python之评分卡学习

之前一直没有实践评分卡模型,今天从网上看到有相关的代码,先保存下,后续需要再细看。

https://blog.csdn.net/axy_shelly/article/details/83274534?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-1.channel_param&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-1.channel_param(如下代码参考来源)

https://blog.csdn.net/LuYi_WeiLin/article/details/85060190?utm_medium=distribute.pc_relevant_download.none-task-blog-searchfrombaidu-4.nonecase&depth_1-utm_source=distribute.pc_relevant_download.none-task-blog-searchfrombaidu-4.nonecas(另一篇参考)

import pandas as pd

import datetime

import collections

import numpy as np

import numbers

import random

import sys

import pickle

from itertools import combinations

from sklearn.linear_model import LinearRegression

from sklearn.ensemble import RandomForestClassifier

from sklearn.model_selection import train_test_split

from sklearn.metrics import roc_curve

from sklearn.metrics import roc_auc_score

import statsmodels.api as sm

from importlib import reload

from matplotlib import pyplot as plt

reload(sys)

sys.setdefaultencoding( "utf-8")

from scorecard_functions import *

from sklearn.linear_model import LogisticRegressionCV

# -*- coding: utf-8 -*-


################################

######## UDF: 自定义函数 ########

################################

### 对时间窗口,计算累计产比 ###

def TimeWindowSelection(df, daysCol, time_windows):

    '''

    :param df: the dataset containg variabel of days

    :param daysCol: the column of days

    :param time_windows: the list of time window

    :return:

    '''

    freq_tw = {}

    for tw in time_windows:

        freq = sum(df[daysCol].apply(lambda x: int(x<=tw)))

        freq_tw[tw] = freq

    return freq_tw



def DeivdedByZero(nominator, denominator):

    '''

    当分母为0时,返回0;否则返回正常值

    '''

    if denominator == 0:

        return 0

    else:

        return nominator*1.0/denominator



#对某些统一的字段进行统一

def ChangeContent(x):

    y = x.upper()

    if y == '_MOBILEPHONE':

        y = '_PHONE'

    return y


def MissingCategorial(df,x):

    missing_vals = df[x].map(lambda x: int(x!=x))

    return sum(missing_vals)*1.0/df.shape[0]


def MissingContinuous(df,x):

    missing_vals = df[x].map(lambda x: int(np.isnan(x)))

    return sum(missing_vals) * 1.0 / df.shape[0]


def MakeupRandom(x, sampledList):

    if x==x:

        return x

    else:

        randIndex = random.randint(0, len(sampledList)-1)

        return sampledList[randIndex]




############################################################

#Step 0: 数据分析的初始工作, 包括读取数据文件、检查用户Id的一致性等#

############################################################


folderOfData = '/Users/Code/Data Collections/bank default/'

data1 = pd.read_csv(folderOfData+'PPD_LogInfo_3_1_Training_Set.csv', header = 0)

data2 = pd.read_csv(folderOfData+'PPD_Training_Master_GBK_3_1_Training_Set.csv', header = 0,encoding = 'gbk')

data3 = pd.read_csv(folderOfData+'PPD_Userupdate_Info_3_1_Training_Set.csv', header = 0)


#############################################################################################

# Step 1: 从PPD_LogInfo_3_1_Training_Set &  PPD_Userupdate_Info_3_1_Training_Set数据中衍生特征#

#############################################################################################

# compare whether the four city variables match

data2['city_match'] = data2.apply(lambda x: int(x.UserInfo_2 == x.UserInfo_4 == x.UserInfo_8 == x.UserInfo_20),axis = 1)

del data2['UserInfo_2']

del data2['UserInfo_4']

del data2['UserInfo_8']

del data2['UserInfo_20']


### 提取申请日期,计算日期差,查看日期差的分布

data1['logInfo'] = data1['LogInfo3'].map(lambda x: datetime.datetime.strptime(x,'%Y-%m-%d'))

data1['Listinginfo'] = data1['Listinginfo1'].map(lambda x: datetime.datetime.strptime(x,'%Y-%m-%d'))

data1['ListingGap'] = data1[['logInfo','Listinginfo']].apply(lambda x: (x[1]-x[0]).days,axis = 1)

plt.hist(data1['ListingGap'],bins=200)

plt.title('Days between login date and listing date')

ListingGap2 = data1['ListingGap'].map(lambda x: min(x,365))

plt.hist(ListingGap2,bins=200)


timeWindows = TimeWindowSelection(data1, 'ListingGap', range(30,361,30))


'''

使用180天作为最大的时间窗口计算新特征

所有可以使用的时间窗口可以有7 days, 30 days, 60 days, 90 days, 120 days, 150 days and 180 days.

在每个时间窗口内,计算总的登录次数,不同的登录方式,以及每种登录方式的平均次数

'''

time_window = [7, 30, 60, 90, 120, 150, 180]

var_list = ['LogInfo1','LogInfo2']

data1GroupbyIdx = pd.DataFrame({'Idx':data1['Idx'].drop_duplicates()})


for tw in time_window:

    data1['TruncatedLogInfo'] = data1['Listinginfo'].map(lambda x: x + datetime.timedelta(-tw))

    temp = data1.loc[data1['logInfo'] >= data1['TruncatedLogInfo']]

    for var in var_list:

        #count the frequences of LogInfo1 and LogInfo2

        count_stats = temp.groupby(['Idx'])[var].count().to_dict()

        data1GroupbyIdx[str(var)+'_'+str(tw)+'_count'] = data1GroupbyIdx['Idx'].map(lambda x: count_stats.get(x,0))


        # count the distinct value of LogInfo1 and LogInfo2

        Idx_UserupdateInfo1 = temp[['Idx', var]].drop_duplicates()

        uniq_stats = Idx_UserupdateInfo1.groupby(['Idx'])[var].count().to_dict()

        data1GroupbyIdx[str(var) + '_' + str(tw) + '_unique'] = data1GroupbyIdx['Idx'].map(lambda x: uniq_stats.get(x,0))


        # calculate the average count of each value in LogInfo1 and LogInfo2

        data1GroupbyIdx[str(var) + '_' + str(tw) + '_avg_count'] = data1GroupbyIdx[[str(var)+'_'+str(tw)+'_count',str(var) + '_' + str(tw) + '_unique']].\

            apply(lambda x: DeivdedByZero(x[0],x[1]), axis=1)



data3['ListingInfo'] = data3['ListingInfo1'].map(lambda x: datetime.datetime.strptime(x,'%Y/%m/%d'))

data3['UserupdateInfo'] = data3['UserupdateInfo2'].map(lambda x: datetime.datetime.strptime(x,'%Y/%m/%d'))

data3['ListingGap'] = data3[['UserupdateInfo','ListingInfo']].apply(lambda x: (x[1]-x[0]).days,axis = 1)

collections.Counter(data3['ListingGap'])

hist_ListingGap = np.histogram(data3['ListingGap'])

hist_ListingGap = pd.DataFrame({'Freq':hist_ListingGap[0],'gap':hist_ListingGap[1][1:]})

hist_ListingGap['CumFreq'] = hist_ListingGap['Freq'].cumsum()

hist_ListingGap['CumPercent'] = hist_ListingGap['CumFreq'].map(lambda x: x*1.0/hist_ListingGap.iloc[-1]['CumFreq'])


'''

