TensorFlow implementation of Hinton et al's capsule network concept.
Uses TensorFlow's built-in MNIST sample dataset. To run, must have Python 2 and TensorFlow []. This is about as bare-bones of an implementation as I could come up with, so there are minimal checkpoints and evaluation happens at the end of training (unfortunately means have to train each time). Top of the to-do list is to come up with a version that doesn't require this.
Run with "python rewrite_capsule_network.py"