toad评分卡构建与改写问题

Question

我用toad包构建了一个评分卡模型，我想对他进行改造，我把里面用的逻辑回归模型换成了别的模型，但是我发现不管我运不运行这个a模型，toad都可以直接出一个评分卡结果，难道我这个改写是错误的吗？toad是内置的逻辑回归，我是没法简单改写吗？

import pandas as pd
train=pd.read_csv("训练集.csv",index_col=0)
test=pd.read_csv("测试集.csv",index_col=0)

Xtr=train.loc[:,"ficoRangeLow":"n14"]
Ytr=train.loc[:,"isDefault"]

Xts=test.loc[:,"ficoRangeLow":"n14"]
Yts=test.loc[:,"isDefault"]


data_tr = pd.concat([Xtr,Ytr],axis=1)
data_tr['type'] = 'train'

data_ts = pd.concat([Xts,Yts],axis=1)
data_ts['type'] = 'test'
import toad
toad.detector.detect(data_tr).columns
toad.detector.detect(data_tr)

quality = toad.quality(data_tr,'isDefault',iv_only=True)
quality.sort_values('iv',ascending=False)


selected_data, dropped = toad.selection.select(data_tr,target = 'isDefault', empty = 0.5, iv = 0.02, corr = 0.9, return_drop=True,exclude=['type'])
print(dropped)
print(selected_data.shape)



quality = toad.quality(selected_data,'isDefault',iv_only=True)
quality.sort_values('iv',ascending=False)



ivzhi=quality.sort_values('iv',ascending=False)
ivzhi.to_csv("iv值排序.csv")


combiner = toad.transform.Combiner()
combiner.fit(selected_data, y = 'isDefault', method = 'chi', min_samples = 0.05) #empty_separate = False


from toad.plot import bin_plot
for i in range(0,14,1):
    col = selected_data.columns[i]
    bin_plot(combiner.transform(selected_data[[col,'isDefault']], labels=True), x=col, target='isDefault')


bins = combiner.export()
selected_test = data_ts[selected_data.columns]
combiner.set_rules(bins)
binned_data = combiner.transform(selected_data)
transer = toad.transform.WOETransformer()
data_tr_woe = transer.fit_transform(binned_data, binned_data['isDefault'], exclude=['isDefault','type'])
data_ts_woe = transer.transform(combiner.transform(selected_test))
Xtr_woe = data_tr_woe.drop(['isDefault','type'],axis=1)
Ytr_woe = data_tr_woe['isDefault']
Xts_woe = data_ts_woe.drop(['isDefault','type'],axis=1)
Yts_woe = data_ts_woe['isDefault']

import autogluon
from autogluon.tabular import TabularDataset,TabularPredictor  
import pandas as pd
import numpy as np
label='isDefault'
train_data=TabularDataset(data_tr_woe.drop(["type"],axis=1)  )
metric = 'roc_auc'
time_limit=60
predictor=TabularPredictor(label=label,eval_metric=metric).fit(train_data,presets='best_quality',time_limit=time_limit,auto_stack=True)

test_data=TabularDataset(data_ts_woe.drop(['isDefault','type'],axis=1))

predictor.predict_proba(train_data)
train_proba=predictor.predict_proba(train_data)
train_proba=train_proba.values[:,1]
train_proba = np.array(train_proba).flatten() 

predictor.predict_proba(test_data)
auto_proba=predictor.predict_proba(test_data)
auto_proba=auto_proba.values[:,1]
auto_proba = np.array(auto_proba).flatten() 

psi = toad.metrics.PSI(data_tr_woe,data_ts_woe)
psi.sort_values(0,ascending=False)

psizhi=psi.sort_values(0,ascending=False)
psizhi.to_csv("psi值.csv")


tr_bucket = toad.metrics.KS_bucket(train_proba,Ytr,bucket=10,method='quantile')
tr_bucket


selected_data
x_card=selected_data.loc[:,"ficoRangeLow":"n14"]
y_card=selected_data.loc[:,"isDefault"]

card = ScoreCard(combiner=combiner, 
                 transer=transer, 
                 C=0.1, 
                 class_weight='balanced', 
                 base_score=600,
                 base_odds=1,
                 pdo=50,
                 rate=2)
 
# 使用评分卡模型进行拟合
card.fit(x_card,y_card)

最后评分卡生成的这一步和我上面调试的模型没有任何关系，请问该如何改写才能让评分卡的构建基于我的模型。

Answer 1

toad对于模型的改造收不容易的，因为逻辑回归模型才带.coef_,你必须得根据线性模型才能应用于评分卡。你可以结合以下逻辑回归模型与自动机器学习，这样容易改。