Skip to content

Keras

AI & Machine LearningML/AIPython
Official docs ↗GitHub ↗

What it is

Keras is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, Microsoft Cognitive Toolkit (CNTK), or Theano. It allows for easy and fast prototyping of deep learning models with minimal code.

Keras allows you to define deep learning models using Sequential or Functional APIs. You can build layers, compile models, train with fit(), evaluate, and make predictions easily. It supports a wide variety of layers, optimizers, loss functions, and metrics.

Installation

pip install keras

Getting started

The smallest useful thing you can do with it, and what each part means.

Simple Sequential Model
from tensorflow import keras
from tensorflow.keras import layers

model = keras.Sequential([
    layers.Dense(32, activation='relu', input_shape=(784,)),
    layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
Defines a basic feedforward neural network with one hidden layer and a softmax output layer for classification.
Training a model
import numpy as np
X_train = np.random.rand(100,784)
y_train = np.random.randint(0,10,100)
model.fit(X_train, y_train, epochs=5, batch_size=10)
Trains the Keras model on synthetic data for 5 epochs using a batch size of 10.

Advanced usage

Where the library earns its place over a simpler alternative.

Functional API Example
from tensorflow.keras import Input, Model
inputs = Input(shape=(784,))
x = layers.Dense(64, activation='relu')(inputs)
outputs = layers.Dense(10, activation='softmax')(x)
model = Model(inputs=inputs, outputs=outputs)
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
Defines a model using the Functional API, allowing for more flexible architectures than Sequential.
Using Callbacks
from tensorflow.keras.callbacks import EarlyStopping
early_stop = EarlyStopping(monitor='loss', patience=3)
model.fit(X_train, y_train, epochs=50, callbacks=[early_stop])
Demonstrates stopping training early when the loss stops improving to prevent overfitting.
Saving and loading models
model.save('my_keras_model.h5')
new_model = keras.models.load_model('my_keras_model.h5')
Shows how to save a trained Keras model to disk and load it later.

Errors and fixes

The failures you are most likely to hit, and what actually resolves them.

ValueError: Input arrays should have the same number of samples
Ensure that your features and labels arrays have the same number of samples.
InvalidArgumentError
Check that the input shape matches the model's expected input.
ModuleNotFoundError: No module named 'keras'
Install Keras using pip or conda in your current Python environment.

Best practices

  • Normalize input data for faster convergence.
  • Use validation sets and callbacks to prevent overfitting.
  • Leverage the Functional API for complex models like multi-input/multi-output networks.
  • Use pretrained models from Keras Applications for transfer learning.
  • Monitor training with TensorBoard for visual insights.

Background

Why it exists, and what it was reacting to.

Keras was developed by François Chollet and released in March 2015. Its design philosophy focuses on user-friendliness, modularity, and extensibility. Keras became popular for its simple API that abstracts the complexities of deep learning, and it was later integrated tightly with TensorFlow as its official high-level API.

Related libraries

Others in AI & Machine Learning for Python.

← All Python librariesPython guide