collaboration between: Ethan Caballero & Aran Komatsuzaki
This is an attempt to use ENAS (Efficient Neural Architecture Search) to find optimal GAN architectures. Unfortunately, ENAS seems to be unable to find useful GAN architectures.
First download CIFAR-10 python version from https://www.cs.toronto.edu/~kriz/cifar.html to obtain folder cifar-10-batches-py/
to run:
python3 train.py --save True --incept-start-epoch 10
to run ppo:
python3 train.py --save True --incept-start-epoch 10 --ppo True
Make sure to change the names of saved models (or back them up somewhere) before running new models that will be saved, because the new saved models will overwrite the old saved models if old saved models have default names.
pytorch v0.2 or v0.3