The Basics

Dummy dataset

Let's start first with a dummy dataset based on the UCI credit card dataset.

from skorecard.datasets import load_uci_credit_card

X, y = load_uci_credit_card(return_X_y=True)
X.head(4)

	EDUCATION	MARRIAGE	LIMIT_BAL	BILL_AMT1
0	1	2	400000.0	201800.0
1	2	2	80000.0	80610.0
2	1	2	500000.0	499452.0
3	1	1	140000.0	450.0

A basic bucketer

skorecard offers a set of bucketers that have a scikit-learn compatible interface. By default they will bucket all variables into n_bins buckets.

Some bucketers like OptimalBucketer and DecisionTreeBucketer are supervised and can use information from y to find good buckets. You can control the numbers of buckets using max_n_bins instead of n_bins.

from skorecard.bucketers import DecisionTreeBucketer

bucketer = DecisionTreeBucketer(max_n_bins=10)
X_transformed = bucketer.fit_transform(X, y)
X_transformed.head(4)

	EDUCATION	MARRIAGE	LIMIT_BAL	BILL_AMT1
0	0	1	9	9
1	1	1	3	7
2	0	1	9	9
3	0	0	5	0

X_transformed["BILL_AMT1"].value_counts().sort_index()

0    1343
1     404
2     574
3     462
4     400
5     359
6     857
7     789
8     500
9     312
Name: BILL_AMT1, dtype: int64

Bucketing specific variables

Instead of applying a bucketer on all features, you'll likely want to apply it only to specific features. You can use the variables parameter for that:

bucketer = DecisionTreeBucketer(max_n_bins=10, variables=["BILL_AMT1"])
bucketer.fit_transform(X, y).head(4)

	EDUCATION	MARRIAGE	LIMIT_BAL	BILL_AMT1
0	1	2	400000.0	9
1	2	2	80000.0	7
2	1	2	500000.0	9
3	1	1	140000.0	0

Inspecting bucketing results

skorecard bucketers have some methods to help you inspect the result of the bucketing process:

from skorecard.bucketers import EqualWidthBucketer

bucketer = EqualWidthBucketer(n_bins=5, variables=["BILL_AMT1"])
bucketer.fit(X, y)
bucketer.bucket_table("BILL_AMT1")

	bucket	label	Count	Count (%)	Non-event	Event	Event Rate	WoE	IV
0	-1	Missing	0.0	0.00	0.0	0.0	NaN	0.000	0.000
1	0	(-inf, -10319.399999999994]	3.0	0.05	3.0	0.0	0.000000	4.181	-0.003
2	1	(-10319.399999999994, 144941.2]	5408.0	90.13	4188.0	1220.0	0.225592	-0.008	-0.000
3	2	(144941.2, 300201.80000000005]	490.0	8.17	395.0	95.0	0.193878	0.183	-0.003
4	3	(300201.80000000005, 455462.4]	75.0	1.25	55.0	20.0	0.266667	-0.230	-0.001
5	4	(455462.4, inf]	24.0	0.40	14.0	10.0	0.416667	-0.903	-0.004

bucketer.plot_bucket(
    "BILL_AMT1", format="png", scale=2, width=1050, height=525
)  # remove format argument for an interactive plotly plot.

---

Last update: 2023-08-08