检查VIF

[73]:

get_vif(data_train_woe,data_train_woe.columns,return_data=True)

[73]:

(2.97,
     variables  VIF
 0          ID 1.02
 1   LIMIT_BAL 1.48
 2         SEX 1.02
 3   EDUCATION 1.12
 4    MARRIAGE 1.06
 5         AGE 1.03
 6       PAY_0 1.95
 7       PAY_2 2.91
 8       PAY_3 2.73
 9       PAY_4 2.78
 10      PAY_5 2.97
 11      PAY_6 2.17
 12  BILL_AMT1 1.94
 13  BILL_AMT2 1.90
 14  BILL_AMT3 2.10
 15  BILL_AMT4 1.56
 16  BILL_AMT5 2.30
 17  BILL_AMT6 1.93
 18   PAY_AMT1 1.90
 19   PAY_AMT2 1.89
 20   PAY_AMT3 1.91
 21   PAY_AMT4 2.07
 22   PAY_AMT5 1.82
 23   PAY_AMT6 1.68)

[74]:

# 查看评分卡模型的参数
print('a:',a,'','b:',b)

a: 508.4396430011438  b: 72.13475204444818

预测模型分数

[75]:

# 预测训练集模型分数
data_train_score = get_predict_score(data_train,scorecard)
data_train_score

[75]:

	PAY_0	PAY_0_Score	PAY_AMT1	PAY_AMT1_Score	LIMIT_BAL	LIMIT_BAL_Score	PAY_AMT2	PAY_AMT2_Score	PAY_AMT3	PAY_AMT3_Score	PAY_AMT4	PAY_AMT4_Score	PAY_AMT5	PAY_AMT5_Score	PAY_AMT6	PAY_AMT6_Score	y	Score	Proba
0	-1	28.00	1671.00	0.00	120000.00	-6.00	380.00	-0.00	131062.00	14.00	2000.00	2.00	3000.00	3.00	3000.00	1.00	0	642.00	0.14
1	-2	28.00	1468.00	0.00	200000.00	14.00	2321.00	-0.00	163597.00	14.00	6680.00	5.00	3963.00	3.00	2514.00	1.00	0	665.00	0.10
2	0	43.00	4038.00	0.00	80000.00	-6.00	3199.00	-0.00	914.00	0.00	850.00	-1.00	2055.00	3.00	8318.00	1.00	0	640.00	0.14
3	0	43.00	1596.00	0.00	20000.00	-27.00	2000.00	-0.00	3000.00	0.00	0.00	-5.00	1600.00	-1.00	0.00	-2.00	0	608.00	0.20
4	0	43.00	3000.00	0.00	90000.00	-6.00	2000.00	-0.00	2000.00	0.00	2000.00	2.00	2000.00	-1.00	1087.00	-0.00	0	638.00	0.14
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
22495	2	-137.00	3000.00	0.00	50000.00	-6.00	2525.00	-0.00	3900.00	0.00	0.00	-5.00	2000.00	-1.00	4500.00	1.00	1	452.00	0.69
22496	-1	28.00	0.00	-7.00	210000.00	14.00	358.00	-0.00	12816.00	7.00	0.00	-5.00	102.00	-1.00	210.00	-0.00	0	636.00	0.15
22497	1	-38.00	0.00	-7.00	390000.00	28.00	1266.00	-0.00	0.00	-8.00	0.00	-5.00	0.00	-5.00	0.00	-2.00	0	563.00	0.32
22498	0	43.00	1700.00	0.00	30000.00	-27.00	1600.00	-0.00	1287.00	0.00	1296.00	-1.00	500.00	-1.00	1550.00	-0.00	0	614.00	0.19
22499	2	-137.00	6300.00	5.00	150000.00	14.00	6100.00	8.00	4900.00	7.00	0.00	-5.00	10200.00	9.00	5100.00	1.00	0	502.00	0.52

22500 rows × 19 columns

[76]:

# 预测测试集模型分数
data_test_score = get_predict_score(data_test,scorecard)
data_test_score

[76]:

