Saturday, October 27, 2018

Tensorflow on Hold After Weeks of Wasted Effort


Time that could have been spent learning machine learning has instead been spent trying to get Tensorflow to work on Centos 7.5.  Its been a neverending nightmare, a labyrinth of conflicting requirements, installation programs, neurotic programs that require specific versions, complications of using a GPU and so forth.  I feel like I am living the Reagan "Star Wars" nightmare of software that is one bridge too far.

Let there be no doubt I made this worse for myself in a variety of ways.  This computer has to serve a lot of purposes, not just one.  Google Tensorflow makes a precompiled version for Ubuntu, which I despise, and makes the source available for other Linux's including Centos (Centos == Redhat for those of you outside that world).

Without doubt, the biggest problem here involves Python and its wacky Pip installer.  Look, I know that if you are immersed in the Python culture that this all works for you kids.  But I am not immerse, I want to use Keras and Tensorflow, Python is just along for the ride.

I had it working for normal CPU work and it was painful but not impossible.  When I went to use it with a GPU, though, it all fell apart and now I can not get back.

The good news is that I do have CUDA working, which is a non trivial accomplishment.  Of course, X is broken.

This nightmare has to end.

No comments:

Post a Comment