如何高效WOE平衡DataFrame？

1. 导入必要的库和数据

import pandas as pd import numpy as np

df = pd.read_csv('filename.csv')

2. 计算每个变量的WOE值

def get_woe(data, target, var): lst = [] for i in range(data[var].nunique()): val = list(data[var].unique())[i] var_count = data[data[var] == val].count()[var] good_count = data[(data[var] == val) & (data[target] == 0)].count()[var] bad_count = data[(data[var] == val) & (data[target] == 1)].count()[var] if good_count == 0: good_count = 0.5 if bad_count == 0: bad_count = 0.5 woe = np.log((good_count / bad_count) / (data[target].mean() / (1 - data[target].mean()))) lst.append({'Value': val, 'Var': var, 'Count': var_count, 'Good': good_count, 'Bad': bad_count, 'WOE': woe}) woe_data = pd.DataFrame(lst) return woe_data

woe_dict = {'Var': [], 'WOE': []} for var in df.columns: if var != 'target': woe = get_woe(df, 'target', var) woe = woe[woe['Count'] > 0] woe['Var'] = var woe_dict['Var'].append(var) woe_dict['WOE'].append(woe['WOE'].sum()) woe_df = pd.DataFrame(woe_dict)

3. 计算每个变量的IV值

def get_iv(data, target, var): lst = [] for i in range(data[var].nunique()): val = list(data[var].unique())[i] var_count = data[data[var] == val].count()[var] good_count = data[(data[var] == val) & (data[target] == 0)].count()[var] bad_count = data[(data[var] == val) & (data[target] ==