	PAY_0	PAY_0_Score	PAY_AMT1	PAY_AMT1_Score	LIMIT_BAL	LIMIT_BAL_Score	PAY_AMT2	PAY_AMT2_Score	PAY_AMT3	PAY_AMT3_Score	PAY_AMT4	PAY_AMT4_Score	PAY_AMT5	PAY_AMT5_Score	PAY_AMT6	PAY_AMT6_Score	y	Score	Proba
0	-2	28.00	0.00	-7.00	400000.00	28.00	0.00	-11.00	0.00	-8.00	0.00	-5.00	0.00	-5.00	0.00	-2.00	0	621.00	0.18
1	0	43.00	2600.00	0.00	80000.00	-6.00	4300.00	-0.00	2000.00	0.00	2000.00	2.00	2000.00	-1.00	2000.00	-0.00	0	641.00	0.14
2	1	-38.00	0.00	-7.00	200000.00	14.00	2317.00	-0.00	7588.00	7.00	7614.00	5.00	14053.00	9.00	0.00	-2.00	0	591.00	0.25
3	-1	28.00	1087.00	0.00	20000.00	-27.00	1140.00	-0.00	0.00	-8.00	7014.00	5.00	800.00	-1.00	0.00	-2.00	0	598.00	0.23
4	2	-137.00	5000.00	5.00	70000.00	-6.00	3000.00	-0.00	2000.00	0.00	3000.00	2.00	5000.00	3.00	0.00	-2.00	0	468.00	0.65
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
7495	0	43.00	1128.00	0.00	500000.00	28.00	1000.00	-0.00	0.00	-8.00	8479.00	5.00	236.00	-1.00	2990.00	1.00	0	671.00	0.10
7496	0	43.00	8080.00	5.00	110000.00	-6.00	14298.00	8.00	2519.00	0.00	6616.00	5.00	1953.00	-1.00	5300.00	1.00	0	658.00	0.12
7497	0	43.00	4539.00	0.00	150000.00	14.00	4218.00	-0.00	4204.00	0.00	3296.00	2.00	3408.00	3.00	3416.00	1.00	0	666.00	0.10
7498	0	43.00	5059.00	5.00	140000.00	-6.00	14659.00	8.00	5000.00	7.00	8000.00	5.00	5000.00	3.00	10000.00	3.00	0	671.00	0.10
7499	0	43.00	3500.00	0.00	80000.00	-6.00	2934.00	-0.00	897.00	0.00	929.00	-1.00	935.00	-1.00	984.00	-0.00	0	638.00	0.15

7500 rows × 19 columns

计算AUC 和 KS

[77]:

plot_roc_ks(data_train,scorecard)

../_images/tutorials_tutorials_assess_7_1.png

根据预测得分计算ks

[78]:

data_proba = get_predict_score(data_train,
                                       scorecard,
                                       init_score=600,
                                       pdo=50,
                                       odds=0,
                                       target='y',
                                       precision=2)


plot_ks(data_train,data_proba)

../_images/tutorials_tutorials_assess_9_1.png

[79]:

# 计算auc 和 ks，并返回 ks 结果数据
plot_roc_ks(data_train,scorecard,return_data=True,precision=4)

../_images/tutorials_tutorials_assess_10_1.png

[79]:

	No.	fpr	tpr	thresholds	ks
0	1	0.00	0.01	0.82	0.01
1	2	0.04	0.32	0.51	0.28
2	3	0.12	0.49	0.31	0.37
3	4	0.22	0.61	0.20	0.39
4	5	0.32	0.70	0.17	0.38
5	6	0.45	0.78	0.15	0.33
6	7	0.54	0.83	0.14	0.30
7	8	0.65	0.89	0.12	0.25
8	9	0.77	0.94	0.10	0.17
9	10	0.91	0.98	0.08	0.07
10	11	1.00	1.00	0.04	0.00

AUC 计算

[80]:

#根据训练集和评分卡计算auc
get_auc_by_card(data_train,scorecard)

[80]:

0.76

[81]:

#根据训练集预测得分，计算auc
get_auc(data_train_score)

[81]:

0.76

KS 计算

[82]:

