# -*- coding: utf-8 -*-
"""
Created on Fri Oct  9 13:34:59 2020

@author: Administrator
"""

import scorecardpy as sc
dat = sc.germancredit()
#首先，导入germancredit数据。

#1. 筛选变量
#这个函数可以根据指定的条件筛选变量，例如IV值、缺失率、一致性等
dt_s = sc.var_filter(dt=dat
                     , y="creditability"
                     , x=None
                     , iv_limit=0.02
                     , missing_limit=0.95
                     , identical_limit=0.95
                     , var_rm=None #强制删除变量的名称
                     , var_kp=None #强制保留变量的名称
                     , return_rm_reason=False #是否返回每个变量被删除的原因
                     , positive='bad|1' #坏样本的标签
                     )
# 数据划分
train, test = sc.split_df(dt=dt_s
                          , y='creditability'
                          , ratio=0.7 #默认按照7:3对数据集进行分割
                          , seed=186
                          ).values()
#变量分箱 返回字典
bins = sc.woebin(dt_s, y="creditability")

bins = sc.woebin(dt=dt_s #数据
                 , y="creditability" #目标值
                 , x=None #
                 , var_skip=None # 指定不需要分箱的变量。
                 , breaks_list=None #分割点的List。对分箱进行调整的时候用。可以进行自定义分箱
                 , special_values=None #指定单独的箱。
                 , stop_limit=0.1 #当IV值的增加值小于stop_limit或者卡方值小于qchisq(1-stoplimit, 1)时停止分割。
           , count_distr_limit=0.05 #分箱结果中最小占比。默认0.05
           , bin_num_limit=8 #最大分箱数。
           # min_perc_fine_bin=0.02, min_perc_coarse_bin=0.05, max_num_bin=8, 
           , positive="bad|1"
           , no_cores=None
           , print_step=0
           , method="tree"#分箱方法，可以有"tree" or "chimerge"。
           , ignore_const_cols=True#是否忽略常数列。
           , ignore_datetime_cols=True
           , check_cate_num=True#检查分类变量中类别数是否大于50。
           , replace_blank=True#将空值替换为None。
           , save_breaks_list=None
           )


# 变量分箱
sc.woebin_plot(bins)

# 分箱调整
breaks_adj = sc.woebin_adj(dt_s, "creditability", bins) 
bins_adj = sc.woebin(dt_s, y="creditability", breaks_list=breaks_adj)


#woe转换
train_woe = sc.woebin_ply(train, bins_adj)
test_woe = sc.woebin_ply(test, bins_adj)


#模型建立
y_train = train_woe.loc[:,'creditability']
X_train = train_woe.loc[:,train_woe.columns != 'creditability']
y_test = test_woe.loc[:,'creditability']
X_test = test_woe.loc[:,train_woe.columns != 'creditability']

# logistic regression ------
from sklearn.linear_model import LogisticRegression
lr = LogisticRegression(penalty='l1', C=0.9, solver='saga', n_jobs=-1)
lr.fit(X_train, y_train)
# lr.coef_
# lr.intercept_

# predicted proability
train_pred = lr.predict_proba(X_train)[:,1]
test_pred = lr.predict_proba(X_test)[:,1]

#模型评估
train_perf = sc.perf_eva(y_train, train_pred, title = "train")
test_perf = sc.perf_eva(y_test, test_pred, title = "test")

#评分映射
card = sc.scorecard(bins_adj, lr, X_train.columns)

# credit score
train_score = sc.scorecard_ply(train, card, print_step=0)
test_score = sc.scorecard_ply(test, card, print_step=0)

#评分稳定性评估--PSI
sc.perf_psi(
  score = {'train':train_score, 'test':test_score},
  label = {'train':y_train, 'test':y_test}
)