Implementation:Scikit learn Scikit learn KBinsDiscretizer

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Data Preprocessing, Feature Engineering
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for discretizing continuous features into bins provided by scikit-learn.

Description

KBinsDiscretizer bins continuous data into intervals. It supports three encoding strategies: one-hot, one-hot-dense, and ordinal. The binning strategy can be uniform (equal-width bins), quantile (equal-frequency bins), or kmeans (bins based on 1D k-means clustering).

Usage

Use KBinsDiscretizer when you need to convert continuous features into discrete or categorical features, which can be useful for models that work better with categorical inputs or when you want to introduce non-linearity into linear models.

Code Reference

Source Location

• Repository: scikit-learn
• File: sklearn/preprocessing/_discretization.py

Signature

class KBinsDiscretizer(TransformerMixin, BaseEstimator):
    def __init__(
        self,
        n_bins=5,
        *,
        encode="onehot",
        strategy="quantile",
        quantile_method="warn",
        dtype=None,
        subsample=200_000,
        random_state=None,
    ):

Import

from sklearn.preprocessing import KBinsDiscretizer

I/O Contract

Inputs

Name	Type	Required	Description
n_bins	int or array-like of shape (n_features,)	No	The number of bins to produce. Default is 5. Raises ValueError if n_bins < 2.
encode	str	No	Method used to encode the transformed result: 'onehot', 'onehot-dense', or 'ordinal'. Default is 'onehot'.
strategy	str	No	Strategy used to define the widths of the bins: 'uniform', 'quantile', or 'kmeans'. Default is 'quantile'.
quantile_method	str	No	Method passed to np.percentile when strategy='quantile'. Default is 'linear'.
dtype	np.float32 or np.float64	No	The desired data-type for the output. Default is None (consistent with input).
subsample	int or None	No	Maximum number of samples used to fit the model. Default is 200,000.
random_state	int or RandomState	No	Random state for reproducibility when subsampling.

Outputs

Name	Type	Description
X_transformed	ndarray or sparse matrix	The discretized feature matrix, encoded according to the encode parameter.
bin_edges_	ndarray of ndarray	The edges of each bin for each feature after fitting.
n_bins_	ndarray of shape (n_features,)	Number of bins per feature, which may differ from n_bins if a feature has fewer unique values.

Usage Examples

Basic Usage

from sklearn.preprocessing import KBinsDiscretizer
import numpy as np

X = np.array([[-2, 1, -4, -1],
              [-1, 2, -3, -0.5],
              [0, 3, -2, 0.5],
              [1, 4, -1, 2]])

est = KBinsDiscretizer(n_bins=3, encode='ordinal', strategy='uniform')
est.fit(X)
X_transformed = est.transform(X)
print(X_transformed)

Related Pages

• Principle:Scikit_learn_Scikit_learn_Feature_Transformation