#根据训练集和评分卡计算ks
get_ks_by_card(data_train,scorecard)

[82]:

0.39

[83]:

#根据训练集预测得分，计算ks
get_ks(data_train_score)

[83]:

0.39

[84]:

#根据训练集预测得分，计算ks，并返回数据
ks,ks_data = get_ks(data_train_score,return_data=True)
ks_data

[84]:

	No.	Proba	#Total	#Bad	#Good	%Total	%Bad	%Good	%BadRate	%CumBad	%CumGood	KS
0	1	(0.039, 0.08]	2834	187	2647	12.60%	3.79%	15.07%	6.60%	3.79%	15.07%	0.11
1	2	(0.08, 0.1]	2360	206	2154	10.49%	4.17%	12.26%	8.73%	7.96%	27.33%	0.19
2	3	(0.1, 0.12]	2245	223	2022	9.98%	4.52%	11.51%	9.93%	12.48%	38.85%	0.26
3	4	(0.12, 0.14]	3076	437	2639	13.67%	8.85%	15.03%	14.21%	21.33%	53.87%	0.33
4	5	(0.14, 0.15]	2080	292	1788	9.24%	5.92%	10.18%	14.04%	27.25%	64.05%	0.37
5	6	(0.15, 0.17]	1260	213	1047	5.60%	4.32%	5.96%	16.90%	31.56%	70.01%	0.38
6	7	(0.17, 0.2]	2296	490	1806	10.20%	9.93%	10.28%	21.34%	41.49%	80.29%	0.39
7	8	(0.2, 0.31]	1894	488	1406	8.42%	9.89%	8.01%	25.77%	51.38%	88.30%	0.37
8	9	(0.31, 0.511]	2205	844	1361	9.80%	17.10%	7.75%	38.28%	68.48%	96.05%	0.28
9	10	(0.511, 0.82]	2250	1556	694	10.00%	31.52%	3.95%	69.16%	100.00%	100.00%	0.00

查看评分卡分数分布和提升度

[85]:

# 查看训练集评分卡分数分布 和 提升度
score_dist(data_train_score)

[85]:

	No.	Score Range	#Total	#Bad	#Good	%Total	%Bad	%Good	%BadRate	%BadRate Random	%CumBad	%CumTotal	Lift
0	1	(-inf, 503.9]	2250	1556	694	10.00%	31.52%	3.95%	69.16%	21.94%	69.16%	10.00%	6.92
1	2	(503.9, 567.0]	2276	863	1413	10.12%	17.48%	8.04%	37.92%	21.94%	53.45%	20.12%	2.66
2	3	(567.0, 610.0]	2346	600	1746	10.43%	12.16%	9.94%	25.58%	21.94%	43.93%	30.54%	1.44
3	4	(610.0, 624.0]	2232	433	1799	9.92%	8.77%	10.24%	19.40%	21.94%	37.92%	40.46%	0.94
4	5	(624.0, 635.0]	2615	395	2220	11.62%	8.00%	12.64%	15.11%	21.94%	32.83%	52.08%	0.63
5	6	(635.0, 642.0]	1823	267	1556	8.10%	5.41%	8.86%	14.65%	21.94%	30.38%	60.19%	0.50
6	7	(642.0, 653.0]	2216	292	1924	9.85%	5.92%	10.95%	13.18%	21.94%	27.96%	70.04%	0.40
7	8	(653.0, 666.0]	2364	210	2154	10.51%	4.25%	12.26%	8.88%	21.94%	25.47%	80.54%	0.32
8	9	(666.0, 686.0]	2748	231	2517	12.21%	4.68%	14.33%	8.41%	21.94%	23.22%	92.76%	0.25
9	10	(686.0, inf]	1630	89	1541	7.24%	1.80%	8.77%	5.46%	21.94%	21.94%	100.00%	0.22

[86]:

# 查看训练集评分卡分数分布 和 提升度，分为5组，查看提升度
score_dist(data_train_score,qcut=5)

[86]:

