easy-ml/feature/woe/strategy_woe.py

# -*- coding:utf-8 -*-
"""
@author: yq
@time: 2024/1/2
@desc: iv值及单调性筛选类
"""
import json
import os.path
from itertools import combinations_with_replacement
from typing import Dict, Optional, Union

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import scorecardpy as sc
import seaborn as sns
from pandas.core.dtypes.common import is_numeric_dtype
from tqdm import tqdm

from commom import f_display_images_by_side, NumpyEncoder, GeneralException, f_df_to_image, f_display_title
from entitys import DataSplitEntity, MetricFucResultEntity
from enums import ContextEnum, ResultCodesEnum
from feature.feature_strategy_base import FeatureStrategyBase
from init import context
from .entity import BinInfo, HomologousBinInfo
from .utils import f_monto_shift, f_get_corr, f_get_vif, f_format_bin, f_trend_shift, f_get_psi


class StrategyWoe(FeatureStrategyBase):

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        # woe编码需要的分箱信息，复用scorecardpy的格式
        self.sc_woebin = None

    def _f_get_img_corr(self, train_woe) -> Union[str, None]:
        if len(train_woe.columns.to_list()) <= 1:
            return None
        train_corr = f_get_corr(train_woe)
        plt.figure(figsize=(12, 12))
        sns.heatmap(train_corr, vmax=1, square=True, cmap='RdBu', annot=True)
        plt.title('Variables Correlation', fontsize=15)
        plt.yticks(rotation=0)
        plt.xticks(rotation=90)
        img_path = self.ml_config.f_get_save_path(f"corr.png")
        plt.savefig(img_path)
        return img_path

    def _f_get_img_trend(self, sc_woebin, x_columns, prefix):
        imgs_path = []
        for k in x_columns:
            df_bin = sc_woebin[k]
            # df_bin["bin"] = df_bin["bin"].apply(lambda x: re.sub(r"(\d+\.\d+)",
            #                                                      lambda m: "{:.2f}".format(float(m.group(0))), x))
            sc.woebin_plot(df_bin)
            path = self.ml_config.f_get_save_path(f"{prefix}_{k}.png")
            plt.savefig(path)
            imgs_path.append(path)
        return imgs_path

    def _f_get_sc_woebin(self, data: pd.DataFrame, bin_info_dict: Dict[str, BinInfo]) -> Dict[str, pd.DataFrame]:
        y_column = self.ml_config.y_column
        special_values = self.ml_config.special_values
        x_columns = list(bin_info_dict.keys())
        breaks_list = {column: bin_info.points for column, bin_info in bin_info_dict.items()}
        sc_woebin = sc.woebin(data[x_columns + [y_column]], y=y_column, breaks_list=breaks_list,
                              special_values=special_values, print_info=False)
        return sc_woebin

    def _handle_numeric(self, data: DataSplitEntity, x_column: str) -> HomologousBinInfo:
        # 贪婪搜索【训练集】及【测试集】加起来【iv】值最高的且【单调】的分箱
        def _n0(x):
            return sum(x == 0)

        def _n1(x):
            return sum(x == 1)

        def _get_bins_sv(df, x_column):
            y_column = self.ml_config.y_column
            special_values = self.ml_config.get_special_values(x_column)
            # special_values_bins
            bins_sv = pd.DataFrame()
            for special in special_values:
                dtm = df[df[x_column] == special]
                if len(dtm) != 0:
                    dtm['bin'] = [str(special)] * len(dtm)
                    bin = dtm.groupby(['bin'], group_keys=False)[y_column].agg([_n0, _n1]) \
                        .reset_index().rename(columns={'_n0': 'good', '_n1': 'bad'})
                    bin['is_special_values'] = [True] * len(bin)
                    bins_sv = pd.concat((bins_sv, bin))
            return bins_sv

        def _get_bins_nsv(df, x_column, breaks_list):
            # no_special_values_bins
            def _left_value(bin: str):
                if "," not in bin:
                    return float(bin)
                left = bin.split(",")[0]
                return float(left[1:])