对 QQ和qQ, Idnumber和idNumber,MOBILEPHONE和PHONE 进行统一

在时间切片内,计算

 (1)更新的频率

 (2)每种更新对象的种类个数

 (3)对重要信息如IDNUMBER,HASBUYCAR, MARRIAGESTATUSID, PHONE的更新

'''

data3['UserupdateInfo1'] = data3['UserupdateInfo1'].map(ChangeContent)

data3GroupbyIdx = pd.DataFrame({'Idx':data3['Idx'].drop_duplicates()})


time_window = [7, 30, 60, 90, 120, 150, 180]

for tw in time_window:

    data3['TruncatedLogInfo'] = data3['ListingInfo'].map(lambda x: x + datetime.timedelta(-tw))

    temp = data3.loc[data3['UserupdateInfo'] >= data3['TruncatedLogInfo']]


    #frequency of updating

    freq_stats = temp.groupby(['Idx'])['UserupdateInfo1'].count().to_dict()

    data3GroupbyIdx['UserupdateInfo_'+str(tw)+'_freq'] = data3GroupbyIdx['Idx'].map(lambda x: freq_stats.get(x,0))


    # number of updated types

    Idx_UserupdateInfo1 = temp[['Idx','UserupdateInfo1']].drop_duplicates()

    uniq_stats = Idx_UserupdateInfo1.groupby(['Idx'])['UserupdateInfo1'].count().to_dict()

    data3GroupbyIdx['UserupdateInfo_' + str(tw) + '_unique'] = data3GroupbyIdx['Idx'].map(lambda x: uniq_stats.get(x, x))


    #average count of each type

    data3GroupbyIdx['UserupdateInfo_' + str(tw) + '_avg_count'] = data3GroupbyIdx[['UserupdateInfo_'+str(tw)+'_freq', 'UserupdateInfo_' + str(tw) + '_unique']]. \

        apply(lambda x: x[0] * 1.0 / x[1], axis=1)


    #whether the applicant changed items like IDNUMBER,HASBUYCAR, MARRIAGESTATUSID, PHONE

    Idx_UserupdateInfo1['UserupdateInfo1'] = Idx_UserupdateInfo1['UserupdateInfo1'].map(lambda x: [x])

    Idx_UserupdateInfo1_V2 = Idx_UserupdateInfo1.groupby(['Idx'])['UserupdateInfo1'].sum()

    for item in ['_IDNUMBER','_HASBUYCAR','_MARRIAGESTATUSID','_PHONE']:

        item_dict = Idx_UserupdateInfo1_V2.map(lambda x: int(item in x)).to_dict()

        data3GroupbyIdx['UserupdateInfo_' + str(tw) + str(item)] = data3GroupbyIdx['Idx'].map(lambda x: item_dict.get(x, x))


# Combine the above features with raw features in PPD_Training_Master_GBK_3_1_Training_Set

allData = pd.concat([data2.set_index('Idx'), data3GroupbyIdx.set_index('Idx'), data1GroupbyIdx.set_index('Idx')],axis= 1)

allData.to_csv(folderOfData+'allData_0.csv',encoding = 'gbk')





#######################################

# Step 2: 对类别型变量和数值型变量进行补缺#

######################################

allData = pd.read_csv(folderOfData+'allData_0.csv',header = 0,encoding = 'gbk')

allFeatures = list(allData.columns)

allFeatures.remove('target')

if 'Idx' in allFeatures:

    allFeatures.remove('Idx')

allFeatures.remove('ListingInfo')


#检查是否有常数型变量,并且检查是类别型还是数值型变量

numerical_var = []

for col in allFeatures:

    if len(set(allData[col])) == 1:

        print('delete {} from the dataset because it is a constant'.format(col))

        del allData[col]

        allFeatures.remove(col)

    else:

        uniq_valid_vals = [i for i in allData[col] if i == i]

        uniq_valid_vals = list(set(uniq_valid_vals))

        if len(uniq_valid_vals) >= 10 and isinstance(uniq_valid_vals[0], numbers.Real):

            numerical_var.append(col)


categorical_var = [i for i in allFeatures if i not in numerical_var]



#检查变量的最多值的占比情况,以及每个变量中占比最大的值

records_count = allData.shape[0]

col_most_values,col_large_value = {},{}

for col in allFeatures:

    value_count = allData[col].groupby(allData[col]).count()

    col_most_values[col] = max(value_count)/records_count

    large_value = value_count[value_count== max(value_count)].index[0]

    col_large_value[col] = large_value

col_most_values_df = pd.DataFrame.from_dict(col_most_values, orient = 'index')

col_most_values_df.columns = ['max percent']

col_most_values_df = col_most_values_df.sort_values(by = 'max percent', ascending = False)

pcnt = list(col_most_values_df[:500]['max percent'])

vars = list(col_most_values_df[:500].index)

plt.bar(range(len(pcnt)), height = pcnt)

plt.title('Largest Percentage of Single Value in Each Variable')


#计算多数值占比超过90%的字段中,少数值的坏样本率是否会显著高于多数值

large_percent_cols = list(col_most_values_df[col_most_values_df['max percent']>=0.9].index)

bad_rate_diff = {}

for col in large_percent_cols:

    large_value = col_large_value[col]

    temp = allData[[col,'target']]

    temp[col] = temp.apply(lambda x: int(x[col]==large_value),axis=1)

    bad_rate = temp.groupby(col).mean()

    if bad_rate.iloc[0]['target'] == 0:

        bad_rate_diff[col] = 0

        continue

    bad_rate_diff[col] = np.log(bad_rate.iloc[0]['target']/bad_rate.iloc[1]['target'])

bad_rate_diff_sorted = sorted(bad_rate_diff.items(),key=lambda x: x[1], reverse=True)

bad_rate_diff_sorted_values = [x[1] for x in bad_rate_diff_sorted]

plt.bar(x = range(len(bad_rate_diff_sorted_values)), height = bad_rate_diff_sorted_values)


#由于所有的少数值的坏样本率并没有显著高于多数值,意味着这些变量可以直接剔除

for col in large_percent_cols:

    if col in numerical_var:

        numerical_var.remove(col)

    else:

        categorical_var.remove(col)

    del allData[col]


'''

对类别型变量,如果缺失超过80%, 就删除,否则当成特殊的状态

'''

missing_pcnt_threshould_1 = 0.8

for col in categorical_var:

    missingRate = MissingCategorial(allData,col)

    print('{0} has missing rate as {1}'.format(col,missingRate))

    if missingRate > missing_pcnt_threshould_1:

        categorical_var.remove(col)

        del allData[col]

    if 0 < missingRate < missing_pcnt_threshould_1:

        uniq_valid_vals = [i for i in allData[col] if i == i]

        uniq_valid_vals = list(set(uniq_valid_vals))

        if isinstance(uniq_valid_vals[0], numbers.Real):

            missing_position = allData.loc[allData[col] != allData[col]][col].index

            not_missing_sample = [-1]*len(missing_position)

            allData.loc[missing_position, col] = not_missing_sample

        else:

            # In this way we convert NaN to NAN, which is a string instead of np.nan

            allData[col] = allData[col].map(lambda x: str(x).upper())


allData_bk = allData.copy()

'''

检查数值型变量

'''

missing_pcnt_threshould_2 = 0.8

deleted_var = []

for col in numerical_var:

    missingRate = MissingContinuous(allData, col)

    print('{0} has missing rate as {1}'.format(col, missingRate))

    if missingRate > missing_pcnt_threshould_2:

        deleted_var.append(col)

        print('we delete variable {} because of its high missing rate'.format(col))

    else:

        if missingRate > 0:

            not_missing = allData.loc[allData[col] == allData[col]][col]

            #makeuped = allData[col].map(lambda x: MakeupRandom(x, list(not_missing)))

            missing_position = allData.loc[allData[col] != allData[col]][col].index

            not_missing_sample = random.sample(list(not_missing), len(missing_position))

            allData.loc[missing_position,col] = not_missing_sample

            #del allData[col]

