down.rbm: Reconstruct original data using features

In sdverkoelen/RBM: Train and use (a stack of) Restricted Boltzmann Machines

Description

Reconstruct original data using features

Usage

down.rbm(model, data)

Arguments

model	an RBM object from the single.rbm or multi.rbm functions
data	matrix with features

Value

a matrix with reconstructions of the input features

Examples

# Data availible at https://www.samverkoelen.com/data
data <- as.matrix(read.csv('mnist_sample.csv'));

#train RBM stack
model <- multi.rbm (hidden = c(150, 75, 30), data = data)

par(mfrow=c(1,2))
id <- 57

#Original
original <- matrix(data[ ,id], byrow = T, ncol = 16)
original <- t(original[16:1, 1:16]) #correct orientation
image(original, col = grey(seq(0, 1, 0.001)), main = 'Original')

#visible to hidden
features <- up.rbm(model, data)

#hidden back to visible
reconstructions <- down.rbm(model, features)

#Reconstruction
reconstruction <- matrix(reconstructions[id, ], byrow = T, ncol = 16)
reconstruction <- t(reconstruction[16:1, 1:16]) #correct orientation
image(reconstruction, col = grey(seq(0, 1, 0.001)), main = 'Reconstruction')

sdverkoelen/RBM documentation built on May 29, 2019, 4:23 p.m.