Bernoulli-Restricted-Boltzmann-Machines

A self made implementation of the Bernoulli Restricted Boltzmann Machine using Contrastive Divergence & Gibbs Sampling in Python.

Requirements :

1. Pytorch.

How to use?

1. Put the RBM.py file in your project's folder.
2. Import RBM like you'd import a normal Python Library. (from RBM import *)
3. An example of usage is provided here (Movie Recommendation using Restricted Boltzmann machines).

Class Description :

1. rbm = RBM(nv, nh)

nv = Number of visible nodes. (Determined by number of features in your i/p.)
nh = Number of hidden nodes. (Number of functions to learn in the hidden node.)

2. rbm.sample_h(x)

x = Values in the visible node.
returns : P(h|x), Bernoulli(P(h|x))
  a.P(h|x) = Probability of hidden node's activation given the input values (x) in the visible nodes.
  b.Bernoulli(P(h|x)) = Activations of the hidden neurons based on the Bernoulli Distribution.

3. rbm.sample_v(x)

x = Values in the hidden node.

returns : P(v|x), Bernoulli(P(v|x))
    a.P(v|x) = Probability of visible node's activation given the values (x) in the hidden nodes.
    b.Bernoulli(P(v|x)) = Activations of the visible neurons based on 
      the Bernoulli Distribution and the acquired probability P(v|x).

4. rbm.train(v0, vk, ph0, pk)

   v0 = The correct values in the initial input.
   vk = acquired values in visible nodes after k iterations of the Gibbs Sampling (G.S.).
   ph0 = Probability of hidden node's activation given the input values (v0) in the visible nodes.
   phk = Probability of hidden node's activation given the input values 
         after k iterations of G.S. (vk) in the visible nodes.

   returns : null.
   purpose : Used to train the weights (W), biases from visible nodes to hidden nodes (a) 
             & biases from the hidden nodes to the visible nodes (b) 
             of the RBM according to the Contrastive Divergence formulae given below.