            y_column = self.ml_config.y_column

            dtm = pd.DataFrame({'y': df[y_column], 'value': df[x_column]})
            bstbrks = [-np.inf] + breaks_list + [np.inf]
            labels = ['[{},{})'.format(bstbrks[i], bstbrks[i + 1]) for i in range(len(bstbrks) - 1)]
            dtm.loc[:, 'bin'] = pd.cut(dtm['value'], bstbrks, right=False, labels=labels)
            dtm['bin'] = dtm['bin'].astype(str)
            bins = dtm.groupby(['bin'], group_keys=False)['y'].agg([_n0, _n1]) \
                .reset_index().rename(columns={'_n0': 'good', '_n1': 'bad'})
            bins['is_special_values'] = [False] * len(bins)
            bins["ordered"] = bins['bin'].apply(_left_value)
            # 排序防止计算变量分箱后的单调性错位
            bins = bins.sort_values(by=["ordered"], ascending=[True])
            return bins

        def _get_badprobs(bins):
            bins['count'] = bins['good'] + bins['bad']
            bins['badprob'] = bins['bad'] / bins['count']
            return bins['badprob'].values.tolist()

        def _get_iv(bins):
            infovalue = pd.DataFrame({'good': bins['good'], 'bad': bins['bad']}) \
                .replace(0, 0.9) \
                .assign(DistrBad=lambda x: x.bad / sum(x.bad), DistrGood=lambda x: x.good / sum(x.good)) \
                .assign(iv=lambda x: (x.DistrBad - x.DistrGood) * np.log(x.DistrBad / x.DistrGood)) \
                .iv
            bins['bin_iv'] = infovalue
            bins['total_iv'] = bins['bin_iv'].sum()
            iv = bins['total_iv'].values[0]
            return iv.round(3)

        def _get_points(data_ascending, column):
            def _sampling(raw_list: list, num: int):
                # 按步长采样
                return raw_list[::int(len(raw_list) / num)]

            def _distribute(interval, bin_num):
                parts = int(1 / interval)
                # 穷举分布，隔板法
                total_ways = combinations_with_replacement(range(parts + bin_num - 1), bin_num - 1)
                distributions = []
                # 遍历所有可能的隔板位置
                for combo in total_ways:
                    # 根据隔板位置分配球
                    distribution = [0] * bin_num
                    start = 0
                    for i, divider in enumerate(combo):
                        distribution[i] = divider - start + 1
                        start = divider + 1
                    distribution[-1] = parts - start  # 最后一个箱子的球数
                    # 确保每个箱子至少有一个球
                    if all(x > 0 for x in distribution):
                        distributions.append(distribution)
                return distributions

            interval = self.ml_config.bin_search_interval
            bin_sample_rate = self.ml_config.bin_sample_rate
            format_bin = self.ml_config.format_bin

            data_x = data_ascending[column]
            data_x_describe = data_x.describe(percentiles=[0.1, 0.9])
            data_x_max = data_x.max()

            # 计算 2 - 5 箱的情况
            distributions_list = []
            for bin_num in list(range(2, 6)):
                distributions = _distribute(interval, bin_num)
                # 4箱及以上得采样，不然耗时太久
                sample_num = 1000 * bin_sample_rate
                if bin_sample_rate <= 0.15:
                    sample_num *= 2
                if bin_num == 5:
                    sample_num = 4000 * bin_sample_rate
                if bin_num in (4, 5) and len(distributions) >= sample_num:
                    distributions = _sampling(distributions, sample_num)
                distributions_list.extend(distributions)

            points_list = []
            for distributions in distributions_list:
                points = []
                point_percentile = [sum(distributions[0:idx + 1]) * interval for idx, _ in
                                    enumerate(distributions[0:-1])]
                for percentile in point_percentile:
                    point = data_x.iloc[int(len(data_x) * percentile)]
                    point = float(point)
                    if format_bin:
                        point = f_format_bin(data_x_describe, point)
                    point = round(point, 2)
                    if point == 0:
                        continue
                    # 排除粗分箱后越界的情况
                    if point not in points and point < data_x_max:
                        points.append(point)
                if points not in points_list and len(points) != 0:
                    points_list.append(points)
            return points_list

