New blog post: "install keras environment from scratch"

[FabienLavocat]

New blog post on Installing Keras on a new OS, from installing Ubuntu to PyCharm

[fchollet]

Looks great, but I think we need basically the following two sections (each with 2 sub-sections):

• setting up Keras on a workstation with a GPU
  • install Cuda
  • install cuDNN (v3)
  • Theano version:
  • GPU-specific Theano installation instructions
  • make sure Keras is running on Theano with cuDNN/on GPU
  • TensorFlow version:
  • install the GPU-specific TF version (which is harder to install)
  • make sure Keras is running with cuDNN/on GPU
• setting up Keras to run only on CPU:
  • Theano version (this one bullet point is really what the tutorial is currently about)
  • TF version

[FabienLavocat]

I 100% agree with you @fchollet about the two new sections. As I said in a previous comment. I think we should keep this article as the very minimal to start playing with Keras. A second article to let people install TensorFlow and use the GPU will be required, but in a second article (IMO)

[fchollet]

I think we should keep this article as the very minimal to start playing with Keras. A second article to let people install TensorFlow and use the GPU will be required, but in a second article (IMO)

Fair point. But I think it would be easier to make the different sections all part of a single "manual" (which would mean releasing the two articles at the same time, and joining them either through highlighted links ("looking for the GP version? Here") or by literally putting them on the same page.

It would be a bad strategy to release info about CPU installs (which people are generally able to do without issue in 2 lines of bash) and nothing about GPU. People don't like that.

PS: what do you think about using Anaconda to simplify the install? With Anaconda, which a lot of people use, the install becomes just two lines of bash. Pros/cons?