def test_PSI_frame():
result = PSI(
test_df,
base_df,
combiner={
'A': [0.3, 0.5, 0.7],
'B': [0.4, 0.8],
},
)
assert result['B'] == 0.014528279995858708
def test_PSI_return_frame():
result, frame = PSI(
test_df,
base_df,
combiner={
'A': [0.3, 0.5, 0.7],
'B': [0.4, 0.8],
},
return_frame=True,
)
assert frame.loc[4, 'test'] == 0.38
def var_psi_chart(train_data, test_data, var_list, month, q=10):
'''
以训练集为预期分布,跨时间窗监控变量稳定性
var_list:需要查看的变量
'''
data_lst = []
for i in tqdm(var_list):
df_lst = []
for j, k in test_data.groupby(month):
if len(k) < 50:
pass
else:
df = pd.DataFrame()
df['var'] = [i]
df.set_index('var', inplace=True)
df[j] = PSI(k[i], train_data[i])
df_lst.append(df)
data = pd.concat(df_lst, axis=1)
data_lst.append(data)
data_all = pd.concat(data_lst, axis=0)
return data_all
def test_PSI():
result = PSI(feature, base_feature, combiner=[0.3, 0.5, 0.7])
assert result == 0.018630024627491467
print('precision_train:{:.3f} | precision_test:{:.3f}'.format(
precision_train, precision_test))
print(
'Recall_train:{:.3f} | Recall_test:{:.3f}'.format(Recall_train,
Recall_test), '\n' * 2)
vs.roc(fpr_train, tpr_train, fpr_test, tpr_test)
# ----------------------------------------------------------------------------------------------------------------------------------------------------
# 报告
print('报告'.center(62, '—'))
# EDA报告，参与模型训练的变量情况
EDA = toad.detect(X_train)
EDA.to_csv('EDA.csv')
print('EDA done!')
# PSI
PSI = PSI(X_train, X_test)
PSI.to_csv('PSI.csv')
print('PSI done!')
# KS报告
KS_train = KS_bucket(y_train_prob, y_train, bucket=20, method='quantile')
KS_test = KS_bucket(y_test_prob, y_test, bucket=20, method='quantile')
KS_train.to_csv('KS_train.csv')
KS_test.to_csv('KS_test.csv')
print('KS done!')
end = time.time()
print('总耗时：{:.2f}'.format((end - start)))