Restricted Boltzman Machine

This is a project created to be a learning experience for myself about how Restricted Boltzmann Machines (RBMs) work. It implements CD-k with numpy along with Gibb's Block Sampling.

The weights are formulated in this machine as $v^T \cdot W \cdot h$

All vectors that are passed in should be column vectors.

Resources

This RBM currently implements:

• Normal Gaussian Initial Weights
• Mean Squared Reconstruction Error Logging
• Mini Batch Size of 100
• Learning rate of 0.01

all advised from the following paper: A Practical Guide to Training Restricted Boltzmann Machines