Experiments on pruning neural networks on MNIST

I use the model optimization suite of tensorflow, which requires at least tensorflow 1.14, and tkkerassurgeon to actually remove the weights. Unfortunately the latter requires tensorflow < 1.14, so I need two distinct environments to make this work. This is far from optimal, but for this experiment I did not mind. So first, you need to recreate those environments. To make things easier, I used conda and I provide to text files that you can use to create the two environments::

> conda create -n tensorflow --file tensorflow_env.txt
> conda create -n tfkerassurgeon --file tfkerassurgeon_env.txt

Go into the "scripts" directory and run "one_run.sh".