            #allData[col] = makeuped

            missingRate2 = MissingContinuous(allData, col)

            print('missing rate after making up is:{}'.format(str(missingRate2)))


if deleted_var != []:

    for col in deleted_var:

        numerical_var.remove(col)

        del allData[col]



allData.to_csv(folderOfData+'allData_1.csv', header=True,encoding='gbk', columns = allData.columns, index=False)


allData = pd.read_csv(folderOfData+'allData_1.csv', header=0,encoding='gbk')





###################################

# Step 3: 基于卡方分箱法对变量进行分箱#

###################################

'''

对不同类型的变量,分箱的处理是不同的:

(1)数值型变量可直接分箱

(2)取值个数较多的类别型变量,需要用bad rate做编码转换成数值型变量,再分箱

(3)取值个数较少的类别型变量不需要分箱,但是要检查是否每个类别都有好坏样本。如果有类别只有好或坏,需要合并

'''


#for each categorical variable, if it has distinct values more than 5, we use the ChiMerge to merge it


trainData = pd.read_csv(folderOfData+'allData_1.csv',header = 0, encoding='gbk',dtype=object)

allFeatures = list(trainData.columns)

allFeatures.remove('ListingInfo')

allFeatures.remove('target')

#allFeatures.remove('Idx')


#将特征区分为数值型和类别型

numerical_var = []

for var in allFeatures:

    uniq_vals = list(set(trainData[var]))

    if np.nan in uniq_vals:

        uniq_vals.remove( np.nan)

    if len(uniq_vals) >= 10 and isinstance(uniq_vals[0],numbers.Real):

        numerical_var.append(var)


categorical_var = [i for i in allFeatures if i not in numerical_var]


for col in categorical_var:

    #for Chinese character, upper() is not valid

    if col not in ['UserInfo_7','UserInfo_9','UserInfo_19']:

        trainData[col] = trainData[col].map(lambda x: str(x).upper())



'''

对于类别型变量,按照以下方式处理

1,如果变量的取值个数超过5,计算bad rate进行编码

2,除此之外,其他任何类别型变量如果有某个取值中,对应的样本全部是坏样本或者是好样本,进行合并。

'''

deleted_features = []  #将处理过的变量删除,防止对后面建模的干扰

encoded_features = {}  #将bad rate编码方式保存下来,在以后的测试和生产环境中需要使用

merged_features = {}    #将类别型变量合并方案保留下来

var_IV = {}  #save the IV values for binned features      #将IV值保留和WOE值

var_WOE = {}

for col in categorical_var:

    print('we are processing {}'.format(col))

    if len(set(trainData[col]))>5:

        print('{} is encoded with bad rate'.format(col))

        col0 = str(col)+'_encoding'


        #(1),计算坏样本率并进行编码

        encoding_result = BadRateEncoding(trainData, col, 'target')

        trainData[col0], br_encoding = encoding_result['encoding'],encoding_result['bad_rate']


        #(2),将(1)中的编码后的变量也加入数值型变量列表中,为后面的卡方分箱做准备

        numerical_var.append(col0)


        #(3),保存编码结果

        encoded_features[col] = [col0, br_encoding]


        #(4),删除原始值


        deleted_features.append(col)

    else:

        bad_bin = trainData.groupby([col])['target'].sum()

        #对于类别数少于5个,但是出现0坏样本的特征需要做处理

        if min(bad_bin) == 0:

            print('{} has 0 bad sample!'.format(col))

            col1 = str(col) + '_mergeByBadRate'

            #(1),找出最优合并方式,使得每一箱同时包含好坏样本

            mergeBin = MergeBad0(trainData, col, 'target')

            #(2),依照(1)的结果对值进行合并

            trainData[col1] = trainData[col].map(mergeBin)

            maxPcnt = MaximumBinPcnt(trainData, col1)

            #如果合并后导致有箱占比超过90%,就删除。

            if maxPcnt > 0.9:

                print('{} is deleted because of large percentage of single bin'.format(col))

                deleted_features.append(col)

                categorical_var.remove(col)

                del trainData[col]

                continue

            #(3)如果合并后的新的变量满足要求,就保留下来

            merged_features[col] = [col1, mergeBin]

            WOE_IV = CalcWOE(trainData, col1, 'target')

            var_WOE[col1] = WOE_IV['WOE']

            var_IV[col1] = WOE_IV['IV']

            #del trainData[col]

            deleted_features.append(col)

        else:

            WOE_IV = CalcWOE(trainData, col, 'target')

            var_WOE[col] = WOE_IV['WOE']

            var_IV[col] = WOE_IV['IV']



'''

对于连续型变量,处理方式如下:

1,利用卡方分箱法将变量分成5个箱

2,检查坏样本率的单带性,如果发现单调性不满足,就进行合并,直到满足单调性

'''

var_cutoff = {}

for col in numerical_var:

    print("{} is in processing".format(col))

    col1 = str(col) + '_Bin'


    #(1),用卡方分箱法进行分箱,并且保存每一个分割的端点。例如端点=[10,20,30]表示将变量分为x<10,10<x<20,20<x<30和x>30.

    #特别地,缺失值-1不参与分箱

    if -1 in set(trainData[col]):

        special_attribute = [-1]

    else:

        special_attribute = []

    cutOffPoints = ChiMerge(trainData, col, 'target',special_attribute=special_attribute)

    var_cutoff[col] = cutOffPoints

    trainData[col1] = trainData[col].map(lambda x: AssignBin(x, cutOffPoints,special_attribute=special_attribute))


    #(2), check whether the bad rate is monotone

    BRM = BadRateMonotone(trainData, col1, 'target',special_attribute=special_attribute)

    if not BRM:

        if special_attribute == []:

            bin_merged = Monotone_Merge(trainData, 'target', col1)

            removed_index = []

            for bin in bin_merged:

                if len(bin)>1:

                    indices = [int(b.replace('Bin ','')) for b in bin]

                    removed_index = removed_index+indices[0:-1]

            removed_point = [cutOffPoints[k] for k in removed_index]

            for p in removed_point:

                cutOffPoints.remove(p)

            var_cutoff[col] = cutOffPoints

            trainData[col1] = trainData[col].map(lambda x: AssignBin(x, cutOffPoints, special_attribute=special_attribute))

        else:

            cutOffPoints2 = [i for i in cutOffPoints if i not in special_attribute]

            temp = trainData.loc[~trainData[col].isin(special_attribute)]

            bin_merged = Monotone_Merge(temp, 'target', col1)

            removed_index = []

            for bin in bin_merged:

                if len(bin) > 1:

                    indices = [int(b.replace('Bin ', '')) for b in bin]

                    removed_index = removed_index + indices[0:-1]

            removed_point = [cutOffPoints2[k] for k in removed_index]

            for p in removed_point:

                cutOffPoints2.remove(p)

            cutOffPoints2 = cutOffPoints2 + special_attribute

            var_cutoff[col] = cutOffPoints2

            trainData[col1] = trainData[col].map(lambda x: AssignBin(x, cutOffPoints2, special_attribute=special_attribute))


    #(3),分箱后再次检查是否有单一的值占比超过90%。如果有,删除该变量

    maxPcnt = MaximumBinPcnt(trainData, col1)

    if maxPcnt > 0.9:

        # del trainData[col1]

        deleted_features.append(col)

        numerical_var.remove(col)

        print('we delete {} because the maximum bin occupies more than 90%'.format(col))

        continue


    WOE_IV = CalcWOE(trainData, col1, 'target')

    var_IV[col] = WOE_IV['IV']

    var_WOE[col] = WOE_IV['WOE']

