n",
|
||
" )\n",
|
||
"\n",
|
||
" # 过滤掉无法划分分位数 (NaN) 的行，进行分位数 IC 计算\n",
|
||
" # 创建一个临时 DataFrame df_binned_analysis\n",
|
||
" df_binned_analysis = df.dropna(subset=[mcap_bin_col_name]).copy()\n",
|
||
"\n",
|
||
" if df_binned_analysis.empty:\n",
|
||
" print(\"错误: 划分市值分位数后数据为空，无法计算分位数上的 IC。\")\n",
|
||
" avg_ic_by_bin = pd.DataFrame(index=range(mcap_bins), columns=feature_columns) # Placeholder\n",
|
||
" else:\n",
|
||
" # 按日期和市值分位数分组，计算每个分组内的因子与目标变量的截面相关性 (分位数IC)\n",
|
||
" binned_ic_by_day = df_binned_analysis.groupby([trade_date_col, mcap_bin_col_name]).apply(\n",
|
||
" lambda group: {\n",
|
||
" feature: group[feature].corr(group[target_column], method='spearman')\n",
|
||
" for feature in feature_columns if group.shape[0] > 1 # 确保分位数组内数据点多于1\n",
|
||
" }\n",
|
||
" ).apply(pd.Series) # 将嵌套结果转为 DataFrame\n",
|
||
"\n",
|
||
" # 对每个分位数组的每日 IC 求平均\n",
|
||
" # unstack(level=mcap_bin_col_name) 将 mcap_bin 作为列\n",
|
||
" # mean(axis=0) 对日期索引求平均\n",
|
||
" avg_ic_by_bin = binned_ic_by_day.unstack(level=mcap_bin_col_name).mean(axis=0).unstack()\n",
|
||
"\n",
|
||
" # 重命名索引和列，使表格更清晰\n",
|
||
" if not avg_ic_by_bin.empty:\n",
|
||
" # Index name will be the original column name used for grouping ('_mcap_bin_X')\n",
|
||
" # Rename the index name explicitly\n",
|
||
" avg_ic_by_bin.index.name = 'MarketCap_Bin'\n",
|
||
" avg_ic_by_bin.columns.name = 'Feature'\n",
|
||
" # 可以根据需要对分位数 bin 索引进行排序 (虽然 pd.qcut labels=False usually sorts)\n",
|
||
" avg_ic_by_bin = avg_ic_by_bin.sort_index()\n",
|
||
"\n",
|
||
" print(avg_ic_by_bin)\n",
|
||
"\n",
|
||
"\n",
|
||
" # --- 6. 汇总所有指标 ---\n",
|
||
" # 将基本统计、整体相关性、IC/ICIR 合并到一个 DataFrame\n",
|
||
" # 注意：合并时需要根据索引进行对齐 (因子名称)\n",
|
||
" summary_df = basic_stats\n",
|
||
" summary_df = summary_df.merge(overall_corr_df, left_index=True, right_index=True, how='left')\n",
|
||
" summary_df = summary_df.merge(ic_stats, left_index=True, right_index=True, how='left')\n",
|
||
"\n",
|
||
" # print(\"\\n--- 因子分析汇总报告 ---\")\n",
|
||
" # print(summary_df)\n",
|
||
"\n",
|
||
" # --- 清理临时列 'mcap_bin' ---\n",
|
||
" # 修正：在函数结束时从我们一直在操作的 df 副本中删除临时列\n",
|
||
" if mcap_bin_col_name in df.columns:\n",
|
||
" df.drop(columns=[mcap_bin_col_name], inplace=True)\n",
|
||
"\n",
|
||
"\n",
|
||
" return summary_df # 主要返回汇总报告，分位数IC单独打印\n",
|
||
"\n",
|
||
"# # 运行分析函数\n",
|
||
"# factor_analysis_report = analyze_factors(test_data.copy(), feature_columns, 'future_return')\n",
|
||
"\n",
|
||
"# print(\"\\n--- 最终汇总报告 DataFrame ---\")\n",
|
||
"# print(factor_analysis_report)"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 31,
|
||
"id": "a0000d75",
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"ename": "ModuleNotFoundError",
|
||