        special_values = self.ml_config.get_special_values(x_column)
        breaks_list = self.ml_config.get_breaks_list(x_column)
        iv_threshold = self.ml_config.iv_threshold
        psi_threshold = self.ml_config.psi_threshold
        monto_shift_threshold = self.ml_config.monto_shift_threshold
        trend_shift_threshold = self.ml_config.trend_shift_threshold

        train_data = data.train_data
        test_data = data.test_data

        train_data_ascending_nsv = train_data[~train_data[x_column].isin(special_values)] \
            .sort_values(by=x_column, ascending=True)
        test_data_ascending_nsv = test_data[~test_data[x_column].isin(special_values)] \
            .sort_values(by=x_column, ascending=True)

        train_bins_sv = _get_bins_sv(train_data, x_column)
        test_bins_sv = _get_bins_sv(test_data, x_column)

        # 获取每种分箱的信息
        # 构造数据切分点
        is_auto_bins = 1
        if len(breaks_list) != 0:
            points_list_nsv = [breaks_list]
            is_auto_bins = 0
        else:
            points_list_nsv = _get_points(train_data_ascending_nsv, x_column)
        homo_bin_info = HomologousBinInfo(x_column, is_auto_bins, self.ml_config.is_include(x_column))
        # 计算iv psi monto_shift等
        for points in points_list_nsv:
            bin_info = BinInfo()
            bin_info.x_column = x_column
            bin_info.bin_num = len(points) + 1
            bin_info.points = points
            bin_info.is_auto_bins = is_auto_bins

            # 变量iv，与special_values合并计算iv
            train_bins_nsv = _get_bins_nsv(train_data_ascending_nsv, x_column, points)
            train_bins = pd.concat((train_bins_nsv, train_bins_sv))
            train_iv = _get_iv(train_bins)

            test_bins_nsv = _get_bins_nsv(test_data_ascending_nsv, x_column, points)
            test_bins = pd.concat((test_bins_nsv, test_bins_sv))
            test_iv = _get_iv(test_bins)

            bin_info.train_iv = train_iv
            bin_info.test_iv = test_iv
            bin_info.iv = train_iv + test_iv
            bin_info.is_qualified_iv_train = 1 if train_iv > iv_threshold else 0

            # 变量单调性变化次数
            train_badprobs_nsv = _get_badprobs(train_bins_nsv)
            monto_shift_train_nsv = f_monto_shift(train_badprobs_nsv)
            bin_info.monto_shift_nsv = monto_shift_train_nsv
            bin_info.is_qualified_monto_train_nsv = 0 if monto_shift_train_nsv > monto_shift_threshold else 1

            # 变量趋势一致性
            test_badprobs_nsv = _get_badprobs(test_bins_nsv)
            trend_shift_nsv = f_trend_shift(train_badprobs_nsv, test_badprobs_nsv)
            bin_info.trend_shift_nsv = trend_shift_nsv
            bin_info.is_qualified_trend_nsv = 0 if trend_shift_nsv > trend_shift_threshold else 1

            # 变量psi
            psi = f_get_psi(train_bins, test_bins)
            bin_info.psi = psi
            bin_info.is_qualified_psi = 1 if psi < psi_threshold else 0
            homo_bin_info.add(bin_info)
        return homo_bin_info

    def _f_fast_filter(self, data: DataSplitEntity) -> Dict[str, BinInfo]:
        # 通过iv值粗筛变量
        train_data = data.train_data
        test_data = data.test_data
        y_column = self.ml_config.y_column
        x_columns = self.ml_config.x_columns
        columns_exclude = self.ml_config.columns_exclude
        special_values = self.ml_config.special_values
        breaks_list = self.ml_config.breaks_list.copy()
        iv_threshold = self.ml_config.iv_threshold
        psi_threshold = self.ml_config.psi_threshold

        if len(x_columns) == 0:
            x_columns = train_data.columns.tolist()
        if y_column in x_columns:
            x_columns.remove(y_column)
        for column in columns_exclude:
            if column in x_columns:
                x_columns.remove(column)

