Python Reinforcement Learning Projects

cifar10_processor.py

This module contains code for processing CIFAR-10 data, which we use to train our child networks. In particular, we construct an input data pipeline using TensorFlow's native tf.data.Dataset API. Those who have used TensorFlow for some time may be more familiar with creating tf.placeholder tensors and feeding data via sess.run(..., feed_dict={...}). However, this is no longer the preferred way of feeding data into the network; in fact, it is the slowest way to train a network, for the repetitive conversions from data in numpy format to a native TensorFlow format cause significant computational overhead. tf.data.Dataset alleviates this problem by turning the input pipeline into TensorFlow operations that are part of the ...