    #del trainData[col]




trainData.to_csv(folderOfData+'allData_2.csv', header=True,encoding='gbk', columns = trainData.columns, index=False)




with open(folderOfData+'var_WOE.pkl',"wb") as f:

    f.write(pickle.dumps(var_WOE))


with open(folderOfData+'var_IV.pkl',"wb") as f:

    f.write(pickle.dumps(var_IV))



with open(folderOfData+'var_cutoff.pkl',"wb") as f:

    f.write(pickle.dumps(var_cutoff))



with open(folderOfData+'merged_features.pkl',"wb") as f:

    f.write(pickle.dumps(merged_features))


########################################

# Step 4: WOE编码后的单变量分析与多变量分析#

########################################

trainData = pd.read_csv(folderOfData+'allData_2.csv', header=0, encoding='gbk')



with open(folderOfData+'var_WOE.pkl',"rb") as f:

    var_WOE = pickle.load(f)


with open(folderOfData+'var_IV.pkl',"rb") as f:

    var_IV = pickle.load(f)



with open(folderOfData+'var_cutoff.pkl',"rb") as f:

    var_cutoff = pickle.load(f)



with open(folderOfData+'merged_features.pkl',"rb") as f:

    merged_features = pickle.load(f)


#将一些看起来像数值变量实际上是类别变量的字段转换成字符

num2str = ['SocialNetwork_13','SocialNetwork_12','UserInfo_6','UserInfo_5','UserInfo_10','UserInfo_17']

for col in num2str:

    trainData[col] = trainData[col].map(lambda x: str(x))



for col in var_WOE.keys():

    print(col)

    col2 = str(col)+"_WOE"

    if col in var_cutoff.keys():

        cutOffPoints = var_cutoff[col]

        special_attribute = []

        if - 1 in cutOffPoints:

            special_attribute = [-1]

        binValue = trainData[col].map(lambda x: AssignBin(x, cutOffPoints,special_attribute=special_attribute))

        trainData[col2] = binValue.map(lambda x: var_WOE[col][x])

    else:

        trainData[col2] = trainData[col].map(lambda x: var_WOE[col][x])


trainData.to_csv(folderOfData+'allData_3.csv', header=True,encoding='gbk', columns = trainData.columns, index=False)




### (i) 选择IV高于阈值的变量

trainData = pd.read_csv(folderOfData+'allData_3.csv', header=0,encoding='gbk')

all_IV = list(var_IV.values())

all_IV = sorted(all_IV, reverse=True)

plt.bar(x=range(len(all_IV)), height = all_IV)

iv_threshould = 0.02

varByIV = [k for k, v in var_IV.items() if v > iv_threshould]




### (ii) 检查WOE编码后的变量的两两线性相关性


var_IV_selected = {k:var_IV[k] for k in varByIV}

var_IV_sorted = sorted(var_IV_selected.items(), key=lambda d:d[1], reverse = True)

var_IV_sorted = [i[0] for i in var_IV_sorted]


removed_var  = []

roh_thresould = 0.6

for i in range(len(var_IV_sorted)-1):

    if var_IV_sorted[i] not in removed_var:

        x1 = var_IV_sorted[i]+"_WOE"

        for j in range(i+1,len(var_IV_sorted)):

            if var_IV_sorted[j] not in removed_var:

                x2 = var_IV_sorted[j] + "_WOE"

                roh = np.corrcoef([trainData[x1], trainData[x2]])[0, 1]

                if abs(roh) >= roh_thresould:

                    print('the correlation coeffient between {0} and {1} is {2}'.format(x1, x2, str(roh)))

                    if var_IV[var_IV_sorted[i]] > var_IV[var_IV_sorted[j]]:

                        removed_var.append(var_IV_sorted[j])

                    else:

                        removed_var.append(var_IV_sorted[i])


var_IV_sortet_2 = [i for i in var_IV_sorted if i not in removed_var]


### (iii)检查是否有变量与其他所有变量的VIF > 10

for i in range(len(var_IV_sortet_2)):

    x0 = trainData[var_IV_sortet_2[i]+'_WOE']

    x0 = np.array(x0)

    X_Col = [k+'_WOE' for k in var_IV_sortet_2 if k != var_IV_sortet_2[i]]

    X = trainData[X_Col]

    X = np.matrix(X)

    regr = LinearRegression()

    clr= regr.fit(X, x0)

    x_pred = clr.predict(X)

    R2 = 1 - ((x_pred - x0) ** 2).sum() / ((x0 - x0.mean()) ** 2).sum()

    vif = 1/(1-R2)

    if vif > 10:

        print("Warning: the vif for {0} is {1}".format(var_IV_sortet_2[i], vif))




#########################

# Step 5: 应用逻辑回归模型#

#########################

multi_analysis = [i+'_WOE' for i in var_IV_sortet_2]

y = trainData['target']

X = trainData[multi_analysis].copy()

X['intercept'] = [1]*X.shape[0]



LR = sm.Logit(y, X).fit()

summary = LR.summary2()

pvals = LR.pvalues.to_dict()

params = LR.params.to_dict()


#发现有变量不显著,因此需要单独检验显著性

varLargeP = {k: v for k,v in pvals.items() if v >= 0.1}

varLargeP = sorted(varLargeP.items(), key=lambda d:d[1], reverse = True)

varLargeP = [i[0] for i in varLargeP]

p_value_list = {}

for var in varLargeP:

    X_temp = trainData[var].copy().to_frame()

    X_temp['intercept'] = [1] * X_temp.shape[0]

    LR = sm.Logit(y, X_temp).fit()

    p_value_list[var] = LR.pvalues[var]

for k,v in p_value_list.items():

    print("{0} has p-value of {1} in univariate regression".format(k,v))



#发现有变量的系数为正,因此需要单独检验正确性

varPositive = [k for k,v in params.items() if v >= 0]

coef_list = {}

for var in varPositive:

    X_temp = trainData[var].copy().to_frame()

    X_temp['intercept'] = [1] * X_temp.shape[0]

    LR = sm.Logit(y, X_temp).fit()

    coef_list[var] = LR.params[var]

for k,v in coef_list.items():

    print("{0} has coefficient of {1} in univariate regression".format(k,v))



selected_var = [multi_analysis[0]]

for var in multi_analysis[1:]:

    try_vars = selected_var+[var]

    X_temp = trainData[try_vars].copy()

    X_temp['intercept'] = [1] * X_temp.shape[0]

    LR = sm.Logit(y, X_temp).fit()

    #summary = LR.summary2()

    pvals, params = LR.pvalues, LR.params

    del params['intercept']

    if max(pvals)<0.1 and max(params)<0:

        selected_var.append(var)


LR.summary2()


y_pred = LR.predict(X_temp)

y_result = pd.DataFrame({'y_pred':y_pred, 'y_real':list(trainData['target'])})

KS(y_result,'y_pred','y_real')


roc_auc_score(trainData['target'], y_pred)




################

# Step 6: 尺度化#

################

scores = Prob2Score(y_pred, 200, 100)

plt.hist(score,bins=100)

===============================

import numpy as np

import pandas as pd


def SplitData(df, col, numOfSplit, special_attribute=[]):

    '''

    :param df:按照col排序后的数据集

    :param col:待分箱的变量

    :param numOfSplit:切分的组别数

    :param special_attribute:在切分数据集的时候,某些特殊值需要排除在外

    :return:在原数据集上增加一列,把原始细粒度的col重新划分成粗粒度的值,便于分箱中的合并处理

    '''

    df2 = df.copy()

    if special_attribute != []:

        df2 = df.loc[~df[col].isin(special_attribute)]

    N = df2.shape[0]

    n = int(N/numOfSplit)

    splitPointIndex = [i*n for i in range(1,numOfSplit)]

    rawValues = sorted(list(df2[col]))

    splitPoint = [rawValues[i] for i in splitPointIndex]

    splitPoint = sorted(list(set(splitPoint)))

    return splitPoint


def MaximumBinPcnt(df,col):

    '''

    :return:数据集df中,变量col的分布占比

    '''

    N = df.shape[0]

    total = df.groupby([col])[col].count()

    pcnt = total*1.0/N

    return max(pcnt)




def Chi2(df, total_col, bad_col):

    '''

    :param df:包含全部样本总计与坏样本总计的数据框

    :param total_col:全部样本的个数

    :param bad_col:坏样本的个数

    :return:卡方值

    '''

    df2 = df.copy()

    #求出df中,总体的坏样本率和好样本率

    badRate = sum(df2[bad_col])*1.0/sum(df2[total_col])

    #当全部样本只有好或者坏样本时,卡方值为0

    if badRate in [0,1]:

        return 0

    df2['good'] = df2.apply(lambda x: x[total_col] - x[bad_col], axis = 1)

    goodRate = sum(df2['good']) * 1.0 / sum(df2[total_col])

    #期望坏(好)样本个数=全部样本个数*平均坏(好)样本占比

    df2['badExpected'] = df[total_col].apply(lambda x: x*badRate)

    df2['goodExpected'] = df[total_col].apply(lambda x: x * goodRate)

    badCombined = zip(df2['badExpected'], df2[bad_col])

    goodCombined = zip(df2['goodExpected'], df2['good'])

    badChi = [(i[0]-i[1])**2/i[0] for i in badCombined]

    goodChi = [(i[0] - i[1]) ** 2 / i[0] for i in goodCombined]

    chi2 = sum(badChi) + sum(goodChi)

    return chi2




def BinBadRate(df, col, target, grantRateIndicator=0):

    '''

    :param df:需要计算好坏比率的数据集

    :param col:需要计算好坏比率的特征

    :param target:好坏标签

    :param grantRateIndicator: 1返回总体的坏样本率,0不返回

    :return:每箱的坏样本率,以及总体的坏样本率(当grantRateIndicator==1时)

    '''

    total = df.groupby([col])[target].count()

    total = pd.DataFrame({'total': total})

    bad = df.groupby([col])[target].sum()

    bad = pd.DataFrame({'bad': bad})

    regroup = total.merge(bad, left_index=True, right_index=True, how='left')

    regroup.reset_index(level=0, inplace=True)

    regroup['bad_rate'] = regroup.apply(lambda x: x.bad / x.total, axis=1)

    dicts = dict(zip(regroup[col],regroup['bad_rate']))

    if grantRateIndicator==0:

        return (dicts, regroup)

    N = sum(regroup['total'])

    B = sum(regroup['bad'])

    overallRate = B * 1.0 / N

    return (dicts, regroup, overallRate)




def AssignGroup(x, bin):

    '''

    :return:数值x在区间映射下的结果。例如,x=2,bin=[0,3,5], 由于0<x<3,x映射成3

    '''

    N = len(bin)

    if x<=min(bin):

        return min(bin)

    elif x>max(bin):

        return 10e10

    else:

        for i in range(N-1):

            if bin[i] < x <= bin[i+1]:

                return bin[i+1]



def ChiMerge(df, col, target, max_interval=5,special_attribute=[],minBinPcnt=0):

    '''

    :param df:包含目标变量与分箱属性的数据框

    :param col:需要分箱的属性

    :param target:目标变量,取值0或1

    :param max_interval:最大分箱数。如果原始属性的取值个数低于该参数,不执行这段函数

    :param special_attribute:不参与分箱的属性取值

    :param minBinPcnt:最小箱的占比,默认为0

    :return:分箱结果

    '''

    colLevels = sorted(list(set(df[col])))

    N_distinct = len(colLevels)

    if N_distinct <= max_interval:  #如果原始属性的取值个数低于max_interval,不执行这段函数

        print("The number of original levels for {} is less than or equal to max intervals".format(col))

        return colLevels[:-1]

    else:

        if len(special_attribute)>=1:

            df1 = df.loc[df[col].isin(special_attribute)]

            df2 = df.loc[~df[col].isin(special_attribute)]

        else:

            df2 = df.copy()

        N_distinct = len(list(set(df2[col])))


        #步骤一: 通过col对数据集进行分组,求出每组的总样本数与坏样本数

        if N_distinct > 100:

            split_x = SplitData(df2, col, 100)

            df2['temp'] = df2[col].map(lambda x: AssignGroup(x, split_x))

        else:

            df2['temp'] = df2[col]

        #总体bad rate将被用来计算expected bad count

        (binBadRate, regroup, overallRate) = BinBadRate(df2, 'temp', target, grantRateIndicator=1)


        #首先,每个单独的属性值将被分为单独的一组

        #对属性值进行排序,然后两两组别进行合并

        colLevels = sorted(list(set(df2['temp'])))

        groupIntervals = [[i] for i in colLevels]


        #步骤二:建立循环,不断合并最优的相邻两个组别,直到:

        # 1,最终分裂出来的分箱数<=预设的最大分箱数

        # 2,每箱的占比不低于预设值(可选)

        # 3,每箱同时包含好坏样本

        #如果有特殊属性,那么最终分裂出来的分箱数=预设的最大分箱数-特殊属性的个数

        split_intervals = max_interval - len(special_attribute)

        while (len(groupIntervals) > split_intervals):  #终止条件: 当前分箱数=预设的分箱数

            #每次循环时, 计算合并相邻组别后的卡方值。具有最小卡方值的合并方案,是最优方案

            chisqList = []

            for k in range(len(groupIntervals)-1):

                temp_group = groupIntervals[k] + groupIntervals[k+1]

                df2b = regroup.loc[regroup['temp'].isin(temp_group)]

                chisq = Chi2(df2b, 'total', 'bad')

                chisqList.append(chisq)

            best_comnbined = chisqList.index(min(chisqList))

            groupIntervals[best_comnbined] = groupIntervals[best_comnbined] + groupIntervals[best_comnbined+1]

            #当将最优的相邻的两个变量合并在一起后,需要从原来的列表中将其移除。例如,将[3,4,5] 与[6,7]合并成[3,4,5,6,7]后,需要将[3,4,5] 与[6,7]移除,保留[3,4,5,6,7]

            groupIntervals.remove(groupIntervals[best_comnbined+1])

        groupIntervals = [sorted(i) for i in groupIntervals]

        cutOffPoints = [max(i) for i in groupIntervals[:-1]]


        #检查是否有箱没有好或者坏样本。如果有,需要跟相邻的箱进行合并,直到每箱同时包含好坏样本

        groupedvalues = df2['temp'].apply(lambda x: AssignBin(x, cutOffPoints))

        df2['temp_Bin'] = groupedvalues

        (binBadRate,regroup) = BinBadRate(df2, 'temp_Bin', target)

        [minBadRate, maxBadRate] = [min(binBadRate.values()),max(binBadRate.values())]

        while minBadRate ==0 or maxBadRate == 1:

            #找出全部为好/坏样本的箱

            indexForBad01 = regroup[regroup['bad_rate'].isin([0,1])].temp_Bin.tolist()

            bin=indexForBad01[0]

            #如果是最后一箱,则需要和上一个箱进行合并,也就意味着分裂点cutOffPoints中的最后一个需要移除

            if bin == max(regroup.temp_Bin):

                cutOffPoints = cutOffPoints[:-1]

            #如果是第一箱,则需要和下一个箱进行合并,也就意味着分裂点cutOffPoints中的第一个需要移除

            elif bin == min(regroup.temp_Bin):

                cutOffPoints = cutOffPoints[1:]

            #如果是中间的某一箱,则需要和前后中的一个箱进行合并,依据是较小的卡方值

            else:

                #和前一箱进行合并,并且计算卡方值

                currentIndex = list(regroup.temp_Bin).index(bin)

                prevIndex = list(regroup.temp_Bin)[currentIndex - 1]

                df3 = df2.loc[df2['temp_Bin'].isin([prevIndex, bin])]

                (binBadRate, df2b) = BinBadRate(df3, 'temp_Bin', target)

                chisq1 = Chi2(df2b, 'total', 'bad')

                #和后一箱进行合并,并且计算卡方值

                laterIndex = list(regroup.temp_Bin)[currentIndex + 1]

                df3b = df2.loc[df2['temp_Bin'].isin([laterIndex, bin])]

                (binBadRate, df2b) = BinBadRate(df3b, 'temp_Bin', target)

                chisq2 = Chi2(df2b, 'total', 'bad')

                if chisq1 < chisq2:

                    cutOffPoints.remove(cutOffPoints[currentIndex - 1])

                else:

                    cutOffPoints.remove(cutOffPoints[currentIndex])

            #完成合并之后,需要再次计算新的分箱准则下,每箱是否同时包含好坏样本

            groupedvalues = df2['temp'].apply(lambda x: AssignBin(x, cutOffPoints))

            df2['temp_Bin'] = groupedvalues

            (binBadRate, regroup) = BinBadRate(df2, 'temp_Bin', target)

            [minBadRate, maxBadRate] = [min(binBadRate.values()), max(binBadRate.values())]

        #需要检查分箱后的最小占比

        if minBinPcnt > 0:

            groupedvalues = df2['temp'].apply(lambda x: AssignBin(x, cutOffPoints))

            df2['temp_Bin'] = groupedvalues

            valueCounts = groupedvalues.value_counts().to_frame()

            N = sum(valueCounts['temp'])

            valueCounts['pcnt'] = valueCounts['temp'].apply(lambda x: x * 1.0 / N)

            valueCounts = valueCounts.sort_index()

            minPcnt = min(valueCounts['pcnt'])

            while minPcnt < minBinPcnt and len(cutOffPoints) > 2:

                #找出占比最小的箱

                indexForMinPcnt = valueCounts[valueCounts['pcnt'] == minPcnt].index.tolist()[0]

                #如果占比最小的箱是最后一箱,则需要和上一个箱进行合并,也就意味着分裂点cutOffPoints中的最后一个需要移除

                if indexForMinPcnt == max(valueCounts.index):

                    cutOffPoints = cutOffPoints[:-1]

                #如果占比最小的箱是第一箱,则需要和下一个箱进行合并,也就意味着分裂点cutOffPoints中的第一个需要移除

                elif indexForMinPcnt == min(valueCounts.index):

                    cutOffPoints = cutOffPoints[1:]

                #如果占比最小的箱是中间的某一箱,则需要和前后中的一个箱进行合并,依据是较小的卡方值

                else:

                    #和前一箱进行合并,并且计算卡方值

                    currentIndex = list(valueCounts.index).index(indexForMinPcnt)

                    prevIndex = list(valueCounts.index)[currentIndex - 1]

                    df3 = df2.loc[df2['temp_Bin'].isin([prevIndex, indexForMinPcnt])]

                    (binBadRate, df2b) = BinBadRate(df3, 'temp_Bin', target)

                    chisq1 = Chi2(df2b, 'total', 'bad')

                    #和后一箱进行合并,并且计算卡方值

                    laterIndex = list(valueCounts.index)[currentIndex + 1]

                    df3b = df2.loc[df2['temp_Bin'].isin([laterIndex, indexForMinPcnt])]

                    (binBadRate, df2b) = BinBadRate(df3b, 'temp_Bin', target)

                    chisq2 = Chi2(df2b, 'total', 'bad')

                    if chisq1 < chisq2:

                        cutOffPoints.remove(cutOffPoints[currentIndex - 1])

                    else:

                        cutOffPoints.remove(cutOffPoints[currentIndex])

                groupedvalues = df2['temp'].apply(lambda x: AssignBin(x, cutOffPoints))

                df2['temp_Bin'] = groupedvalues

                valueCounts = groupedvalues.value_counts().to_frame()

                valueCounts['pcnt'] = valueCounts['temp'].apply(lambda x: x * 1.0 / N)

                valueCounts = valueCounts.sort_index()

                minPcnt = min(valueCounts['pcnt'])

        cutOffPoints = special_attribute + cutOffPoints

        return cutOffPoints




def BadRateEncoding(df, col, target):

    '''

    :return:在数据集df中,用坏样本率给col进行编码。target表示坏样本标签

    '''

    regroup = BinBadRate(df, col, target, grantRateIndicator=0)[1]

    br_dict = regroup[[col,'bad_rate']].set_index([col]).to_dict(orient='index')

    for k, v in br_dict.items():

        br_dict[k] = v['bad_rate']

    badRateEnconding = df[col].map(lambda x: br_dict[x])

    return {'encoding':badRateEnconding, 'bad_rate':br_dict}



def AssignBin(x, cutOffPoints,special_attribute=[]):

    '''

    :param x:某个变量的某个取值

    :param cutOffPoints:上述变量的分箱结果,用切分点表示

    :param special_attribute:  不参与分箱的特殊取值

    :return:分箱后的对应的第几个箱,从0开始

    例如, cutOffPoints = [10,20,30], 对于 x = 7, 返回 Bin 0;对于x=23,返回Bin 2; 对于x = 35, return Bin 3。

    对于特殊值,返回的序列数前加"-"

    '''

    cutOffPoints2 = [i for i in cutOffPoints if i not in special_attribute]

    numBin = len(cutOffPoints2)

    if x in special_attribute:

        i = special_attribute.index(x)+1

        return 'Bin {}'.format(0-i)

    if x<=cutOffPoints2[0]:

        return 'Bin 0'

    elif x > cutOffPoints2[-1]:

        return 'Bin {}'.format(numBin)

    else:

        for i in range(0,numBin):

            if cutOffPoints2[i] < x <=  cutOffPoints2[i+1]:

                return 'Bin {}'.format(i+1)




def CalcWOE(df, col, target):

    '''

    :param df:包含需要计算WOE的变量和目标变量

    :param col:需要计算WOE、IV的变量,必须是分箱后的变量,或者不需要分箱的类别型变量

    :param target:目标变量,0、1表示好、坏

    :return:返回WOE和IV

    '''

    total = df.groupby([col])[target].count()

    total = pd.DataFrame({'total': total})

    bad = df.groupby([col])[target].sum()

    bad = pd.DataFrame({'bad': bad})

    regroup = total.merge(bad, left_index=True, right_index=True, how='left')

    regroup.reset_index(level=0, inplace=True)

    N = sum(regroup['total'])

    B = sum(regroup['bad'])

    regroup['good'] = regroup['total'] - regroup['bad']

    G = N - B

    regroup['bad_pcnt'] = regroup['bad'].map(lambda x: x*1.0/B)

    regroup['good_pcnt'] = regroup['good'].map(lambda x: x * 1.0 / G)

    regroup['WOE'] = regroup.apply(lambda x: np.log(x.good_pcnt*1.0/x.bad_pcnt),axis = 1)

    WOE_dict = regroup[[col,'WOE']].set_index(col).to_dict(orient='index')

    for k, v in WOE_dict.items():

        WOE_dict[k] = v['WOE']

    IV = regroup.apply(lambda x: (x.good_pcnt-x.bad_pcnt)*np.log(x.good_pcnt*1.0/x.bad_pcnt),axis = 1)

    IV = sum(IV)

    return {"WOE": WOE_dict, 'IV':IV}



def FeatureMonotone(x):

    '''

    :return:返回序列x中有几个元素不满足单调性,以及这些元素的位置。

    例如,x=[1,3,2,5], 元素3比前后两个元素都大,不满足单调性;元素2比前后两个元素都小,也不满足单调性。

    故返回的不满足单调性的元素个数为2,位置为1和2.

