14. Train LGBM

In [1]:

import featurebyte as fb
import pandas as pd
from sklearn.metrics import roc_auc_score
from modeling_script import LightGBMPipeline, Objective, Metric

import featurebyte as fb import pandas as pd from sklearn.metrics import roc_auc_score from modeling_script import LightGBMPipeline, Objective, Metric

16:50:49 | WARNING  | Service endpoint is inaccessible: http://featurebyte-server:8088/

Activate Catalog¶

In [2]:

# Set your profile to the tutorial environment
fb.use_profile("tutorial")

catalog_name = "Credit Default Dataset SDK Tutorial"
catalog = fb.Catalog.activate(catalog_name)

# Set your profile to the tutorial environment fb.use_profile("tutorial") catalog_name = "Credit Default Dataset SDK Tutorial" catalog = fb.Catalog.activate(catalog_name)

16:51:05 | INFO     | Using profile: tutorial
16:51:05 | INFO     | Using configuration file at: /Users/gxav/.featurebyte/config.yaml
16:51:05 | INFO     | Active profile: tutorial (https://tutorials.featurebyte.com/api/v1)
16:51:05 | INFO     | SDK version: 2.1.0.dev113
16:51:05 | INFO     | No catalog activated.
16:51:05 | INFO     | Catalog activated: Credit Default Dataset SDK Tutorial

Get Training and Holdout data¶

In [3]:

catalog.list_historical_feature_tables()

catalog.list_historical_feature_tables()

Out[3]:

	id	name	feature_store_name	observation_table_name	shape	created_at
0	67c2c9a34e08d83e21381725	51 features for Credit Default - HOLDOUT_2024_1H	playground	CREDIT_DEFAULT_HOLDOUT_2024_1H	[25666, 54]	2025-03-01T08:49:21.154000
1	67c2c8fe3df413286793fb5a	51 features for Credit Default - TRAIN_2019_2023	playground	CREDIT_DEFAULT_TRAIN_2019_2023	[256273, 54]	2025-03-01T08:47:25.134000

In [4]:

training_data_table = catalog.get_historical_feature_table("51 features for Credit Default - TRAIN_2019_2023")
holdout_data_table = catalog.get_historical_feature_table("51 features for Credit Default - HOLDOUT_2024_1H")

training_data_table = catalog.get_historical_feature_table("51 features for Credit Default - TRAIN_2019_2023") holdout_data_table = catalog.get_historical_feature_table("51 features for Credit Default - HOLDOUT_2024_1H")

In [5]:

# download as pandas data frame
training_data = training_data_table.to_pandas()
holdout_data = holdout_data_table.to_pandas()

# download as pandas data frame training_data = training_data_table.to_pandas() holdout_data = holdout_data_table.to_pandas()

Downloading table |████████████████████████████████████████| 256273/256273 [100%
Downloading table |████████████████████████████████████████| 25666/25666 [100%] 

Categorize per feature type¶

In [6]:

