keras-balanced-batch-generator: A Keras-compatible generator for creating balanced batches

Installation

pip install keras-balanced-batch-generator

Overview

This module implements an over-sampling algorithm to address the issue of class imbalance. It generates balanced batches, i.e., batches in which the number of samples from each class is on average the same. Generated batches are also shuffled.

The generator can be easily used with Keras models' fit method.

Currently, only NumPy arrays for single-input, single-output models are supported.

API

make_generator(x, y, batch_size,
               categorical=True,
               seed=None)

• x (numpy.ndarray) Input data. Must have the same length as y.
• y (numpy.ndarray) Target data. Must be a binary class matrix (i.e., shape (num_samples, num_classes)). You can use keras.utils.to_categorical to convert a class vector to a binary class matrix.
• batch_size (int) Batch size.
• categorical (bool) If true, generates binary class matrices (i.e., shape (num_samples, num_classes)) for batch targets. Otherwise, generates class vectors (i.e., shape (num_samples,)).
• seed Random seed (see the docs).
• Returns a Keras-compatible generator yielding batches as (x, y) tuples.

Usage

import keras
from keras_balanced_batch_generator import make_generator

x = ...
y = ...
batch_size = ...
steps_per_epoch = ...
model = keras.models.Sequential(...)

generator = make_generator(x, y, batch_size)
model.fit(generator, steps_per_epoch=steps_per_epoch)

Example: Multiclass Classification

import numpy as np
import keras
from keras_balanced_batch_generator import make_generator

num_samples = 100
num_classes = 3
input_shape = (2,)
batch_size = 16

x = np.random.rand(num_samples, *input_shape)
y = np.random.randint(low=0, high=num_classes, size=num_samples)
y = keras.utils.to_categorical(y)

generator = make_generator(x, y, batch_size)

model = keras.models.Sequential()
model.add(keras.layers.Dense(32, input_shape=input_shape, activation='relu'))
model.add(keras.layers.Dense(num_classes, activation='softmax'))
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(generator, steps_per_epoch=10, epochs=5)

Example: Binary Classification

import numpy as np
import keras
from keras_balanced_batch_generator import make_generator

num_samples = 100
num_classes = 2
input_shape = (2,)
batch_size = 16

x = np.random.rand(num_samples, *input_shape)
y = np.random.randint(low=0, high=num_classes, size=num_samples)
y = keras.utils.to_categorical(y)

generator = make_generator(x, y, batch_size, categorical=False)

model = keras.models.Sequential()
model.add(keras.layers.Dense(32, input_shape=input_shape, activation='relu'))
model.add(keras.layers.Dense(1, activation='sigmoid'))
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['binary_accuracy'])
model.fit(generator, steps_per_epoch=10, epochs=5)