    '''

    monotone = [x[i]x[i+1] and x[i] > x[i-1] for i in range(1,len(x)-1)]

    index_of_nonmonotone = [i+1 for i in range(len(monotone)) if monotone[i]]

    return {'count_of_nonmonotone':monotone.count(True), 'index_of_nonmonotone':index_of_nonmonotone}


## 判断某变量的坏样本率是否单调

def BadRateMonotone(df, sortByVar, target,special_attribute = []):

    '''

    :param df:包含检验坏样本率的变量,和目标变量

    :param sortByVar:需要检验坏样本率的变量

    :param target:目标变量,0、1表示好、坏

    :param special_attribute:不参与检验的特殊值

    :return:坏样本率单调与否

    '''

    df2 = df.loc[~df[sortByVar].isin(special_attribute)]

    if len(set(df2[sortByVar])) <= 2:

        return True

    regroup = BinBadRate(df2, sortByVar, target)[1]

    combined = zip(regroup['total'],regroup['bad'])

    badRate = [x[1]*1.0/x[0] for x in combined]

    badRateNotMonotone = FeatureMonotone(badRate)['count_of_nonmonotone']

    if badRateNotMonotone > 0:

        return False

    else:

        return True


def MergeBad0(df,col,target, direction='bad'):

    '''

     :param df:包含检验0%或者100%坏样本率

     :param col:分箱后的变量或者类别型变量。检验其中是否有一组或者多组没有坏样本或者没有好样本。如果是,则需要进行合并

     :param target:目标变量,0、1表示好、坏

     :return:合并方案,使得每个组里同时包含好坏样本

     '''

    regroup = BinBadRate(df, col, target)[1]

    if direction == 'bad':

        #如果是合并0坏样本率的组,则跟最小的非0坏样本率的组进行合并

        regroup = regroup.sort_values(by  = 'bad_rate')

    else:

        #如果是合并0好样本率的组,则跟最小的非0好样本率的组进行合并

        regroup = regroup.sort_values(by='bad_rate',ascending=False)

    regroup.index = range(regroup.shape[0])

    col_regroup = [[i] for i in regroup[col]]

    del_index = []

    for i in range(regroup.shape[0]-1):

        col_regroup[i+1] = col_regroup[i] + col_regroup[i+1]

        del_index.append(i)

        if direction == 'bad':

            if regroup['bad_rate'][i+1] > 0:

                break

        else:

            if regroup['bad_rate'][i+1] < 1:

                break

    col_regroup2 = [col_regroup[i] for i in range(len(col_regroup)) if i not in del_index]

    newGroup = {}

    for i in range(len(col_regroup2)):

        for g2 in col_regroup2[i]:

            newGroup[g2] = 'Bin '+str(i)

    return newGroup



def Monotone_Merge(df, target, col):

    '''

    :return:将数据集df中,不满足坏样本率单调性的变量col进行合并,使得合并后的新的变量中,坏样本率单调,输出合并方案。

    例如,col=[Bin 0, Bin 1, Bin 2, Bin 3, Bin 4]是不满足坏样本率单调性的。合并后的col是:

    [Bin 0&Bin 1, Bin 2, Bin 3, Bin 4].

    合并只能在相邻的箱中进行。

    迭代地寻找最优合并方案。每一步迭代时,都尝试将所有非单调的箱进行合并,每一次尝试的合并都是跟前后箱进行合并再做比较

    '''

    def MergeMatrix(m, i,j,k):

        '''

        :param m:需要合并行的矩阵

        :param i,j:合并第i和j行

        :param k:删除第k行

        :return:合并后的矩阵

        '''

        m[i, :] = m[i, :] + m[j, :]

        m = np.delete(m, k, axis=0)

        return m


    def Merge_adjacent_Rows(i, bad_by_bin_current, bins_list_current, not_monotone_count_current):

        '''

        :param i:需要将第i行与前、后的行分别进行合并,比较哪种合并方案最佳。判断准则是,合并后非单调性程度减轻,且更加均匀

        :param bad_by_bin_current:合并前的分箱矩阵,包括每一箱的样本个数、坏样本个数和坏样本率

        :param bins_list_current:合并前的分箱方案

        :param not_monotone_count_current:合并前的非单调性元素个数

        :return:分箱后的分箱矩阵、分箱方案、非单调性元素个数和衡量均匀性的指标balance

        '''

        i_prev = i - 1

        i_next = i + 1

        bins_list = bins_list_current.copy()

        bad_by_bin = bad_by_bin_current.copy()

        not_monotone_count = not_monotone_count_current

        #合并方案a:将第i箱与前一箱进行合并

        bad_by_bin2a = MergeMatrix(bad_by_bin.copy(), i_prev, i, i)

        bad_by_bin2a[i_prev, -1] = bad_by_bin2a[i_prev, -2] / bad_by_bin2a[i_prev, -3]

        not_monotone_count2a = FeatureMonotone(bad_by_bin2a[:, -1])['count_of_nonmonotone']

        #合并方案b:将第i行与后一行进行合并

        bad_by_bin2b = MergeMatrix(bad_by_bin.copy(), i, i_next, i_next)

        bad_by_bin2b[i, -1] = bad_by_bin2b[i, -2] / bad_by_bin2b[i, -3]

        not_monotone_count2b = FeatureMonotone(bad_by_bin2b[:, -1])['count_of_nonmonotone']

        balance = ((bad_by_bin[:, 1] / N).T * (bad_by_bin[:, 1] / N))[0, 0]

        balance_a = ((bad_by_bin2a[:, 1] / N).T * (bad_by_bin2a[:, 1] / N))[0, 0]

        balance_b = ((bad_by_bin2b[:, 1] / N).T * (bad_by_bin2b[:, 1] / N))[0, 0]

        #满足下述2种情况时返回方案a:(1)方案a能减轻非单调性而方案b不能;(2)方案a和b都能减轻非单调性,但是方案a的样本均匀性优于方案b

        if not_monotone_count2a < not_monotone_count_current and not_monotone_count2b >= not_monotone_count_current or \

                                        not_monotone_count2a < not_monotone_count_current and not_monotone_count2b < not_monotone_count_current and balance_a < balance_b:

            bins_list[i_prev] = bins_list[i_prev] + bins_list[i]

            bins_list.remove(bins_list[i])

            bad_by_bin = bad_by_bin2a

            not_monotone_count = not_monotone_count2a

            balance = balance_a

        #同样地,满足下述2种情况时返回方案b:(1)方案b能减轻非单调性而方案a不能;(2)方案a和b都能减轻非单调性,但是方案b的样本均匀性优于方案a

        elif not_monotone_count2a >= not_monotone_count_current and not_monotone_count2b < not_monotone_count_current or \

                                        not_monotone_count2a < not_monotone_count_current and not_monotone_count2b < not_monotone_count_current and balance_a > balance_b:

            bins_list[i] = bins_list[i] + bins_list[i_next]

            bins_list.remove(bins_list[i_next])

            bad_by_bin = bad_by_bin2b

            not_monotone_count = not_monotone_count2b

            balance = balance_b

        #如果方案a和b都不能减轻非单调性,返回均匀性更优的合并方案

        else:

            if balance_a< balance_b:

                bins_list[i] = bins_list[i] + bins_list[i_next]

                bins_list.remove(bins_list[i_next])

                bad_by_bin = bad_by_bin2b

                not_monotone_count = not_monotone_count2b

                balance = balance_b

            else:

                bins_list[i] = bins_list[i] + bins_list[i_next]

                bins_list.remove(bins_list[i_next])

                bad_by_bin = bad_by_bin2b

                not_monotone_count = not_monotone_count2b

                balance = balance_b

        return {'bins_list': bins_list, 'bad_by_bin': bad_by_bin, 'not_monotone_count': not_monotone_count,

