python， 在信用评级中，计算KS statistic值

 

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

# -*- coding: utf-8 -*-   import pandas as pd from sklearn.grid_search import GridSearchCV from sklearn.model_selection import train_test_split from sklearn.preprocessing import StandardScaler from sklearn.utils import shuffle import numpy as np from sklearn import metrics from sklearn.metrics import log_loss, recall_score, precision_score, accuracy_score,f1_score from sklearn.metrics import roc_curve, precision_recall_curve, roc_auc_score # from sklearn.model_selection import cross_val_score import lightgbm     def ks_statistic(Y,Y_hat):     data = {"Y":Y,"Y_hat":Y_hat}     df = pd.DataFrame(data)     bins = np.array([-0.1,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0])     category = pd.cut(df["Y_hat"],bins=bins)     category = category.sort_values()     #max_index = len(np.unique(df["Y_hat"]))     Y = df.ix[category.index,:]['Y']     Y_hat = df.ix[category.index,:]['Y_hat']     df2 = pd.concat([Y,Y_hat],axis=1)     df3 = pd.pivot_table(df2,values = ['Y_hat'],index ='Y_hat',columns='Y',aggfunc=len,fill_value=0)     df4 = np.cumsum(df3)     df5 = df4/df4.iloc[:,1].max()     ks = max(abs(df5.iloc[:,0] - df5.iloc[:,1]))     return ks/len(bins)         df = pd.read_csv('DC_ALL_20170217.csv', header=0) X = df[df.columns.drop(['user_id','overdue'])].fillna(-999) # X = df[['count','time_stamp','credit_limit','credit_card_use_rate','credit_count_x','bank_count','sex','occupation','education','marriage','hukou']] y = df['overdue'] train = X.head(55596) test = X.tail(69495-55596)   train_label = y.head(55596).convert_objects(convert_numeric=True) X_train, X_test, y_train, y_test = train_test_split(\      train.values,  train_label, test_size=0.2, random_state=42)   max_depth = 5 subsample=0.8 learning_rate=0.01 n_estimators=400 random_state=3 nthread=4 is_unbalance=True objective ='binary' LGBM = lightgbm.LGBMClassifier(max_depth=max_depth, learning_rate=learning_rate, n_estimators=n_estimators, objective=objective,is_unbalance=is_unbalance, nthread=nthread,subsample=subsample) LGBM.fit(X_train, y_train) y_test_v = LGBM.predict(X_test) y_test_p = LGBM.predict_proba(X_test)[:, 1]     print 'auc: ', roc_auc_score(y_test, y_test_p) print 'log_loss: ', log_loss(y_test, y_test_p) print 'precision: ', precision_score(y_test, y_test_v) print 'recall: ', recall_score(y_test, y_test_v) print 'accuracy: ', accuracy_score(y_test, y_test_v) print 'f1_score: ', f1_score(y_test, y_test_v) print 'ks_statistic: ', ks_statistic(y_test.values, y_test_v)