"evalue": "No module named 'seaborn'",
|
||
"output_type": "error",
|
||
"traceback": [
|
||
"\u001b[31m---------------------------------------------------------------------------\u001b[39m",
|
||
"\u001b[31mModuleNotFoundError\u001b[39m Traceback (most recent call last)",
|
||
"\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[31]\u001b[39m\u001b[32m, line 2\u001b[39m\n\u001b[32m 1\u001b[39m \u001b[38;5;28;01mimport\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34;01mmatplotlib\u001b[39;00m\u001b[34;01m.\u001b[39;00m\u001b[34;01mpyplot\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mas\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34;01mplt\u001b[39;00m\n\u001b[32m----> \u001b[39m\u001b[32m2\u001b[39m \u001b[38;5;28;01mimport\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34;01mseaborn\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mas\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34;01msns\u001b[39;00m\n\u001b[32m 3\u001b[39m \u001b[38;5;28;01mfrom\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34;01mscipy\u001b[39;00m\u001b[34;01m.\u001b[39;00m\u001b[34;01mstats\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mimport\u001b[39;00m spearmanr\n\u001b[32m 4\u001b[39m \u001b[38;5;28;01mfrom\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34;01mtqdm\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mimport\u001b[39;00m tqdm \u001b[38;5;66;03m# 用于显示进度条 (可选)\u001b[39;00m\n",
|
||
"\u001b[31mModuleNotFoundError\u001b[39m: No module named 'seaborn'"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"import matplotlib.pyplot as plt\n",
|
||
"import seaborn as sns\n",
|
||
"from scipy.stats import spearmanr\n",
|
||
"from tqdm import tqdm # 用于显示进度条 (可选)\n",
|
||
"\n",
|
||
"# 设置 Matplotlib/Seaborn 样式 (可选)\n",
|
||
"sns.set_theme(style=\"whitegrid\")\n",
|
||
"plt.rcParams['font.sans-serif'] = ['SimHei'] # 或者其他支持中文的字体\n",
|
||
"plt.rcParams['axes.unicode_minus'] = False # 解决负号显示问题\n",
|
||
"\n",
|
||
"def analyze_score_performance_2d(score_df: pd.DataFrame,\n",
|
||
" score_col: str = 'score',\n",
|
||
" label_col: str = 'label',\n",
|
||
" condition1_col: str = 'circ_mv',\n",
|
||
" condition2_col: str = 'future_return',\n",
|
||
" n_bins: int = 100,\n",
|
||
" min_samples_per_bin: int = 30): # 每个格子最少样本数\n",
|
||
" \"\"\"\n",
|
||
" 分析 score 在两个条件下 (如市值、未来收益) 的二维分箱表现。\n",
|
||
"\n",
|
||
" Args:\n",
|
||
" score_df (pd.DataFrame): 包含分数、标签和条件列的 DataFrame。\n",
|
||
" score_col (str): 预测分数所在的列名。\n",
|
||
" label_col (str): 目标标签所在的列名 (应为数值或可排序类别)。\n",
|
||
" condition1_col (str): 第一个条件列名 (例如 'circ_mv')。\n",
|
||
" condition2_col (str): 第二个条件列名 (例如 'future_return')。\n",
|
||
" n_bins (int): 每个条件划分的箱数 (分位数数量)。\n",
|
||