feature_types = {
    "NEW_APPLICATION_AMT_ANNUITY": "numeric",
    "NEW_APPLICATION_AMT_ANNUITY_To_AMT_CREDIT": "numeric",
    "NEW_APPLICATION_AMT_ANNUITY_To_AMT_GOODS_VALUE": "numeric",
    "NEW_APPLICATION_AMT_CREDIT": "numeric",
    "NEW_APPLICATION_AMT_GOODS_VALUE_To_AMT_CREDIT": "numeric",
    "NEW_APPLICATION_AMT_REQ_CREDIT_BUREAU_QRT": "numeric",
    "NEW_APPLICATION_Credit-Goods_Gap": "numeric",
    "NEW_APPLICATION_DAYS_EMPLOYED": "numeric",
    "NEW_APPLICATION_DAYS_LAST_PHONE_CHANGE": "numeric",
    "NEW_APPLICATION_DAYS_REGISTRATION": "numeric",
    "NEW_APPLICATION_FLAG_DOCUMENT_3": "numeric",
    "NEW_APPLICATION_FLOORSMAX_MEDI": "numeric",
    "NEW_APPLICATION_REGION_POPULATION_RELATIVE": "numeric",
    "NEW_APPLICATION_Time_To_BIRTHDATE": "numeric",
    "CLIENT_Max_of_Consumer_Loan_terminations_Consumer_Loan_AMT_ANNUITYs_104w": "numeric",
    "CLIENT_Avg_of_Consumer_installments_records_AMT_PAYMENTs_24cMo": "numeric",
    "CLIENT_Avg_of_Late_Status_Consumer_installments_records_AMT_PAYMENTs_24cMo": "numeric",
    "CLIENT_Min_of_Consumer_installments_records_AMT_PAYMENTs_12cMo": "numeric",
    "CLIENT_Min_of_Consumer_installments_records_Consumer_Loan_AMT_CREDIT_To_AMT_APPLICATIONs_24cMo": "numeric",
    "CLIENT_Max_of_Consumer_installments_records_Consumer_Loan_AMT_APPLICATIONs_24cMo": "numeric",
    "CLIENT_Max_of_Consumer_installments_records_Payment_Delays_24cMo": "numeric",
    "CLIENT_Temporal_Mean_of_Sums_of_Consumer_installments_records_Consumer_Loan_AMT_ANNUITYs_24cMo": "numeric",
    "CLIENT_Avg_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_104w": "numeric",
    "CLIENT_Avg_of_Approved_Status_Prior_Applications_CNT_PAYMENTs_104w": "numeric",
    "CLIENT_Avg_of_Prior_Applications_CNT_PAYMENTs_104w": "numeric",
    "CLIENT_Max_of_Time_between_2_Approved_Status_Prior_Applications_for_the_Client_104w": "numeric",
    "CLIENT_Max_of_Time_between_2_Prior_Applications_for_the_Client_104w": "numeric",
    "CLIENT_Latest_Revolving_loans_Contract_type_Prior_Application_AMT_APPLICATION_104w": "numeric",
    "CLIENT_Min_of_Approved_Status_Cash_loans_Contract_type_Prior_Applications_AMT_ANNUITY_To_AMT_CREDITs_104w": "numeric",
    "CLIENT_Max_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_104w": "numeric",
    "CLIENT_Min_of_Cash_loans_Contract_type_Prior_Applications_CNT_PAYMENTs_104w": "numeric",
    "CLIENT_Max_of_Prior_Applications_AMT_CREDIT_To_AMT_APPLICATIONs_104w": "numeric",
    "CLIENT_Max_of_Prior_Applications_Application-Credit_Gaps_104w": "numeric",
    "CLIENT_Max_of_Prior_Applications_CNT_PAYMENTs_104w": "numeric",
    "CLIENT_Pct_of_Prior_Applications_is_Refused_Status_52w": "numeric",
    "CLIENT_Pct_of_Prior_Applications_is_Refused_Status_Cash_loans_Contract_type_104w": "numeric",
    "CLIENT_Sum_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_To_Sum_of_Approved_Status_Prior_Applications_AMT_CREDITs_104w": "numeric",
    "CLIENT_Time_Since_Latest_Prior_Application_DECISION_DATE_104w": "numeric",
    "CLIENT_vs_OVERALL_Count_of_Approved_Status_Cash_loans_Contract_type_Prior_Applications_by_Prior_Application_PRODUCT_COMBINATION_104w": "numeric",
    "NEW_APPLICATION_CODE_GENDER": "categorical",
    "NEW_APPLICATION_EDUCATION_TYPE": "categorical",
    "NEW_APPLICATION_INCOME_TYPE": "categorical",
    "NEW_APPLICATION_OCCUPATION_TYPE": "categorical",
    "NEW_APPLICATION_ORGANIZATION_TYPE": "categorical",
    "CLIENT_Consumer_Loan_GOODS_CATEGORY_with_Lowest_sum_of_active_Consumer_Loans_AMT_CREDITs": "categorical",
    "CLIENT_Latest_Refused_Status_Cash_loans_Contract_type_Prior_Application_YIELD_GROUP_104w": "categorical",
    "CLIENT_Consumer_installments_records_AMT_PAYMENTs_by_Consumer_installments_record_INSTALLMENT_STATUS_6cMo": "dictionary",
    "CLIENT_Count_of_Consumer_installments_records_by_Consumer_Loan_YIELD_GROUP_24cMo": "dictionary",
    "CLIENT_Approved_Status_Prior_Applications_AMT_CREDITs_by_Prior_Application_YIELD_GROUP_104w": "dictionary",
    "CLIENT_Count_of_Cash_loans_Contract_type_Prior_Applications_by_Prior_Application_YIELD_GROUP_104w": "dictionary",
    "CLIENT_Count_of_Prior_Applications_by_Prior_Application_YIELD_GROUP_104w": "dictionary",
}

