pruning simplenet

this repo is about training and pruning Simple-Net an explanation can be found here: https://arxiv.org/abs/1608.06037 there are 3 stages:

1. train.py this script trains the simplenet on cifar 10 data base, if you train with the defualt HP it will get to 90% ( give or take 1%) test accuracy. use --help when initiating the script to see easy to change parameters note: the learning rate is defined in configs.py has a static method which is correlated to the number of epoch passed
2. prune.py this script prunes the trained model from train.py and outputs a pruned model, there are some HP which can be explained here: https://arxiv.org/pdf/1710.01878.pdf and here https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/model_pruning/README.md use --help to learn more you can always use the defualt parameters just be careful with the sparsity_goal which defines the output model sparsity level
3. create_sparse.py this script creates a sparse model from the pruned model, basiclly it tears down all the pruning nodes from the graph and create a pb model of the sparse model.

Thank you.