                'balance': balance}



    N = df.shape[0]

    [badrate_bin, bad_by_bin] = BinBadRate(df, col, target)

    bins = list(bad_by_bin[col])

    bins_list = [[i] for i in bins]

    badRate = sorted(badrate_bin.items(), key=lambda x: x[0])

    badRate = [i[1] for i in badRate]

    not_monotone_count, not_monotone_position = FeatureMonotone(badRate)['count_of_nonmonotone'], FeatureMonotone(badRate)['index_of_nonmonotone']

    #迭代地寻找最优合并方案,终止条件是:当前的坏样本率已经单调,或者当前只有2箱

    while (not_monotone_count > 0 and len(bins_list)>2):

        #当非单调的箱的个数超过1个时,每一次迭代中都尝试每一个箱的最优合并方案

        all_possible_merging = []

        for i in not_monotone_position:

            merge_adjacent_rows = Merge_adjacent_Rows(i, np.mat(bad_by_bin), bins_list, not_monotone_count)

            all_possible_merging.append(merge_adjacent_rows)

        balance_list = [i['balance'] for i in all_possible_merging]

        not_monotone_count_new = [i['not_monotone_count'] for i in all_possible_merging]

        #如果所有的合并方案都不能减轻当前的非单调性,就选择更加均匀的合并方案

        if min(not_monotone_count_new) >= not_monotone_count:

            best_merging_position = balance_list.index(min(balance_list))

        #如果有多个合并方案都能减轻当前的非单调性,也选择更加均匀的合并方案

        else:

            better_merging_index = [i for i in range(len(not_monotone_count_new)) if not_monotone_count_new[i] < not_monotone_count]

            better_balance = [balance_list[i] for i in better_merging_index]

            best_balance_index = better_balance.index(min(better_balance))

            best_merging_position = better_merging_index[best_balance_index]

        bins_list = all_possible_merging[best_merging_position]['bins_list']

        bad_by_bin = all_possible_merging[best_merging_position]['bad_by_bin']

        not_monotone_count = all_possible_merging[best_merging_position]['not_monotone_count']

        not_monotone_position = FeatureMonotone(bad_by_bin[:, 3])['index_of_nonmonotone']

    return bins_list






def Prob2Score(prob, basePoint, PDO):

    #将概率转化成分数且为正整数

    y = np.log(prob/(1-prob))

    y2 = basePoint+PDO/np.log(2)*(-y)

    score = y2.astype("int")

    return score




### 计算KS值

def KS(df, score, target):

    '''

    :param df:包含目标变量与预测值的数据集

    :param score:得分或者概率

    :param target:目标变量

    :return: KS值

    :return: KS值

    '''

    total = df.groupby([score])[target].count()

    bad = df.groupby([score])[target].sum()

    all = pd.DataFrame({'total':total, 'bad':bad})

    all['good'] = all['total'] - all['bad']

    all[score] = all.index

    all = all.sort_values(by=score,ascending=False)

    all.index = range(len(all))

    all['badCumRate'] = all['bad'].cumsum() / all['bad'].sum()

    all['goodCumRate'] = all['good'].cumsum() / all['good'].sum()

    KS = all.apply(lambda x: x.badCumRate - x.goodCumRate, axis=1)

    return max(KS)

最后编辑于
©著作权归作者所有,转载或内容合作请联系作者
  • 序言:七十年代末,一起剥皮案震惊了整个滨河市,随后出现的几起案子,更是在滨河造成了极大的恐慌,老刑警刘岩,带你破解...
    沈念sama阅读 211,639评论 6 492
  • 序言:滨河连续发生了三起死亡事件,死亡现场离奇诡异,居然都是意外死亡,警方通过查阅死者的电脑和手机,发现死者居然都...
    沈念sama阅读 90,277评论 3 385
  • 文/潘晓璐 我一进店门,熙熙楼的掌柜王于贵愁眉苦脸地迎上来,“玉大人,你说我怎么就摊上这事。” “怎么了?”我有些...
    开封第一讲书人阅读 157,221评论 0 348
  • 文/不坏的土叔 我叫张陵,是天一观的道长。 经常有香客问我,道长,这世上最难降的妖魔是什么? 我笑而不...
    开封第一讲书人阅读 56,474评论 1 283
  • 正文 为了忘掉前任,我火速办了婚礼,结果婚礼上,老公的妹妹穿的比我还像新娘。我一直安慰自己,他们只是感情好,可当我...
    茶点故事阅读 65,570评论 6 386
  • 文/花漫 我一把揭开白布。 她就那样静静地躺着,像睡着了一般。 火红的嫁衣衬着肌肤如雪。 梳的纹丝不乱的头发上,一...
    开封第一讲书人阅读 49,816评论 1 290
  • 那天,我揣着相机与录音,去河边找鬼。 笑死,一个胖子当着我的面吹牛,可吹牛的内容都是我干的。 我是一名探鬼主播,决...
    沈念sama阅读 38,957评论 3 408
  • 文/苍兰香墨 我猛地睁开眼,长吁一口气:“原来是场噩梦啊……” “哼!你这毒妇竟也来了?” 一声冷哼从身侧响起,我...
    开封第一讲书人阅读 37,718评论 0 266
  • 序言:老挝万荣一对情侣失踪,失踪者是张志新(化名)和其女友刘颖,没想到半个月后,有当地人在树林里发现了一具尸体,经...
    沈念sama阅读 44,176评论 1 303
  • 正文 独居荒郊野岭守林人离奇死亡,尸身上长有42处带血的脓包…… 初始之章·张勋 以下内容为张勋视角 年9月15日...
    茶点故事阅读 36,511评论 2 327
  • 正文 我和宋清朗相恋三年,在试婚纱的时候发现自己被绿了。 大学时的朋友给我发了我未婚夫和他白月光在一起吃饭的照片。...
    茶点故事阅读 38,646评论 1 340
  • 序言:一个原本活蹦乱跳的男人离奇死亡,死状恐怖,灵堂内的尸体忽然破棺而出,到底是诈尸还是另有隐情,我是刑警宁泽,带...
    沈念sama阅读 34,322评论 4 330
  • 正文 年R本政府宣布,位于F岛的核电站,受9级特大地震影响,放射性物质发生泄漏。R本人自食恶果不足惜,却给世界环境...
    茶点故事阅读 39,934评论 3 313
  • 文/蒙蒙 一、第九天 我趴在偏房一处隐蔽的房顶上张望。 院中可真热闹,春花似锦、人声如沸。这庄子的主人今日做“春日...
    开封第一讲书人阅读 30,755评论 0 21
  • 文/苍兰香墨 我抬头看了看天上的太阳。三九已至,却和暖如春,着一层夹袄步出监牢的瞬间,已是汗流浃背。 一阵脚步声响...
    开封第一讲书人阅读 31,987评论 1 266
  • 我被黑心中介骗来泰国打工, 没想到刚下飞机就差点儿被人妖公主榨干…… 1. 我叫王不留,地道东北人。 一个月前我还...
    沈念sama阅读 46,358评论 2 360
  • 正文 我出身青楼,却偏偏与公主长得像,于是被迫代替她去往敌国和亲。 传闻我的和亲对象是个残疾皇子,可洞房花烛夜当晚...
    茶点故事阅读 43,514评论 2 348