        bins_train = sc.woebin(train_data[x_columns + [y_column]], y=y_column, bin_num_limit=5,
                               special_values=special_values, breaks_list=breaks_list, print_info=False)

        for column, bin in bins_train.items():
            breaks_list[column] = list(bin[bin["is_special_values"] == False]['breaks'])

        bins_test = sc.woebin(test_data[x_columns + [y_column]], y=y_column,
                              special_values=special_values, breaks_list=breaks_list, print_info=False)
        bin_info_fast: Dict[str, BinInfo] = {}
        filter_fast_overview = ""
        for column, bin_train in bins_train.items():
            train_iv = bin_train['total_iv'][0].round(3)
            if train_iv <= iv_threshold and not self.ml_config.is_include(column):
                filter_fast_overview = f"{filter_fast_overview}{column} 因为train_iv【{train_iv}】小于阈值被剔除\n"
                continue
            bin_test = bins_test[column]
            test_iv = bin_test['total_iv'][0].round(3)
            iv = round(train_iv + test_iv, 3)
            psi = f_get_psi(bin_train, bin_test)
            # if psi >= psi_threshold and not self.ml_config.is_include(column):
            #     filter_fast_overview = f"{filter_fast_overview}{column} 因为psi【{psi}】大于阈值被剔除\n"
            #     continue
            bin_info_fast[column] = BinInfo.ofConvertByDict(
                {"x_column": column, "train_iv": train_iv, "iv": iv, "psi": psi, "points": breaks_list[column]}
            )

        context.set_filter_info(ContextEnum.FILTER_FAST,
                                f"筛选前变量数量:{len(x_columns)}\n{x_columns}\n"
                                f"快速筛选剔除变量数量:{len(x_columns) - len(bin_info_fast)}\n{filter_fast_overview}")
        return bin_info_fast

    def _f_corr_filter(self, data: DataSplitEntity, bin_info_dict: Dict[str, BinInfo]) -> Dict[str, BinInfo]:
        # 相关性剔除变量
        corr_threshold = self.ml_config.corr_threshold
        train_data = data.train_data

        x_columns = list(bin_info_dict.keys())
        sc_woebin = self._f_get_sc_woebin(train_data, bin_info_dict)
        train_woe = sc.woebin_ply(train_data[x_columns], sc_woebin, print_info=False)
        corr_df = f_get_corr(train_woe)
        corr_dict = corr_df.to_dict()
        filter_corr_overview = ""
        filter_corr_detail = {}
        # 依次判断每个变量对于其它变量的相关性
        for column, corr in corr_dict.items():
            column = column.replace("_woe", "")
            column_remove = []
            overview = f"{column}: "
            if column not in x_columns:
                continue
            for challenger_column, challenger_corr in corr.items():
                challenger_corr = round(challenger_corr, 3)
                challenger_column = challenger_column.replace("_woe", "")
                if challenger_corr < corr_threshold or column == challenger_column \
                        or challenger_column not in x_columns:
                    continue
                # 相关性大于阈值的情况，选择iv值大的
                iv = bin_info_dict[column].iv
                challenger_iv = bin_info_dict[challenger_column].iv
                if iv > challenger_iv:
                    if not self.ml_config.is_include(challenger_column):
                        column_remove.append(challenger_column)
                        overview = f"{overview}【{challenger_column}_iv{challenger_iv}_corr{challenger_corr}】 "
                else:
                    # 自己被剔除的情况下不再记录
                    column_remove = []
                    overview = ""
                    break
            # 剔除与自己相关的变量
            for c in column_remove:
                if c in x_columns:
                    x_columns.remove(c)
            if len(column_remove) != 0:
                filter_corr_overview = f"{filter_corr_overview}{overview}\n"
                filter_corr_detail[column] = column_remove
        for column in list(bin_info_dict.keys()):
            if column not in x_columns:
                bin_info_dict.pop(column)
        context.set_filter_info(ContextEnum.FILTER_CORR, filter_corr_overview, filter_corr_detail)
        return bin_info_dict