	No.	Score Range	#Total	#Bad	#Good	%Total	%Bad	%Good	%BadRate	%BadRate Random	%CumBad	%CumTotal	Lift
0	1	(-inf, 567.0]	4526	2419	2107	20.12%	49.01%	12.00%	53.45%	21.94%	53.45%	20.12%	2.66
1	2	(567.0, 624.0]	4578	1033	3545	20.35%	20.93%	20.18%	22.56%	21.94%	37.92%	40.46%	0.94
2	3	(624.0, 642.0]	4438	662	3776	19.72%	13.41%	21.50%	14.92%	21.94%	30.38%	60.19%	0.50
3	4	(642.0, 666.0]	4580	502	4078	20.36%	10.17%	23.22%	10.96%	21.94%	25.47%	80.54%	0.32
4	5	(666.0, inf]	4378	320	4058	19.46%	6.48%	23.10%	7.31%	21.94%	21.94%	100.00%	0.22

[87]:

# 查看训练集评分卡分数分布 和 提升度，分为5组，查看提升度
view_score_dist(data_train_score,qcut=5)

[87]:

	No.	Score Range	#Total	#Bad	#Good	%Total	%Bad	%Good	%BadRate	%BadRate Random	%CumBad	%CumTotal	Lift	Lift.
0	1	(-inf, 567.0]	4526	2419	2107	20.12%	49.01%	12.00%	53.45%	21.94%	53.45%	20.12%	2.660000	2.660000
1	2	(567.0, 624.0]	4578	1033	3545	20.35%	20.93%	20.18%	22.56%	21.94%	37.92%	40.46%	0.940000	0.940000
2	3	(624.0, 642.0]	4438	662	3776	19.72%	13.41%	21.50%	14.92%	21.94%	30.38%	60.19%	0.500000	0.500000
3	4	(642.0, 666.0]	4580	502	4078	20.36%	10.17%	23.22%	10.96%	21.94%	25.47%	80.54%	0.320000	0.320000
4	5	(666.0, inf]	4378	320	4058	19.46%	6.48%	23.10%	7.31%	21.94%	21.94%	100.00%	0.220000	0.220000

[88]:

# 查看训练集评分卡分数分布 和 提升度，分为5组，查看提升度
view_score_dist(data_train_score,qcut=5,color='green')

[88]:

	No.	Score Range	#Total	#Bad	#Good	%Total	%Bad	%Good	%BadRate	%BadRate Random	%CumBad	%CumTotal	Lift	Lift.
0	1	(-inf, 567.0]	4526	2419	2107	20.12%	49.01%	12.00%	53.45%	21.94%	53.45%	20.12%	2.660000	2.660000
1	2	(567.0, 624.0]	4578	1033	3545	20.35%	20.93%	20.18%	22.56%	21.94%	37.92%	40.46%	0.940000	0.940000
2	3	(624.0, 642.0]	4438	662	3776	19.72%	13.41%	21.50%	14.92%	21.94%	30.38%	60.19%	0.500000	0.500000
3	4	(642.0, 666.0]	4580	502	4078	20.36%	10.17%	23.22%	10.96%	21.94%	25.47%	80.54%	0.320000	0.320000
4	5	(666.0, inf]	4378	320	4058	19.46%	6.48%	23.10%	7.31%	21.94%	21.94%	100.00%	0.220000	0.220000

[89]:

# 查看训练集评分卡分数分布 和 提升度，分为5组，查看提升度
view_score_dist(data_train_score,qcut=5,color='#02B057')

[89]:

	No.	Score Range	#Total	#Bad	#Good	%Total	%Bad	%Good	%BadRate	%BadRate Random	%CumBad	%CumTotal	Lift	Lift.
0	1	(-inf, 567.0]	4526	2419	2107	20.12%	49.01%	12.00%	53.45%	21.94%	53.45%	20.12%	2.660000	2.660000
1	2	(567.0, 624.0]	4578	1033	3545	20.35%	20.93%	20.18%	22.56%	21.94%	37.92%	40.46%	0.940000	0.940000
2	3	(624.0, 642.0]	4438	662	3776	19.72%	13.41%	21.50%	14.92%	21.94%	30.38%	60.19%	0.500000	0.500000
3	4	(642.0, 666.0]	4580	502	4078	20.36%	10.17%	23.22%	10.96%	21.94%	25.47%	80.54%	0.320000	0.320000
4	5	(666.0, inf]	4378	320	4058	19.46%	6.48%	23.10%	7.31%	21.94%	21.94%	100.00%	0.220000	0.220000

