Python for Deep Learning: Keras

Deep Learning is a fascinating and rapidly advancing field of artificial intelligence that has been changing the way we interact with our digital devices. It is a subset of machine learning that involves training artificial neural networks to recognize patterns in data and make predictions based on that data. With deep learning, machines can now perform complex tasks like image recognition, speech recognition and natural language processing, which were once considered the exclusive domain of humans. One of the most popular tools used for deep learning is Keras, a high-level neural networks API that is written in Python. In this article, we will explore how Python can be used with Keras to create powerful deep-learning models.


Deep learning is a subfield of machine learning that focuses on neural networks with many layers. These deep networks have shown outstanding performance in a wide range of tasks, including image recognition, natural language processing and reinforcement learning. In this article, we will delve into Python’s popular deep learning library, Keras and explore its core components, its integration with TensorFlow and how to build and train deep learning models using the Keras API.

What is Keras?

Keras is an open-source deep-learning library written in Python. It was developed by François Chollet and was first released in March 2015. Keras is a high-level neural network API designed for fast experimentation and ease of use. It is capable of running on top of TensorFlow, Microsoft Cognitive Toolkit and Theano, though, since TensorFlow 2.0, Keras has been integrated as the official high-level API for TensorFlow.

Why choose Keras for Deep Learning?

Keras has become one of the go-to deep learning libraries for several reasons:

  1. User-friendly: Keras offers a high-level, intuitive API that abstracts complex details, making it easy for beginners to get started with deep learning.
  2. Modularity: Keras models are built by connecting configurable building blocks called layers, allowing for easy and flexible model design.
  3. Extensibility: Keras is highly customizable, allowing users to create custom layers, loss functions and optimizers if needed.
  4. Compatible with TensorFlow: As mentioned earlier, Keras is now the official high-level API for TensorFlow, ensuring seamless integration and support.

Installing Keras

As Keras is now integrated with TensorFlow, you only need to install TensorFlow to get started. You can install TensorFlow using pip:

pip install tensorflow

To verify that TensorFlow and Keras are properly installed, you can run the following Python code:

import tensorflow as tf
from tensorflow import keras

print("TensorFlow version:", tf.__version__)
print("Keras version:", keras.__version__)

Keras Core Components

Keras has several core components that you’ll interact with when building deep learning models:


In Keras, a model is a container for layers that make up a neural network. The most commonly used model is the Sequential model, which is a linear stack of layers. You can also use the functional API for more complex architectures or subclass the Model class for full customization.


Layers are the building blocks of neural networks in Keras. They define the architecture and perform the necessary computations in the network. Keras provides various types of layers, such as dense (fully connected), convolutional, pooling and recurrent layers.

Loss Functions and Optimizers

A loss function measures the difference between the predicted output and the actual output (ground truth). During training, the goal is to minimize this loss. Common loss functions include mean squared error, categorical crossentropy and binary crossentropy.

Optimizers are algorithms that adjust the model’s weights based on the loss function’s value. Popular optimizers include stochastic gradient descent (SGD), Adam and RMSprop.


Metrics are used to evaluate the performance of a model. They are similar to loss functions but are not used for training. Common metrics include accuracy, precision and recall.

Building a Simple Neural Network with Keras

Let’s build a simple neural network using Keras to classify handwritten digits from the MNIST dataset. The following steps demonstrate how to create, compile and train the model:

  1. Import the necessary libraries and load the dataset:
import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

# Load the MNIST dataset
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
  1. Preprocess the data:
# Normalize the pixel values
x_train = x_train.astype("float32") / 255
x_test = x_test.astype("float32") / 255

# One-hot encode the labels
y_train = keras.utils.to_categorical(y_train, num_classes=10)
y_test = keras.utils.to_categorical(y_test, num_classes=10)
  1. Create a simple neural network model:
model = keras.Sequential([
    layers.Flatten(input_shape=(28, 28)),
    layers.Dense(128, activation='relu'),
    layers.Dense(10, activation='softmax')
  1. Compile the model:
  1. Train the model:
history =, y_train, epochs=10, batch_size=32,
                    validation_data=(x_test, y_test))
  1. Evaluate the model:
test_loss, test_accuracy = model.evaluate(x_test, y_test)
print("Test loss:", test_loss)
print("Test accuracy:", test_accuracy)

Using Keras with TensorFlow

Keras is a high-level API for TensorFlow, which means it provides an easy-to-use interface for building and training models with TensorFlow as the backend. The code above already demonstrates how Keras is used with TensorFlow. You can also use TensorFlow’s lower-level APIs alongside Keras for more control over the computation and training process.

Keras Callbacks

Callbacks are useful tools to customize the training process in Keras. They are functions that can be called at various stages during training, such as after each epoch or batch. Some commonly used callbacks include:

  • ModelCheckpoint: Save the model after each epoch.
  • EarlyStopping: Stop training when a monitored metric has stopped improving.
  • ReduceLROnPlateau: Reduce the learning rate when a metric has stopped improving.
  • TensorBoard: Log training metrics for visualization with TensorBoard.

Keras Applications

Keras Applications is a module that provides pre-trained models for popular deep learning architectures, such as VGG, ResNet and Inception. These models can be used for transfer learning, which leverages knowledge learned from one task to improve performance on another task.

from tensorflow.keras.applications import VGG16

# Load the pre-trained VGG16 model
vgg16 = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))


In this guide, we introduced Keras, a popular deep learning library for Python. We covered its core components, such as models, layers, loss functions and optimizers and demonstrated how to build and train a simple neural network using Keras. Additionally, we discussed the integration of Keras with TensorFlow, callbacks and pre-trained models available in Keras Applications. With Keras, you can quickly prototype and experiment with deep learning models, making it an invaluable tool for both beginners and experienced practitioners.