" min_samples_per_bin (int): 计算指标所需的最小样本数，小于此数目的格子结果将被屏蔽。\n",
|
||
"\n",
|
||
" Returns:\n",
|
||
" tuple: 包含 (performance_pivot, count_pivot, fig)\n",
|
||
" performance_pivot: 以二维分箱为索引/列的 Spearman 相关系数矩阵。\n",
|
||
" count_pivot: 每个二维分箱的样本数量矩阵。\n",
|
||
" fig: 生成的热力图 Matplotlib Figure 对象。\n",
|
||
" \"\"\"\n",
|
||
" print(f\"开始分析 '{score_col}' 在 '{condition1_col}' 和 '{condition2_col}' 下的表现...\")\n",
|
||
"\n",
|
||
" required_cols = [score_col, label_col, condition1_col, condition2_col]\n",
|
||
" if not all(col in score_df.columns for col in required_cols):\n",
|
||
" missing = [col for col in required_cols if col not in score_df.columns]\n",
|
||
" raise ValueError(f\"输入 DataFrame 缺少必需列: {missing}\")\n",
|
||
"\n",
|
||
" # --- 1. 数据准备和清洗 ---\n",
|
||
" print(\"准备数据，处理 NaN 值...\")\n",
|
||
" # 只保留需要的列，并移除包含 NaN 的行，避免影响分箱和计算\n",
|
||
" analysis_df = score_df[required_cols].dropna().copy()\n",
|
||
" n_original = len(score_df)\n",
|
||
" n_after_drop = len(analysis_df)\n",
|
||
" print(f\"原始数据 {n_original} 行，移除 NaN 后剩余 {n_after_drop} 行用于分析。\")\n",
|
||
"\n",
|
||
" if n_after_drop < min_samples_per_bin * n_bins: # 检查数据量是否过少\n",
|
||
" print(f\"警告: 清理 NaN 后数据量 ({n_after_drop}) 可能不足以支持 {n_bins}x{n_bins} 的精细分箱分析。\")\n",
|
||
" if n_after_drop < min_samples_per_bin:\n",
|
||
" print(\"错误: 有效数据过少，无法进行分析。\")\n",
|
||
" return None, None, None\n",
|
||
"\n",
|
||
" # --- 2. 二维分箱 ---\n",
|
||
" print(f\"对 '{condition1_col}' 和 '{condition2_col}' 进行 {n_bins} 分位数分箱...\")\n",
|
||
" bin1_col = f'{condition1_col}_bin'\n",
|
||
" bin2_col = f'{condition2_col}_bin'\n",
|
||
"\n",
|
||
" try:\n",
|
||
" # 使用 qcut 进行分位数分箱，labels=False 返回 0 到 n_bins-1 的整数标签\n",
|
||
" # duplicates='drop' 会丢弃导致边界不唯一的重复值所在的箱子，可能导致某些箱号缺失\n",
|
||
" # 对于可视化，这通常可以接受，但如果需要严格的等分，需先 rank\n",
|
||
" analysis_df[bin1_col] = pd.qcut(analysis_df[condition1_col], q=n_bins, labels=False, duplicates='drop')\n",
|
||
" analysis_df[bin2_col] = pd.qcut(analysis_df[condition2_col], q=n_bins, labels=False, duplicates='drop')\n",
|
||
" except Exception as e:\n",
|
||
" print(f\"错误: 分箱失败，请检查数据分布或减少 n_bins。错误信息: {e}\")\n",
|
||
" # 可以尝试先 rank 再 qcut\n",
|
||
" # analysis_df[bin1_col] = pd.qcut(analysis_df[condition1_col].rank(method='first'), q=n_bins, labels=False, duplicates='raise')\n",
|
||
" # analysis_df[bin2_col] = pd.qcut(analysis_df[condition2_col].rank(method='first'), q=n_bins, labels=False, duplicates='raise')\n",
|
||
" return None, None, None\n",
|
||
"\n",
|
||
" # --- 3. 分组计算表现指标 (Spearman Rank IC) ---\n",
|
||
" print(\"按二维分箱分组计算 Spearman Rank IC...\")\n",
|
||
"\n",
|
||
" def safe_spearmanr(x, y):\n",
|
||
" \"\"\"安全计算 Spearman 相关性，处理数据量过少的情况\"\"\"\n",
|
||
" if len(x) < max(2, min_samples_per_bin): # 要求至少有 min_samples_per_bin 个点才计算\n",