feature_types = { "NEW_APPLICATION_AMT_ANNUITY": "numeric", "NEW_APPLICATION_AMT_ANNUITY_To_AMT_CREDIT": "numeric", "NEW_APPLICATION_AMT_ANNUITY_To_AMT_GOODS_VALUE": "numeric", "NEW_APPLICATION_AMT_CREDIT": "numeric", "NEW_APPLICATION_AMT_GOODS_VALUE_To_AMT_CREDIT": "numeric", "NEW_APPLICATION_AMT_REQ_CREDIT_BUREAU_QRT": "numeric", "NEW_APPLICATION_Credit-Goods_Gap": "numeric", "NEW_APPLICATION_DAYS_EMPLOYED": "numeric", "NEW_APPLICATION_DAYS_LAST_PHONE_CHANGE": "numeric", "NEW_APPLICATION_DAYS_REGISTRATION": "numeric", "NEW_APPLICATION_FLAG_DOCUMENT_3": "numeric", "NEW_APPLICATION_FLOORSMAX_MEDI": "numeric", "NEW_APPLICATION_REGION_POPULATION_RELATIVE": "numeric", "NEW_APPLICATION_Time_To_BIRTHDATE": "numeric", "CLIENT_Max_of_Consumer_Loan_terminations_Consumer_Loan_AMT_ANNUITYs_104w": "numeric", "CLIENT_Avg_of_Consumer_installments_records_AMT_PAYMENTs_24cMo": "numeric", "CLIENT_Avg_of_Late_Status_Consumer_installments_records_AMT_PAYMENTs_24cMo": "numeric", "CLIENT_Min_of_Consumer_installments_records_AMT_PAYMENTs_12cMo": "numeric", "CLIENT_Min_of_Consumer_installments_records_Consumer_Loan_AMT_CREDIT_To_AMT_APPLICATIONs_24cMo": "numeric", "CLIENT_Max_of_Consumer_installments_records_Consumer_Loan_AMT_APPLICATIONs_24cMo": "numeric", "CLIENT_Max_of_Consumer_installments_records_Payment_Delays_24cMo": "numeric", "CLIENT_Temporal_Mean_of_Sums_of_Consumer_installments_records_Consumer_Loan_AMT_ANNUITYs_24cMo": "numeric", "CLIENT_Avg_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_104w": "numeric", "CLIENT_Avg_of_Approved_Status_Prior_Applications_CNT_PAYMENTs_104w": "numeric", "CLIENT_Avg_of_Prior_Applications_CNT_PAYMENTs_104w": "numeric", "CLIENT_Max_of_Time_between_2_Approved_Status_Prior_Applications_for_the_Client_104w": "numeric", "CLIENT_Max_of_Time_between_2_Prior_Applications_for_the_Client_104w": "numeric", "CLIENT_Latest_Revolving_loans_Contract_type_Prior_Application_AMT_APPLICATION_104w": "numeric", "CLIENT_Min_of_Approved_Status_Cash_loans_Contract_type_Prior_Applications_AMT_ANNUITY_To_AMT_CREDITs_104w": "numeric", "CLIENT_Max_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_104w": "numeric", "CLIENT_Min_of_Cash_loans_Contract_type_Prior_Applications_CNT_PAYMENTs_104w": "numeric", "CLIENT_Max_of_Prior_Applications_AMT_CREDIT_To_AMT_APPLICATIONs_104w": "numeric", "CLIENT_Max_of_Prior_Applications_Application-Credit_Gaps_104w": "numeric", "CLIENT_Max_of_Prior_Applications_CNT_PAYMENTs_104w": "numeric", "CLIENT_Pct_of_Prior_Applications_is_Refused_Status_52w": "numeric", "CLIENT_Pct_of_Prior_Applications_is_Refused_Status_Cash_loans_Contract_type_104w": "numeric", "CLIENT_Sum_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_To_Sum_of_Approved_Status_Prior_Applications_AMT_CREDITs_104w": "numeric", "CLIENT_Time_Since_Latest_Prior_Application_DECISION_DATE_104w": "numeric", "CLIENT_vs_OVERALL_Count_of_Approved_Status_Cash_loans_Contract_type_Prior_Applications_by_Prior_Application_PRODUCT_COMBINATION_104w": "numeric", "NEW_APPLICATION_CODE_GENDER": "categorical", "NEW_APPLICATION_EDUCATION_TYPE": "categorical", "NEW_APPLICATION_INCOME_TYPE": "categorical", "NEW_APPLICATION_OCCUPATION_TYPE": "categorical", "NEW_APPLICATION_ORGANIZATION_TYPE": "categorical", "CLIENT_Consumer_Loan_GOODS_CATEGORY_with_Lowest_sum_of_active_Consumer_Loans_AMT_CREDITs": "categorical", "CLIENT_Latest_Refused_Status_Cash_loans_Contract_type_Prior_Application_YIELD_GROUP_104w": "categorical", "CLIENT_Consumer_installments_records_AMT_PAYMENTs_by_Consumer_installments_record_INSTALLMENT_STATUS_6cMo": "dictionary", "CLIENT_Count_of_Consumer_installments_records_by_Consumer_Loan_YIELD_GROUP_24cMo": "dictionary", "CLIENT_Approved_Status_Prior_Applications_AMT_CREDITs_by_Prior_Application_YIELD_GROUP_104w": "dictionary", "CLIENT_Count_of_Cash_loans_Contract_type_Prior_Applications_by_Prior_Application_YIELD_GROUP_104w": "dictionary", "CLIENT_Count_of_Prior_Applications_by_Prior_Application_YIELD_GROUP_104w": "dictionary", }

