Added run configurations used in the NeurIPS paper
Updated readme with more details, including examples of Tensorboard plots.
Fixed a few bugs that came from merging the pull request into the base repository. Small note: please do not replace the line "if len(indices)==0" with "if not indices" in trainer_classification_gcn.py, because it does not have the same behavior.
Made some changes to the Wide Resnet model implementation to make it compatible again with Tensorflow 1. To achieve this, the "random translations" part of the data augmentation was temporarily removed, since Tensorflow 1 did not support the existing implementation without using some features from tf.contib.
This pull request contains the following updates: