Keras Backend Functions and Utilities

Keras is a high-level neural network API. It is simple and easy and offers ample features to simplify the task of the user. It is user-friendly offering high readability and focuses on the idea of a Model. Keras is built for fast experiments. It is the most popular Python library supporting Deep Learning. Let us learn about Keras Backend Functions and Utilities.

Keras backend-Tensorflow & Theano

Keeping you updated with latest technology trends, Join TechVidvan on Telegram

Keras Backend Basics

Knowing that Keras is a high-level neural network API. Clearly, it does not support low-level computations. It depends on other platforms and libraries to do these low-level tasks. There are some specialized and optimized manipulation libraries that help Keras. These libraries are called the ‘Backend Engines’ of Keras. Keras have such three libraries supporting Keras in the backend. These are TensorFlow, Theano, and Microsoft CNTK. Keras offers you an opportunity to choose among these backends. Because Keras handles problems in a modular way. You can also choose multiple backends if you want. Keras runs on the top of these backend engines.

Commonly Used Keras Backends

Although Keras support three backend engines. TensorFlow and Theano have commonly used backends:

1. TensorFlow

TensorFlow is an open-source library used for Machine Learning and developed by Google. It focuses on symbolic tensor manipulations and is mainly used for mathematical computational tasks.

2. Theano

Theano is another backend Python library developed by the LISA lab at the University of Montreal. It is also an open-source Deep Learning library. It offers you an opportunity to evaluate the multi-dimensional arrays and allows you to manipulate and evaluate mathematical expressions.

How to Select/Change a Backend in Keras

To check about the backend you are using, you can use a backend function.


Although Keras opts TensorFlow as its default backend. Still, it offers you an opportunity to change and switch between backends. If you want to change the backend from TensorFlow to Theano or CNTK, you need to execute a small code of two lines. Using use_backend() and passing the name of the backend as its argument, you can accomplish your task.

Code to switch to Theano:


Code to switch to CNTK:
To use CNTK as your backend engine, follow the installation instructions.


Configuration File of Keras

If you have run Keras at least once, and if not, you can create it. You will find the configuration file:


The default configuration files look like:

    "image_data_format": "channels_last",
    "epsilon": 1e-07,
    "floatx": "float32",
    "backend": "tensorflow"

To change the backend, edit TensorFlow, theano, or cntk in the backend, according to your choice. Now, save the file and restart your terminal. On starting the Keras again, you will find your backend changed.

Keras Backend Functions

There are ample backend functions that enable you to perform effectively. Let’s study a few of them:
1. Symbolic(): it is useful in TensorFlow 2.0. It enables you to enter the Keras graph. It returns a decorated graph.

2. Eager(): It is useful in TensorFlow 2.0. it helps you to exit the Keras graph. It returns the decorated function.

3. Backend(): It returns the name of the backend you are currently using. It returns the name as a string.

4. Clear_session(): It enables you to destroy the current Keras graph and enables you to create a new graph.

5. Shape(): It returns the shape of variable or tensor in symbolic form.


Keras is the most famous Python library for Deep Learning. It is very simple and enables fast implementation. It runs on the top of TensorFlow and Theano. But it does not perform low-level computations. So, it takes the help of backend engines. These Keras backend engines are TensorFlow, Theano, and Microsoft CNTK. TensorFlow is the default backend for Keras. Keras is glooming these days.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.