[90]:

# 查看训练集评分卡分数分布 和 提升度,  标题显示为中文
score_dist(data_train_score,language='cn')

[90]:

	序号	分数区间	#合计	#坏	#好	%合计	%坏	%好	%坏件率	%随机坏件率	%累计坏	%累计合计	提升度
0	1	(-inf, 503.9]	2250	1556	694	10.00%	31.52%	3.95%	69.16%	21.94%	69.16%	10.00%	6.92
1	2	(503.9, 567.0]	2276	863	1413	10.12%	17.48%	8.04%	37.92%	21.94%	53.45%	20.12%	2.66
2	3	(567.0, 610.0]	2346	600	1746	10.43%	12.16%	9.94%	25.58%	21.94%	43.93%	30.54%	1.44
3	4	(610.0, 624.0]	2232	433	1799	9.92%	8.77%	10.24%	19.40%	21.94%	37.92%	40.46%	0.94
4	5	(624.0, 635.0]	2615	395	2220	11.62%	8.00%	12.64%	15.11%	21.94%	32.83%	52.08%	0.63
5	6	(635.0, 642.0]	1823	267	1556	8.10%	5.41%	8.86%	14.65%	21.94%	30.38%	60.19%	0.50
6	7	(642.0, 653.0]	2216	292	1924	9.85%	5.92%	10.95%	13.18%	21.94%	27.96%	70.04%	0.40
7	8	(653.0, 666.0]	2364	210	2154	10.51%	4.25%	12.26%	8.88%	21.94%	25.47%	80.54%	0.32
8	9	(666.0, 686.0]	2748	231	2517	12.21%	4.68%	14.33%	8.41%	21.94%	23.22%	92.76%	0.25
9	10	(686.0, inf]	1630	89	1541	7.24%	1.80%	8.77%	5.46%	21.94%	21.94%	100.00%	0.22

[91]:

# 绘制提升度图
plot_lift(data_train_score)

../_images/tutorials_tutorials_assess_25_0.png

[92]:

# 绘制提升度图，并输出提升度表格
plot_lift(data_train_score,return_data=True)

../_images/tutorials_tutorials_assess_26_0.png

[92]:

	No.	Score Range	#Total	#Bad	#Good	%Total	%Bad	%Good	%BadRate	%BadRate Random	%CumBad	%CumTotal	Lift
0	1	(-inf, 503.9]	2250	1556	694	10.00%	31.52%	3.95%	69.16%	21.94%	69.16%	10.00%	6.92
1	2	(503.9, 567.0]	2276	863	1413	10.12%	17.48%	8.04%	37.92%	21.94%	53.45%	20.12%	2.66
2	3	(567.0, 610.0]	2346	600	1746	10.43%	12.16%	9.94%	25.58%	21.94%	43.93%	30.54%	1.44
3	4	(610.0, 624.0]	2232	433	1799	9.92%	8.77%	10.24%	19.40%	21.94%	37.92%	40.46%	0.94
4	5	(624.0, 635.0]	2615	395	2220	11.62%	8.00%	12.64%	15.11%	21.94%	32.83%	52.08%	0.63
5	6	(635.0, 642.0]	1823	267	1556	8.10%	5.41%	8.86%	14.65%	21.94%	30.38%	60.19%	0.50
6	7	(642.0, 653.0]	2216	292	1924	9.85%	5.92%	10.95%	13.18%	21.94%	27.96%	70.04%	0.40
7	8	(653.0, 666.0]	2364	210	2154	10.51%	4.25%	12.26%	8.88%	21.94%	25.47%	80.54%	0.32
8	9	(666.0, 686.0]	2748	231	2517	12.21%	4.68%	14.33%	8.41%	21.94%	23.22%	92.76%	0.25
9	10	(686.0, inf]	1630	89	1541	7.24%	1.80%	8.77%	5.46%	21.94%	21.94%	100.00%	0.22

