RBM: Restricted Boltzmann Machine

In TimoMatzen/RBM: Package for fitting RBM and DBN models in R.

Description

Trains a Restricted Boltzmann Machine on binary data, either supervised or unsupervised. This function uses contrastive diversion with k = 1 for training the system.

Usage

RBM(x, y, n.iter = 100, n.hidden = 30, learning.rate = 0.1,
  plot = FALSE, size.minibatch = 10, momentum = 0.5,
  lambda = 0.001)

Arguments

x	A matrix with binary features of shape samples * features.
y	A matrix with labels for the data, only when training a classification RBM. (Optional)
n.iter	Defines the number of epochs to run contrastive diversion.
n.hidden	The number of nodes in the hidden layer.
learning.rate	The learning rate, alpha, for training the system.
plot	Whether to plot the learning progress of the weights
size.minibatch	The size of the minibatches used for training.
momentum	Speeds up the gradient descent learning.
lambda	The sparsity penalty lambda to prevent the system from overfitting.

Value

A list with the trained weights of the RBM that can be used for the predict RBM function when supervised learning was applied or the ReconstructRBM function to reconstruct data with the model.

Examples

# Load the MNIST data
data(MNIST)

# Fit unsupervised RBM with the MNIST data
modelRBM <- RBM(MNIST$trainX, n.iter = 1000, n.hidden = 30, learning.rate = 0.1, 
plot = FALSE, size.minibatch = 10, momentum = 0.5, lambda = 0.001)

# Fit a supervised RBM with the MNIST data
modelRBMSup <- RBM(MNIST$trainX, MNIST$trainY, n.iter = 1000, n.hidden = 30, learning.rate = 0.1, 
plot = FALSE, size.minibatch = 10, momentum = 0.5, lambda = 0.001)

# Add more weight decay when training more parameters
modelRBMSup <- RBM(MNIST$trainX, MNIST$trainY, n.iter = 1000, n.hidden = 100, learning.rate = 0.1, 
plot = FALSE, size.minibatch = 10, momentum = 0.5, lambda = 0.1)

TimoMatzen/RBM documentation built on June 1, 2019, 8:35 a.m.