Updated 28 Jun 2017.

Keras is an amazing wrapper for Tensorflow (and Torch) that makes it simple to start playing with Neural Networks.

Using environment manager like Anaconda makes life easier. But sometimes due to different dependencies it takes additional steps to unserstand how to install needed packages.

I assume that you have Anaconda installed.

Since there is no tensorflow package in an Anaconda Package List one have to use conda-forge - community supported repository of packages.

But as of February 27, 2017 the latest Python version is 3.6 and conda-forge lacks tensorflow package for that version.

So first of all, let’s create environment with the Python, and name it a ‘tf’. I also advice to install pandas, matplotlib, jupyter and nb_conda packages for data manipulation and visualization of the result.

conda config --add channels conda-forge
conda create -n tf python=3 keras tensorflow pandas matplotlib jupyter nb_conda

Then we make new environment active:

source activate tf

Testing that Tensorflow is working

python
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))

There would be warnings that The TensorFlow library wasn't compiled to use <...> instructions, .... That is ok. We don’t want to build libraries from the the source code here.

The succesfull output should be:

Hello, TensorFlow!

Set up Keras

To work with Tensorflow as backend, please make sure that you have the following in the ~/.keras/keras.json file:

{
    "image_dim_ordering": "th",
    "floatx": "float32",
    "epsilon": 1e-07,
    "backend": "tensorflow"
}

That’s it, you are ready to use Keras with Tensorflow!
Let’s do some “Hello, World!” handwritten digits recognition.