[93]:

# 查看训练集分数分布

import matplotlib.pyplot as plt
plt.style.use('default')
fig=plt.figure()
ax1=fig.add_subplot(1,1,1)
ax1.hist(data_train_score['Score'] )
plt.show()

../_images/tutorials_tutorials_assess_27_0.png

模型稳定型 PSI计算

[94]:

data_train_score = get_predict_score(data_train,scorecard)
data_test_score = get_predict_score(data_test,scorecard)

# 按照等频分箱，分为10组，计算模型得分 Score 的PSI
get_psi(data_train_score,data_test_score)

[94]:

	No.	Name	Bins Range	#Total	#Actual	#Expected	%Total	%Actual	%Expected	PSI	Total PSI
0	1	Score	(-inf, 503.9]	3018	2250	768	10.06%	10.00%	10.24%	0.00	0.07
1	2	Score	(503.9, 567.0]	3026	2276	750	10.09%	10.12%	10.00%	0.00	0.07
2	3	Score	(567.0, 610.0]	3018	2346	672	10.06%	10.43%	8.96%	0.00	0.07
3	4	Score	(610.0, 624.0]	2945	2232	713	9.82%	9.92%	9.51%	0.00	0.07
4	5	Score	(624.0, 635.0]	3065	2615	450	10.22%	11.62%	6.00%	0.04	0.07
5	6	Score	(635.0, 642.0]	2706	1823	883	9.02%	8.10%	11.77%	0.01	0.07
6	7	Score	(642.0, 653.0]	3054	2216	838	10.18%	9.85%	11.17%	0.00	0.07
7	8	Score	(653.0, 666.0]	3151	2364	787	10.50%	10.51%	10.49%	0.00	0.07
8	9	Score	(666.0, 686.0]	3555	2748	807	11.85%	12.21%	10.76%	0.00	0.07
9	10	Score	(686.0, inf]	2462	1630	832	8.21%	7.24%	11.09%	0.02	0.07

[95]:

# 按照等频分箱，分为10组，计算模型得分 Score 的PSI
view_psi(data_train_score,data_test_score)

[95]:

	No.	Name	Bins Range	#Total	#Actual	#Expected	%Total	%Actual	%Expected	PSI	Total PSI	PSI.
0	1	Score	(-inf, 503.9]	3018	2250	768	10.06%	10.00%	10.24%	0.000000	0.070000	0.000000
1	2	Score	(503.9, 567.0]	3026	2276	750	10.09%	10.12%	10.00%	0.000000	0.070000	0.000000
2	3	Score	(567.0, 610.0]	3018	2346	672	10.06%	10.43%	8.96%	0.000000	0.070000	0.000000
3	4	Score	(610.0, 624.0]	2945	2232	713	9.82%	9.92%	9.51%	0.000000	0.070000	0.000000
4	5	Score	(624.0, 635.0]	3065	2615	450	10.22%	11.62%	6.00%	0.040000	0.070000	0.040000
5	6	Score	(635.0, 642.0]	2706	1823	883	9.02%	8.10%	11.77%	0.010000	0.070000	0.010000
6	7	Score	(642.0, 653.0]	3054	2216	838	10.18%	9.85%	11.17%	0.000000	0.070000	0.000000
7	8	Score	(653.0, 666.0]	3151	2364	787	10.50%	10.51%	10.49%	0.000000	0.070000	0.000000
8	9	Score	(666.0, 686.0]	3555	2748	807	11.85%	12.21%	10.76%	0.000000	0.070000	0.000000
9	10	Score	(686.0, inf]	2462	1630	832	8.21%	7.24%	11.09%	0.020000	0.070000	0.020000

[96]:

# 按照等频分箱，分为10组，计算模型得分 Score 的PSI
view_psi(data_train_score,data_test_score,color='green')

[96]:

