Objective is to map images of MNIST dataset (each image comprises of 784 pixels) to a lower dimension by reducting the Reconstruction error. Also, we want to demonstrate the reconstruction of each image. All the codes written are based on this great paper.
This Repo contains three files:
- boltzman.py : Contains utility function for one RBM.
- encoder.py : Contains utility functions for stacking RBMs.
- mnist.py : Contains the driver code to minimize reconstruction error and hence encoding the images.
Just keep every file in same path and run mnist.py.
(Dependencies : Tensor-flow, numpy).