一般模型评估与特征可视化

eval.py

# coding:utf-8
import lightgbm as lgb
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from woe.eval import eval_segment_metrics
from sklearn.metrics import precision_recall_curve
from sklearn.metrics import roc_auc_score
import sys

# THERSHOLD = 0.49
# THERSHOLD = 0.693920
THRESHOLD = float(sys.argv[1])
segment_cnt = 50

def assign_grp_idx(df):
    total_sample_cnt = len(df)
    grp_idx = 1
    start_idx = 0
    proba_descend_idx = np.argsort(y_pred_prob)[::-1]
    segment_sample_cnt = int(len(y_pred_prob) / segment_cnt)
    while start_idx < total_sample_cnt:
        segment_idx_list = proba_descend_idx[start_idx: start_idx + segment_sample_cnt]
        segment_sample_cnt = len(segment_idx_list)
        df.loc[segment_idx_list, 'grp'] = grp_idx
        grp_idx += 1
        start_idx += segment_sample_cnt


def model_stat(train, test, df_opt, dump_report=False):
    res = pd.DataFrame()
    res['样本量'] = [len(train), len(test), len(df_opt), len(df_opt) / len(test), test['new_user'].sum(),
                  df_opt['new_user'].sum()]
    # res['样本量'] = [len(train), len(test), len(df_opt), len(df_opt) / len(test), len(test['uid'].unique()),
    #              len(df_opt['uid'].unique())]
    res['正样本量'] = [train['d7_open'].sum(), test['d7_open'].sum(), df_opt['d7_open'].sum(),
                   df_opt['d7_open'].sum() / test['d7_open'].sum(), None, None]
    res.index = ['训练数据', '测试数据', '测试门槛召回数据', '占比', '原始打开用户', '优化打开用户']
    if dump_report:
        res.to_excel('../../reports/model_stat.xlsx', index=True)
    return res


def plot_pr_curve(y_test, y_score):
    precision, recall, _ = precision_recall_curve(y_test, y_score)
    plt.step(recall, precision, color='b', alpha=0.2, where='post')
    plt.fill_between(recall, precision, step='post', alpha=0.2, color='b')
    plt.xlabel('Recall')
    plt.ylabel('Precision')
    plt.ylim([0.0, 0.0175])
    plt.xlim([0.0, 1.0])
    plt.savefig('../../reports/figures/PR-curve.png')


if __name__ == '__main__':
    print('THRESHOLD = %.4f' % THRESHOLD)
    train = pd.read_csv('../../data/processed/lx_new_users_algo2_nf_train.tsv', sep='\t')
    test = pd.read_csv('../../data/processed/lx_new_users_algo2_nf_test.tsv', sep='\t')
    feature_names = ['up_cnt', 'if_333', 'open_1d_fuids_cnt', 'interact_hb_14d_fuids_rate', 
        'interact_hb_30d_fuids_ra