	No.	Name	Bins Range	#Total	#Actual	#Expected	%Total	%Actual	%Expected	PSI	Total PSI	PSI.
0	1	Score	(-inf, 503.9]	3018	2250	768	10.06%	10.00%	10.24%	0.000000	0.070000	0.000000
1	2	Score	(503.9, 567.0]	3026	2276	750	10.09%	10.12%	10.00%	0.000000	0.070000	0.000000
2	3	Score	(567.0, 610.0]	3018	2346	672	10.06%	10.43%	8.96%	0.000000	0.070000	0.000000
3	4	Score	(610.0, 624.0]	2945	2232	713	9.82%	9.92%	9.51%	0.000000	0.070000	0.000000
4	5	Score	(624.0, 635.0]	3065	2615	450	10.22%	11.62%	6.00%	0.040000	0.070000	0.040000
5	6	Score	(635.0, 642.0]	2706	1823	883	9.02%	8.10%	11.77%	0.010000	0.070000	0.010000
6	7	Score	(642.0, 653.0]	3054	2216	838	10.18%	9.85%	11.17%	0.000000	0.070000	0.000000
7	8	Score	(653.0, 666.0]	3151	2364	787	10.50%	10.51%	10.49%	0.000000	0.070000	0.000000
8	9	Score	(666.0, 686.0]	3555	2748	807	11.85%	12.21%	10.76%	0.000000	0.070000	0.000000
9	10	Score	(686.0, inf]	2462	1630	832	8.21%	7.24%	11.09%	0.020000	0.070000	0.020000

[97]:

# 按照等频分箱，分为10组，计算模型得分 Score 的PSI
view_psi(data_train_score,data_test_score,color='#02B057')

[97]:

	No.	Name	Bins Range	#Total	#Actual	#Expected	%Total	%Actual	%Expected	PSI	Total PSI	PSI.
0	1	Score	(-inf, 503.9]	3018	2250	768	10.06%	10.00%	10.24%	0.000000	0.070000	0.000000
1	2	Score	(503.9, 567.0]	3026	2276	750	10.09%	10.12%	10.00%	0.000000	0.070000	0.000000
2	3	Score	(567.0, 610.0]	3018	2346	672	10.06%	10.43%	8.96%	0.000000	0.070000	0.000000
3	4	Score	(610.0, 624.0]	2945	2232	713	9.82%	9.92%	9.51%	0.000000	0.070000	0.000000
4	5	Score	(624.0, 635.0]	3065	2615	450	10.22%	11.62%	6.00%	0.040000	0.070000	0.040000
5	6	Score	(635.0, 642.0]	2706	1823	883	9.02%	8.10%	11.77%	0.010000	0.070000	0.010000
6	7	Score	(642.0, 653.0]	3054	2216	838	10.18%	9.85%	11.17%	0.000000	0.070000	0.000000
7	8	Score	(653.0, 666.0]	3151	2364	787	10.50%	10.51%	10.49%	0.000000	0.070000	0.000000
8	9	Score	(666.0, 686.0]	3555	2748	807	11.85%	12.21%	10.76%	0.000000	0.070000	0.000000
9	10	Score	(686.0, inf]	2462	1630	832	8.21%	7.24%	11.09%	0.020000	0.070000	0.020000

[98]:

data_train_score = get_predict_score(data_train,scorecard)
data_test_score = get_predict_score(data_test,scorecard)

# 按照等频分箱，分为5组，计算PSI
get_psi(data_train_score,data_test_score,col='LIMIT_BAL',qcut=5,precision=4)

[98]:

	No.	Name	Bins Range	#Total	#Actual	#Expected	%Total	%Actual	%Expected	PSI	Total PSI
0	1	LIMIT_BAL	(-inf, 50000.0]	7676	5855	1821	25.5867%	26.0222%	24.2800%	0.00	0.00
1	2	LIMIT_BAL	(50000.0, 100000.0]	4822	3593	1229	16.0733%	15.9689%	16.3867%	0.00	0.00
2	3	LIMIT_BAL	(100000.0, 180000.0]	6123	4548	1575	20.4100%	20.2133%	21.0000%	0.00	0.00
3	4	LIMIT_BAL	(180000.0, 270000.0]	5421	4035	1386	18.0700%	17.9333%	18.4800%	0.00	0.00
4	5	LIMIT_BAL	(270000.0, inf]	5958	4469	1489	19.8600%	19.8622%	19.8533%	0.00	0.00