Run LightGBM¶

In [7]:

target_train = training_data.Loan_Default
features_train = training_data.drop(columns=['POINT_IN_TIME', 'NEW_APPLICATION_ID', 'Loan_Default'])

target_test = holdout_data.Loan_Default
features_test = holdout_data.drop(columns=['POINT_IN_TIME', 'NEW_APPLICATION_ID', 'Loan_Default'])

lgbm_pipeline = LightGBMPipeline(objective=Objective.BINARY, eval_metric=Metric.AUC)
lgbm_pipeline.train(
    df_train=features_train,
    df_test=features_test,
    y_train=target_train,
    y_test=target_test,
    feature_types=feature_types,
)
predictions = lgbm_pipeline.predict(features_test)
print(f"AUC: {roc_auc_score(target_test, predictions)}")

target_train = training_data.Loan_Default features_train = training_data.drop(columns=['POINT_IN_TIME', 'NEW_APPLICATION_ID', 'Loan_Default']) target_test = holdout_data.Loan_Default features_test = holdout_data.drop(columns=['POINT_IN_TIME', 'NEW_APPLICATION_ID', 'Loan_Default']) lgbm_pipeline = LightGBMPipeline(objective=Objective.BINARY, eval_metric=Metric.AUC) lgbm_pipeline.train( df_train=features_train, df_test=features_test, y_train=target_train, y_test=target_test, feature_types=feature_types, ) predictions = lgbm_pipeline.predict(features_test) print(f"AUC: {roc_auc_score(target_test, predictions)}")

AUC: 0.7405204792668864

SHAP Importance¶

In [8]:

pd.set_option('display.max_rows', None)  # Show all rows
pd.set_option('display.max_colwidth', 100)  # Show full column content

pd.set_option('display.max_rows', None) # Show all rows pd.set_option('display.max_colwidth', 100) # Show full column content

In [9]:

_, feature_importance = lgbm_pipeline.compute_shap(features_train.sample(n=10000, random_state=88))
feature_importance[["feature", "cumulative_importance_percent"]]

_, feature_importance = lgbm_pipeline.compute_shap(features_train.sample(n=10000, random_state=88)) feature_importance[["feature", "cumulative_importance_percent"]]

Out[9]:

	feature	cumulative_importance_percent
0	NEW_APPLICATION_DAYS_EMPLOYED	0.067910
1	NEW_APPLICATION_AMT_ANNUITY_To_AMT_CREDIT	0.120227
2	NEW_APPLICATION_CODE_GENDER	0.171455
3	NEW_APPLICATION_DAYS_LAST_PHONE_CHANGE	0.220847
4	NEW_APPLICATION_EDUCATION_TYPE	0.265415
5	NEW_APPLICATION_Time_To_BIRTHDATE	0.307860
6	NEW_APPLICATION_REGION_POPULATION_RELATIVE	0.346151
7	NEW_APPLICATION_AMT_GOODS_VALUE_To_AMT_CREDIT	0.380739
8	NEW_APPLICATION_Credit-Goods_Gap	0.407075
9	NEW_APPLICATION_AMT_ANNUITY_To_AMT_GOODS_VALUE	0.433297
10	NEW_APPLICATION_FLOORSMAX_MEDI	0.459445
11	CLIENT_Max_of_Prior_Applications_AMT_CREDIT_To_AMT_APPLICATIONs_104w	0.484454
12	NEW_APPLICATION_AMT_ANNUITY	0.508202
13	CLIENT_Count_of_Consumer_installments_records_by_Consumer_Loan_YIELD_GROUP_24cMo	0.528123
14	CLIENT_Consumer_installments_records_AMT_PAYMENTs_by_Consumer_installments_record_INSTALLMENT_ST...	0.547329
15	CLIENT_Latest_Revolving_loans_Contract_type_Prior_Application_AMT_APPLICATION_104w	0.566244
16	CLIENT_Max_of_Consumer_installments_records_Payment_Delays_24cMo	0.584885
17	CLIENT_Max_of_Consumer_Loan_terminations_Consumer_Loan_AMT_ANNUITYs_104w	0.602186
18	CLIENT_Count_of_Prior_Applications_by_Prior_Application_YIELD_GROUP_104w	0.619358
19	CLIENT_Max_of_Prior_Applications_CNT_PAYMENTs_104w	0.636254
20	NEW_APPLICATION_FLAG_DOCUMENT_3	0.653004
21	NEW_APPLICATION_ORGANIZATION_TYPE	0.669505
22	CLIENT_Time_Since_Latest_Prior_Application_DECISION_DATE_104w	0.685927
23	CLIENT_Max_of_Time_between_2_Prior_Applications_for_the_Client_104w	0.701935
24	NEW_APPLICATION_AMT_REQ_CREDIT_BUREAU_QRT	0.717486
25	CLIENT_Min_of_Consumer_installments_records_Consumer_Loan_AMT_CREDIT_To_AMT_APPLICATIONs_24cMo	0.732970
26	CLIENT_Avg_of_Consumer_installments_records_AMT_PAYMENTs_24cMo	0.748362
27	CLIENT_Avg_of_Late_Status_Consumer_installments_records_AMT_PAYMENTs_24cMo	0.763642
28	CLIENT_Count_of_Cash_loans_Contract_type_Prior_Applications_by_Prior_Application_YIELD_GROUP_104w	0.777910
29	CLIENT_vs_OVERALL_Count_of_Approved_Status_Cash_loans_Contract_type_Prior_Applications_by_Prior_...	0.791514
30	NEW_APPLICATION_INCOME_TYPE	0.805046
31	CLIENT_Approved_Status_Prior_Applications_AMT_CREDITs_by_Prior_Application_YIELD_GROUP_104w	0.818565
32	CLIENT_Avg_of_Approved_Status_Prior_Applications_CNT_PAYMENTs_104w	0.831819
33	NEW_APPLICATION_OCCUPATION_TYPE	0.844805
34	CLIENT_Max_of_Consumer_installments_records_Consumer_Loan_AMT_APPLICATIONs_24cMo	0.856838
35	CLIENT_Sum_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_To_Sum_of_Approved_Status_Prior_Ap...	0.868432
36	NEW_APPLICATION_AMT_CREDIT	0.879934
37	CLIENT_Pct_of_Prior_Applications_is_Refused_Status_52w	0.891300
38	CLIENT_Pct_of_Prior_Applications_is_Refused_Status_Cash_loans_Contract_type_104w	0.901749
39	CLIENT_Min_of_Consumer_installments_records_AMT_PAYMENTs_12cMo	0.912137
40	CLIENT_Max_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_104w	0.921494
41	CLIENT_Max_of_Time_between_2_Approved_Status_Prior_Applications_for_the_Client_104w	0.930813
42	CLIENT_Avg_of_Approved_Status_Prior_Applications_AMT_ANNUITYs_104w	0.940113
43	CLIENT_Latest_Refused_Status_Cash_loans_Contract_type_Prior_Application_YIELD_GROUP_104w	0.949050
44	CLIENT_Min_of_Approved_Status_Cash_loans_Contract_type_Prior_Applications_AMT_ANNUITY_To_AMT_CRE...	0.957822
45	CLIENT_Temporal_Mean_of_Sums_of_Consumer_installments_records_Consumer_Loan_AMT_ANNUITYs_24cMo	0.966514
46	NEW_APPLICATION_DAYS_REGISTRATION	0.975172
47	CLIENT_Max_of_Prior_Applications_Application-Credit_Gaps_104w	0.983529
48	CLIENT_Avg_of_Prior_Applications_CNT_PAYMENTs_104w	0.991302
49	CLIENT_Min_of_Cash_loans_Contract_type_Prior_Applications_CNT_PAYMENTs_104w	0.998942
50	CLIENT_Consumer_Loan_GOODS_CATEGORY_with_Lowest_sum_of_active_Consumer_Loans_AMT_CREDITs	1.000000

Iterate¶

Check out Credit Default UI Tutorials to get more ideas.

Explore features with the two source tables we left for you to explore.

Table	Description
CASH_LOAN_STATUS	Tracks cash loans status.
CASH_INSTALLMENTS	Logs monthly installments for cash loans at the time of payment.

In [ ]: