Implementation:Scikit learn contrib Imbalanced learn BalancedBatchGenerator

Knowledge Sources	imbalanced-learn imbalanced-learn Docs
Domains	Deep_Learning, Data_Preprocessing, Imbalanced_Learning
Last Updated	2026-02-09 03:00 GMT

Overview

Concrete tool for generating balanced mini-batches for Keras model training provided by the imbalanced-learn library.

Description

The BalancedBatchGenerator class creates a Keras Sequence (or PyDataset) that yields balanced mini-batches. On initialization, it applies a sampler (default: RandomUnderSampler) to create balanced indices, shuffles them, and partitions them into batches. Compatible with model.fit() for seamless Keras integration.

Usage

Import this class when training Keras models on imbalanced data. Pass it directly to model.fit() as the training data generator.

Code Reference

Source Location

• Repository: imbalanced-learn
• File: imblearn/keras/_generator.py
• Lines: L64-202

Signature

class BalancedBatchGenerator(*ParentClass):
    def __init__(
        self,
        X,
        y,
        *,
        sample_weight=None,
        sampler=None,
        batch_size=32,
        keep_sparse=False,
        random_state=None,
    ):
        """
        Args:
            X: ndarray of shape (n_samples, n_features) - Training data.
            y: ndarray of shape (n_samples,) or (n_samples, n_classes) - Targets.
            sample_weight: ndarray or None - Sample weights.
            sampler: sampler with sample_indices_ or None - Balancing sampler
                (default: RandomUnderSampler).
            batch_size: int - Samples per batch (default: 32).
            keep_sparse: bool - Preserve sparse input (default: False).
            random_state: int, RandomState, or None - Seed.
        """

Import

from imblearn.keras import BalancedBatchGenerator

I/O Contract

Inputs

Name	Type	Required	Description
X	ndarray of shape (n_samples, n_features)	Yes	Training features
y	ndarray of shape (n_samples,) or (n_samples, n_classes)	Yes	Targets (can be one-hot encoded)
sampler	sampler with sample_indices_ or None	No	Balancing sampler (default: RandomUnderSampler)
batch_size	int	No	Samples per gradient update (default: 32)

Outputs

Name	Type	Description
__getitem__(index)	tuple of (X_batch, y_batch)	Balanced mini-batch
__len__()	int	Number of batches per epoch

Usage Examples

import tensorflow
from sklearn.datasets import load_iris
from imblearn.datasets import make_imbalance
from imblearn.keras import BalancedBatchGenerator

# Prepare data
iris = load_iris()
X, y = make_imbalance(
    iris.data, iris.target,
    sampling_strategy={0: 30, 1: 50, 2: 40}
)
y_cat = tensorflow.keras.utils.to_categorical(y, 3)

# Build model
model = tensorflow.keras.models.Sequential([
    tensorflow.keras.layers.Dense(y_cat.shape[1], input_dim=X.shape[1], activation="softmax")
])
model.compile(optimizer="sgd", loss="categorical_crossentropy", metrics=["accuracy"])

# Train with balanced batches
training_generator = BalancedBatchGenerator(X, y_cat, batch_size=10, random_state=42)
model.fit(training_generator, epochs=10, verbose=0)

Related Pages

Implements Principle

• Principle:Scikit_learn_contrib_Imbalanced_learn_Balanced_Batch_Generation

Requires Environment

• Environment:Scikit_learn_contrib_Imbalanced_learn_Keras_TensorFlow