    def _f_vif_filter(self, data: DataSplitEntity, bin_info_dict: Dict[str, BinInfo]) -> Dict[str, BinInfo]:
        vif_threshold = self.ml_config.vif_threshold
        train_data = data.train_data

        x_columns = list(bin_info_dict.keys())
        sc_woebin = self._f_get_sc_woebin(train_data, bin_info_dict)
        train_woe = sc.woebin_ply(train_data[x_columns], sc_woebin, print_info=False)
        df_vif = f_get_vif(train_woe)
        if df_vif is None:
            return bin_info_dict

        filter_vif_overview = ""
        filter_vif_detail = []
        for _, row in df_vif.iterrows():
            column = row["变量"]
            vif = row["vif"]
            if vif < vif_threshold or self.ml_config.is_include(column):
                continue
            filter_vif_overview = f"{filter_vif_overview}{column} 因为vif【{vif}】大于阈值被剔除\n"
            filter_vif_detail.append(column)
            bin_info_dict.pop(column)

        context.set_filter_info(ContextEnum.FILTER_VIF, filter_vif_overview, filter_vif_detail)
        return bin_info_dict

    def post_filter(self, data: DataSplitEntity, bin_info_dict: Dict[str, BinInfo]):
        # 变量之间进行比较的过滤器
        max_feature_num = self.ml_config.max_feature_num
        bin_info_filtered = self._f_corr_filter(data, bin_info_dict)
        bin_info_filtered = self._f_vif_filter(data, bin_info_filtered)
        bin_info_filtered = BinInfo.ivTopN(bin_info_filtered, max_feature_num)
        self.sc_woebin = self._f_get_sc_woebin(data.train_data, bin_info_filtered)
        context.set(ContextEnum.BIN_INFO_FILTERED, bin_info_filtered)
        context.set(ContextEnum.WOEBIN, self.sc_woebin)

    def feature_search(self, data: DataSplitEntity, *args, **kwargs):
        # 粗筛
        bin_info_fast = self._f_fast_filter(data)
        x_columns = list(bin_info_fast.keys())

        bin_info_filtered: Dict[str, BinInfo] = {}
        # 数值型变量多种分箱方式的中间结果
        homo_bin_info_numeric_set: Dict[str, HomologousBinInfo] = {}
        filter_numeric_overview = ""
        filter_numeric_detail = []
        for x_column in tqdm(x_columns):
            if is_numeric_dtype(data.train_data[x_column]):
                # 数值型变量筛选
                homo_bin_info_numeric: HomologousBinInfo = self._handle_numeric(data, x_column)
                if homo_bin_info_numeric.is_auto_bins:
                    homo_bin_info_numeric_set[x_column] = homo_bin_info_numeric
                # iv psi 变量单调性 变量趋势一致性 筛选
                bin_info: Optional[BinInfo] = homo_bin_info_numeric.filter()
                if bin_info is not None:
                    bin_info_filtered[x_column] = bin_info
                else:
                    # 不满足要求被剔除
                    filter_numeric_overview = f"{filter_numeric_overview}{x_column} {homo_bin_info_numeric.drop_reason()}\n"
                    filter_numeric_detail.append(x_column)
            else:
                # 字符型暂时用scorecardpy来处理
                bin_info_filtered[x_column] = bin_info_fast[x_column]

        self.post_filter(data, bin_info_filtered)

        context.set(ContextEnum.HOMO_BIN_INFO_NUMERIC_SET, homo_bin_info_numeric_set)
        context.set_filter_info(ContextEnum.FILTER_NUMERIC, filter_numeric_overview, filter_numeric_detail)

    def feature_save(self, *args, **kwargs):
        if self.sc_woebin is None:
            GeneralException(ResultCodesEnum.NOT_FOUND, message=f"feature不存在")
        df_woebin = pd.concat(self.sc_woebin.values())
        path = self.ml_config.f_get_save_path(f"feature.csv")
        df_woebin.to_csv(path)
        print(f"feature save to【{path}】success. ")

    def feature_load(self, path: str, *args, **kwargs):
        if os.path.isdir(path):
            path = os.path.join(path, "feature.csv")
        if not os.path.isfile(path) or "feature.csv" not in path:
            raise GeneralException(ResultCodesEnum.NOT_FOUND, message=f"特征信息【feature.csv】不存在")

        df_woebin = pd.read_csv(path)
        variables = df_woebin["variable"].unique().tolist()
        self.sc_woebin = {}
        for variable in variables:
            self.sc_woebin[variable] = df_woebin[df_woebin["variable"] == variable]
        print(f"feature load from【{path}】success.")

    def feature_generate(self, data: pd.DataFrame, *args, **kwargs) -> pd.DataFrame:
        x_columns = list(self.sc_woebin.keys())
        # 排个序，防止因为顺序原因导致的可能的bug
        x_columns.sort()
        data_woe = sc.woebin_ply(data[x_columns], self.sc_woebin, print_info=False)
        return data_woe

    def feature_report(self, data: DataSplitEntity, *args, **kwargs) -> Dict[str, MetricFucResultEntity]:
        y_column = self.ml_config.y_column
        columns_anns = self.ml_config.columns_anns
        x_columns = list(self.sc_woebin.keys())

        train_data = data.train_data
        test_data = data.test_data
        # 跨模块调用中间结果，所以从上下文里取
        bin_info_filtered: Dict[str, BinInfo] = context.get(ContextEnum.BIN_INFO_FILTERED)

        metric_value_dict = {}
        # 样本分布
        metric_value_dict["样本分布"] = MetricFucResultEntity(table=data.get_distribution(y_column), table_font_size=10,
                                                          table_cell_width=3)

        # 变量相关性
        sc_woebin_train = self._f_get_sc_woebin(train_data, bin_info_filtered)
        train_woe = sc.woebin_ply(train_data[x_columns], sc_woebin_train, print_info=False)
        img_path_corr = self._f_get_img_corr(train_woe)
        metric_value_dict["变量相关性"] = MetricFucResultEntity(image_path=img_path_corr)

        # 变量iv、psi、vif
        df_iv_psi_vif = pd.DataFrame()
        train_iv = [bin_info_filtered[column].train_iv for column in x_columns]
        psi = [bin_info_filtered[column].psi for column in x_columns]
        anns = [columns_anns.get(column, "-") for column in x_columns]
        df_iv_psi_vif["变量"] = x_columns
        df_iv_psi_vif["iv"] = train_iv
        df_iv_psi_vif["psi"] = psi

        df_vif = f_get_vif(train_woe)
        if df_vif is not None:
            df_iv_psi_vif = pd.merge(df_iv_psi_vif, df_vif, on="变量", how="left")

        df_iv_psi_vif["释义"] = anns
        df_iv_psi_vif.sort_values(by=["iv"], ascending=[False], inplace=True)
        img_path_iv = self.ml_config.f_get_save_path(f"iv.png")
        f_df_to_image(df_iv_psi_vif, img_path_iv)
        metric_value_dict["变量iv"] = MetricFucResultEntity(table=df_iv_psi_vif, image_path=img_path_iv)

        # 变量趋势-训练集
        imgs_path_trend_train = self._f_get_img_trend(sc_woebin_train, x_columns, "train")
        metric_value_dict["变量趋势-训练集"] = MetricFucResultEntity(image_path=imgs_path_trend_train, image_size=4)

        # 变量趋势-测试集
        sc_woebin_test = self._f_get_sc_woebin(test_data, bin_info_filtered)
        imgs_path_trend_test = self._f_get_img_trend(sc_woebin_test, x_columns, "test")
        metric_value_dict["变量趋势-测试集"] = MetricFucResultEntity(image_path=imgs_path_trend_test, image_size=4)

        # context.set(ContextEnum.METRIC_FEATURE.value, metric_value_dict)

        if self.ml_config.jupyter_print:
            self.jupyter_print(data, metric_value_dict)

        return metric_value_dict

    def jupyter_print(self, data: DataSplitEntity, metric_value_dict=Dict[str, MetricFucResultEntity]):
        from IPython import display

        def detail_print(detail):
            if isinstance(detail, str):
                detail = [detail]

            if isinstance(detail, list):
                for column in detail:
                    homo_bin_info_numeric = homo_bin_info_numeric_set.get(column)
                    if homo_bin_info_numeric is None:
                        continue
                    bins_info = homo_bin_info_numeric.get_best_bins()
                    print(f"-----【{column}】不同分箱数下变量的推荐切分点-----")
                    imgs_path_trend_train = []
                    imgs_path_trend_test = []
                    for bin_info in bins_info:
                        print(json.dumps(bin_info.points, ensure_ascii=False, cls=NumpyEncoder))
                        breaks_list = [str(i) for i in bin_info.points]
                        sc_woebin_train = self._f_get_sc_woebin(train_data, {column: bin_info})
                        image_path = self._f_get_img_trend(sc_woebin_train, [column],
                                                           f"train_{column}_{'_'.join(breaks_list)}")
                        imgs_path_trend_train.append(image_path[0])
                        sc_woebin_test = self._f_get_sc_woebin(test_data, {column: bin_info})
                        image_path = self._f_get_img_trend(sc_woebin_test, [column],
                                                           f"test_{column}_{'_'.join(breaks_list)}")
                        imgs_path_trend_test.append(image_path[0])
                    f_display_images_by_side(display, imgs_path_trend_train, title=f"训练集",
                                             image_path_list2=imgs_path_trend_test, title2="测试集")
            if isinstance(detail, dict):
                for column, challenger_columns in detail.items():
                    print(f"-----相关性筛选保留的【{column}】-----")
                    detail_print(column)
                    detail_print(challenger_columns)

        def filter_print(filter, title, notes=""):
            f_display_title(display, title)
            print(notes)
            print(filter.get("overview"))
            detail = filter.get("detail")
            if detail is not None and self.ml_config.bin_detail_print:
                detail_print(detail)

        train_data = data.train_data
        test_data = data.test_data

        bin_info_filtered: Dict[str, BinInfo] = context.get(ContextEnum.BIN_INFO_FILTERED)
        homo_bin_info_numeric_set: Dict[str, HomologousBinInfo] = context.get(ContextEnum.HOMO_BIN_INFO_NUMERIC_SET)
        filter_fast = context.get(ContextEnum.FILTER_FAST)
        filter_numeric = context.get(ContextEnum.FILTER_NUMERIC)
        filter_corr = context.get(ContextEnum.FILTER_CORR)
        filter_vif = context.get(ContextEnum.FILTER_VIF)
        filter_ivtop = context.get(ContextEnum.FILTER_IVTOP)

        f_display_title(display, "样本分布")
        display.display(metric_value_dict["样本分布"].table)

        # 打印变量iv
        f_display_title(display, "变量iv")
        display.display(metric_value_dict["变量iv"].table)
        # 打印变量相关性
        f_display_images_by_side(display, metric_value_dict["变量相关性"].image_path, width=800)

        # 打印变量趋势
        f_display_title(display, "变量趋势")
        imgs_path_trend_train = metric_value_dict["变量趋势-训练集"].image_path
        imgs_path_trend_test = metric_value_dict.get("变量趋势-测试集").image_path
        f_display_images_by_side(display, imgs_path_trend_train, title="训练集", image_path_list2=imgs_path_trend_test,
                                 title2="测试集")

        # 打印breaks_list
        breaks_list = {column: bin_info.points for column, bin_info in bin_info_filtered.items()}
        print("变量切分点:")
        print(json.dumps(breaks_list, ensure_ascii=False, indent=2, cls=NumpyEncoder))
        print("选中变量不同分箱数下变量的推荐切分点:")
        detail_print(list(bin_info_filtered.keys()))

        # 打印fast_filter筛选情况
        filter_print(filter_fast, "快速筛选过程", "剔除train_iv小于阈值")
        filter_print(filter_numeric, "数值变量筛选过程")
        filter_print(filter_corr, "相关性筛选过程")
        filter_print(filter_vif, "vif筛选过程")
        filter_print(filter_ivtop, "ivtop筛选过程", "iv = train_iv + test_iv")