This is an implementation of the paper Generative Moment Matching Networks, ICML 2015. The paper can be found here: https://arxiv.org/abs/1502.02761. This implementation is in Python using Tensorflow.
The implementation depends on the following Python libraries:
argparse, cPickle, math, matplotlib, numpy, random, tensorflow
- Extract the data from
data.tar.gzinto the same folder as that of the implementationgenerativeMomentMatchingNetworks.py. The data contains two filesmnist.pklandlfw.npy, for the MNIST and LFW datasets respectively. The implementation uses LFW as the TFD (which is used in the paper) is not publicly available. - The implementation
generativeMomentMatchingNetworks.pyneeds two command line arguments to work, the dataset (mnist, lfw) and the network to be used (data_space, code_space; more in the paper). These can be specified by the-d (or --dataset)and-n (or --network)respectively.
Example Usage: python generativeMomentMatchingNetworks.py -d mnist -n code_space



