By

From 0 to Tensorflow

Read in the dark

Set everything up

STATUS : being written, be patient

Here I’ll quickly go over the installation you need to do to run Theano Keras and Tensorflow

Python

Start with Python and scientific libraries. If you don’t, check out Anaconda or this blog post (for Mac users but interesting for everyone).

Regarding Theano, least check the requirements if you have a doubt : http://deeplearning.net/software/theano/install.html#requirements

Theano

It may seem obvious but do install Theano before Keras and/or Lasagne.

http://deeplearning.net/software/theano/install.html#install

CUDA

If you have a Nvidia GPU Theano can use it to do much faster computations; be sure to install all Theano’s required CUDA dependencies. See also “What is GPU accelerated computing?” on Nvidia’s website.

cuDNN

Once CUDA is installed you need a Deep Learning specific library for your GPU called cuDNN. Go download it there. (you have to register and answer a few questions but no restriction)

Also, a very nice tutorial (but careful some things aren’t up to date and if you’re not comfortable with linux these Cuda/cuDNN libraries are going to be painful to install)

Lasagne

http://lasagne.readthedocs.org/en/stable/user/installation.html

Keras

Keras can use either Theano or Google’s Tensorflow as a processing backend. I have not tried Tensorflow so everything I say might or not be portable to the Tensorflow backend. My feeling is that it is.

http://keras.io/#installation