[99]:

data_train_score = get_predict_score(data_train,scorecard)
data_test_score = get_predict_score(data_test,scorecard)

# 按照指定的切分点，计算PSI
get_psi(data_train_score,data_test_score,col='LIMIT_BAL',bins=[-inf,30000,50000,100000,inf],precision=4)

[99]:

	No.	Name	Bins Range	#Total	#Actual	#Expected	%Total	%Actual	%Expected	PSI	Total PSI
0	1	LIMIT_BAL	(-inf, 30000.0]	4111	3109	1002	13.7033%	13.8178%	13.3600%	0.00	0.00
1	2	LIMIT_BAL	(30000.0, 50000.0]	3649	2746	903	12.1633%	12.2044%	12.0400%	0.00	0.00
2	3	LIMIT_BAL	(50000.0, 100000.0]	4794	3593	1201	15.9800%	15.9689%	16.0133%	0.00	0.00
3	4	LIMIT_BAL	(100000.0, inf]	17446	13052	4394	58.1533%	58.0089%	58.5867%	0.00	0.00

[100]:

data_train_score = get_predict_score(data_train,scorecard)
data_test_score = get_predict_score(data_test,scorecard)

# 按照指定的切分点，计算PSI，标题显示为中文
get_psi(data_train_score,data_test_score,col='LIMIT_BAL',bins=[-inf,30000,50000,100000,inf],precision=4,language='cn')

[100]:

	序号	名称	分组	#合计	#实际	#期望	%合计	%实际	%期望	PSI	PSI 合计
0	1	LIMIT_BAL	(-inf, 30000.0]	4111	3109	1002	13.7033%	13.8178%	13.3600%	0.00	0.00
1	2	LIMIT_BAL	(30000.0, 50000.0]	3649	2746	903	12.1633%	12.2044%	12.0400%	0.00	0.00
2	3	LIMIT_BAL	(50000.0, 100000.0]	4794	3593	1201	15.9800%	15.9689%	16.0133%	0.00	0.00
3	4	LIMIT_BAL	(100000.0, inf]	17446	13052	4394	58.1533%	58.0089%	58.5867%	0.00	0.00

[101]:

# 批量计算所有特征的PSI
data_train_score = get_predict_score(data_train,scorecard)
data_test_score = get_predict_score(data_test,scorecard)

get_data_psi(data_train_score,data_test_score,precision=4)

[101]:

	Name	PSI
16	Score	0.07
8	PAY_AMT3	0.00
4	LIMIT_BAL	0.00
17	Proba	0.00
6	PAY_AMT2	0.00
14	PAY_AMT6	0.00
10	PAY_AMT4	0.00
12	PAY_AMT5	0.00
2	PAY_AMT1	0.00
7	PAY_AMT2_Score	0.00
13	PAY_AMT5_Score	0.00
0	PAY_0	0.00
1	PAY_0_Score	0.00
5	LIMIT_BAL_Score	0.00
9	PAY_AMT3_Score	0.00
15	PAY_AMT6_Score	0.00
11	PAY_AMT4_Score	0.00
3	PAY_AMT1_Score	0.00

模型应用分析

决策树分析

[102]:

plot_tree(data_train_score[['Score','y']],max_depth=3,criterion='gini')

[102]:

../_images/tutorials_tutorials_assess_39_0.svg

[103]:

# 绘制决策树，并保存图片到本地
# plot_tree(data_train_score[['Score','y']],max_depth=3,criterion='gini',out_file='test.svg')

[103]:

[104]:

# 绘制决策树，并保存图片到本地
# plot_tree(data_train_score[['Score','y']],max_depth=3,criterion='gini',out_file='test.pdf')

[104]:

Read the Docs v: latest

版本: latest

下载

托管于 Read the Docs: 项目主页; 构建