|
||
" return np.nan\n",
|
||
" corr, p_value = spearmanr(x, y)\n",
|
||
" return corr if not np.isnan(corr) else np.nan # 确保返回 NaN 而不是 None 或其他\n",
|
||
"\n",
|
||
" # 按两个分箱列分组\n",
|
||
" grouped = analysis_df.groupby([bin1_col, bin2_col])\n",
|
||
"\n",
|
||
" # 计算每个格子的 Spearman 相关系数\n",
|
||
" # apply 可能较慢，但计算相关性通常需要 apply\n",
|
||
" performance_series = grouped.apply(lambda sub: safe_spearmanr(sub[score_col], sub[label_col]))\n",
|
||
"\n",
|
||
" # 计算每个格子的样本数量\n",
|
||
" count_series = grouped.size()\n",
|
||
"\n",
|
||
" # --- 4. 结果整理成 Pivot Table (用于绘图) ---\n",
|
||
" print(\"整理结果用于绘图...\")\n",
|
||
" try:\n",
|
||
" # 将 performance_series 转换成二维矩阵\n",
|
||
" # index 为 condition1_bin, columns 为 condition2_bin\n",
|
||
" performance_pivot = performance_series.unstack(level=0) # level=0 对应第一个 groupby key (bin1_col)\n",
|
||
" count_pivot = count_series.unstack(level=0)\n",
|
||
"\n",
|
||
" # 可选：按列和索引排序，确保顺序正确\n",
|
||
" performance_pivot = performance_pivot.sort_index(axis=0).sort_index(axis=1)\n",
|
||
" count_pivot = count_pivot.sort_index(axis=0).sort_index(axis=1)\n",
|
||
" \n",
|
||
" print(performance_pivot)\n",
|
||
"\n",
|
||
" except Exception as e:\n",
|
||
" print(f\"错误: 无法将结果转换为二维矩阵，可能因为分箱不均匀或数据问题: {e}\")\n",
|
||
" return None, None, None\n",
|
||
"\n",
|
||
" # --- 5. 可视化：绘制热力图 ---\n",
|
||
" print(\"生成热力图...\")\n",
|
||
" fig, ax = plt.subplots(figsize=(16, 12)) # 调整图像大小\n",
|
||
"\n",
|
||
" # 使用 count_pivot 创建一个 mask，屏蔽掉样本量过小的格子\n",
|
||
" mask = count_pivot < min_samples_per_bin\n",
|
||
"\n",
|
||
" # 绘制热力图\n",
|
||
" sns.heatmap(performance_pivot,\n",
|
||
" annot=False, # 100x100 个格子加注释会太密集\n",
|
||
" fmt=\".2f\",\n",
|
||
" cmap=\"viridis\", # 选择颜色映射, 'viridis', 'coolwarm', 'RdYlGn' 等都不错\n",
|
||
" linewidths=.5,\n",
|
||
" linecolor='lightgray',\n",
|
||
" # mask=mask, # 应用 mask\n",
|
||
" ax=ax,\n",
|
||
" cbar_kws={'label': f'Spearman Rank IC ({score_col} vs {label_col})'}) # 颜色条标签\n",
|
||
"\n",
|
||
" # 设置标题和轴标签\n",
|
||
" ax.set_title(f'{score_col} 表现分析 (Rank IC vs {label_col})\\n基于 {condition1_col} 和 {condition2_col} {n_bins}x{n_bins} 分箱', fontsize=16)\n",
|
||
" ax.set_xlabel(f'{condition1_col} 分位数 (0 -> 高)', fontsize=12)\n",
|
||
" ax.set_ylabel(f'{condition2_col} 分位数 (0 -> 高)', fontsize=12)\n",
|
||
"\n",
|
||
" # 可选：调整刻度标签，避免显示所有 100 个刻度\n",
|
||
" if n_bins > 20:\n",
|
||
" tick_interval = n_bins // 10 # 大约显示 10 个刻度\n",
|
||
" ax.set_xticks(np.arange(0, n_bins, tick_interval) + 0.5)\n",
|
||
" ax.set_yticks(np.arange(0, n_bins, tick_interval) + 0.5)\n",
|
||
" ax.set_xticklabels(np.arange(0, n_bins, tick_interval))\n",
|
||
" ax.set_yticklabels(np.arange(0, n_bins, tick_interval))\n",
|
