[FIXED] Does it always need to install TensorFlow before Keras? and Why?

Issue

I read the article as below. https://www.pyimagesearch.com/2016/11/14/installing-keras-with-tensorflow-backend/

It seems to install TensorFlow before Keras. As far as I researched, it is because we can specify CPU or GPU parameter for installing.

In your view, does it always need to install TensorFlow before Keras? And why?

At the moment, I installed Keras directly as the figure shown below. I got TensorFlow version 1.14.

my keras & tensorflow installation

In https://www.tensorflow.org/versions I found that latest TensorFlow is v2.2.

Thank you.

Pongthorn.

Solution

UPDATE DEC 2023:

Keras 3 was now released, being a multi-framework deep learning API (JAX, TensorFlow, PyTorch). To use Keras 3, you need to install either of those: for a more detailed guide, follow this link: https://keras.io/getting_started/

This was a requirement before Keras was implemented in TensorFlow.

Even on version 1.14, Keras is already present in TensorFlow package.

The recommended approach as of now and in the foreseeable future is to use the keras inside Tensorflow, as even Francois Chollet, the creator of Keras mentions this.

Practically, you have to install only TensorFlow, and make all your imports like from tensorflow.keras.X import Y rather than from keras.X import Y.

The last major release of Keras has already been published (2.3.X), but Keras as a standalone library will not be maintained anymore. This, alongside with the fact that Keras inside TensorFlow is much better supported/maintained/updated (plenty of Software Developers at Google) is the reason why using Keras from TensorFlow is the solution.

Answered By - Timbus Calin