1. 代碼

• 基于jupyter notebook

#導包
import numpy as np
import math
import pandas as pd
from sklearn.utils.multiclass import type_of_target
from scipy import stats

#求woe值和iv值
def woe(X, y, event):
    res_woe = []  #列表存放woe字典
    res_iv = []  #列表存放iv
    X1 = feature_discretion(X) #對連續(xù)型特征進行處理
    for i in range(0, X1.shape[-1]):  #遍歷所有特征
        x = X1[:, i]  #單個特征
        woe_dict, iv1 = woe_single_x(x, y, event)  #計算單個特征的woe值
        res_woe.append(woe_dict)
        res_iv.append(iv1)
    return np.array(res_woe), np.array(res_iv)  #返回數(shù)組

#求單個特征的woe值
def woe_single_x(x, y, event):
    event_total, non_event_total = count_binary(y, event) #計算好人壞人總數(shù)
    x_labels = np.unique(x) #特征中的分段
    woe_dict = {}  #存放每個分段的名稱 以及 其對應(yīng)的woe值
    iv = 0
    for x1 in x_labels: #遍歷每個分段
        y1 = y[np.where(x == x1)[0]]
        event_count, non_event_count = count_binary(y1, event=event)
        rate_event = 1.0 * event_count / event_total
        rate_non_event = 1.0 * non_event_count / non_event_total
        #woe無窮大時處理
        if rate_event == 0:
            print()#print("{'",x1,"'}"+":全是好人") #只輸出不做處理
        elif rate_non_event == 0:
            print()#print("{'",x1,"'}"+":全是壞人")
        else:
            woe1 = math.log(rate_event / rate_non_event)
            woe_dict[x1] = woe1  
            iv += (rate_event - rate_non_event) * woe1
    return woe_dict, iv

#計算個數(shù)
def count_binary(a, event):
    event_count = (a == event).sum()
    non_event_count = a.shape[-1] - event_count
    return event_count, non_event_count

#判斷特征數(shù)據(jù)是否為離散型
def feature_discretion(X):
    temp = []
    for i in range(0, X.shape[-1]):
        x = X[:, i]
        x_type = type_of_target(x)
        if pd.Series(list(x)).dtype != 'O':
            x1 = discrete(x)
            temp.append(x1)
        else:
            temp.append(x)
    return np.array(temp).T

#對連續(xù)型特征進行離散化
def discrete(x):
    res = np.array([0] * x.shape[-1], dtype=int)
    for i in range(5):
        point1 = stats.scoreatpercentile(x, i * 20)
        point2 = stats.scoreatpercentile(x, (i + 1) * 20)
        x1 = x[np.where((x >= point1) & (x <= point2))]
        mask = np.in1d(x, x1)
        res[mask] = (i + 1)
    return res

2. 數(shù)據(jù)

• 讀取數(shù)據(jù)
  df = pd.read_csv("telephone_test.csv")
• 目標變量
  y=df['status'].values
• 所有特征
  X=df.drop(['status','sid', 'uid'],axis=1).values

3. 結(jié)果

• 將特征的IV值和特征構(gòu)建成字典并按IV值對特征進行排序
  dic = dict(zip(a,b)) #a為IV值列表彤蔽，b為特征列表
  dic_sort= sorted(dic.items(),key = lambda x:x[1],reverse = True)

• 部分截圖