||
"\n",
|
||
" plt.xticks(rotation=45, ha='right')\n",
|
||
" plt.yticks(rotation=0)\n",
|
||
" plt.tight_layout() # 调整布局\n",
|
||
"\n",
|
||
" print(\"分析完成。\")\n",
|
||
" return performance_pivot, count_pivot, fig\n",
|
||
"\n",
|
||
"# --- 如何使用 ---\n",
|
||
"# 假设你的包含预测结果和所需列的 DataFrame 是 final_predictions_df\n",
|
||
"# 确保它包含 'score', 'label', 'circ_mv', 'future_return'\n",
|
||
"\n",
|
||
"# # 示例调用 (你需要有实际的 score_df)\n",
|
||
"try:\n",
|
||
" # 确保数据类型正确\n",
|
||
" cols_to_numeric = ['score', 'label', 'circ_mv', 'future_return']\n",
|
||
" for col in cols_to_numeric:\n",
|
||
" if col in score_df.columns:\n",
|
||
" score_df[col] = pd.to_numeric(score_df[col], errors='coerce')\n",
|
||
"\n",
|
||
" # 调用分析函数\n",
|
||
" performance_matrix, count_matrix, heatmap_figure = analyze_score_performance_2d(\n",
|
||
" score_df,\n",
|
||
" n_bins=100, # 你要求的100分箱\n",
|
||
" min_samples_per_bin=50 # 每个格子至少需要50个样本才显示IC，可以调整\n",
|
||
" )\n",
|
||
"\n",
|
||
" # 显示图像\n",
|
||
" if heatmap_figure:\n",
|
||
" plt.show()\n",
|
||
"\n",
|
||
" # 可以查看具体的 performance_matrix 和 count_matrix\n",
|
||
" # print(\"\\nPerformance Matrix (Spearman IC):\")\n",
|
||
" # print(performance_matrix)\n",
|
||
" # print(\"\\nCount Matrix:\")\n",
|
||
" # print(count_matrix)\n",
|
||
"\n",
|
||
"except ValueError as ve:\n",
|
||
" print(f\"数据错误: {ve}\")\n",
|
||
"except Exception as e:\n",
|
||
" print(f\"发生未知错误: {e}\")"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": null,
|
||
"id": "a436dba4",
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"ename": "AttributeError",
|
||
"evalue": "'CatBoostClassifier' object has no attribute 'feature_importance'",
|
||
"output_type": "error",
|
||
"traceback": [
|
||
"\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
|
||
"\u001b[1;31mAttributeError\u001b[0m Traceback (most recent call last)",
|
||
"Cell \u001b[1;32mIn[29], line 1\u001b[0m\n\u001b[1;32m----> 1\u001b[0m \u001b[38;5;28mprint\u001b[39m(model\u001b[38;5;241m.\u001b[39mfeature_importance)\n",
|
||
"\u001b[1;31mAttributeError\u001b[0m: 'CatBoostClassifier' object has no attribute 'feature_importance'"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"print(model.feature_importance())"
|
||
]
|
||
}
|
||
],
|
||
"metadata": {
|
||
"kernelspec": {
|
||
"display_name": "stock",
|
||
"language": "python",
|
||
"name": "python3"
|
||
},
|
||
"language_info": {
|
||
"codemirror_mode": {
|
||
"name": "ipython",
|
||
"version": 3
|
||
},
|
||
"file_extension": ".py",
|
||
"mimetype": "text/x-python",
|
||
"name": "python",
|
||
"nbconvert_exporter": "python",
|
||
"pygments_lexer": "ipython3",
|
||
"version": "3.13.2"
|
||
}
|
||
},
|
||
"nbformat": 4,
|
||
"nbformat_minor": 5
|
||
}
|