c: Combine DeepBeliefNets, RestrictedBolzmannMachines and Layers...
In xrobin/DeepLearning: Deep Learning of neural networks

Description Usage Arguments Value Examples

This function combines one or more DeepBeliefNets, RestrictedBolzmannMachines and Layers into a single DeepBeliefNets.

The layers must be compatible in order, i.e. the output of the previous DBN/RBM must be the same (both in term of size and type) than the input of the next one. The only exception is with Layer objects, where RestrictedBolzmannMachines will be created before and after the layer (only one RestrictedBolzmannMachine is created Layer succeeding immediately an other Layer).

1 2	## S3 method for class 'DeepBeliefNet' c(..., biases.first = TRUE)

`...`	objects to be combined
`biases.first`	whether to use the biases of RBM i (TRUE) or i+1 (FALSE) for shared layers. Defaults to TRUE as we don't care much about the Bs after the pre-training.

the combined DBN. Note that it is not tagged as unrolled and fine-tuned any more. It is tagged as pre-trained if all individual DBNs/RBMs were pre-trained.

# DeepBeliefNets only
dbn1 <- DeepBeliefNet(Layers(c(784, 1000), input="continuous", output="binary"))
dbn2 <- DeepBeliefNet(Layers(c(1000, 500, 250), input="binary", output="binary"))
dbn <- c(dbn1, dbn2)

# RestrictedBolzmannMachines only
dbn <- c(dbn1[[1]], dbn2[[1]], dbn2[[2]])

# Layers only
dbn <- c(Layer(784, "continuous"), Layer(1000, "binary"), Layer(500, "gaussian"))

# Layers only
c(Layer(784, "continuous"), Layer(1000, "binary"), Layer(500, "binary"))

# Mixing it all
rbm3 <- RestrictedBolzmannMachine(Layer(250, "binary"), Layer(30, "binary"))
layer4 <- Layer(2, "gaussian")
c(Layer(2, "gaussian"), dbn1, dbn2, rbm3, layer4)

# The following won't work
## Not run: 
dbn3 <- DeepBeliefNet(Layers(c(250, 500), input="binary", output="binary"))
dbn <- c(dbn1, dbn3)
dbn4 <- DeepBeliefNet(Layers(c(1000, 500), input="continuous", output="binary"))
dbn <- c(dbn1, dbn4)

## End(Not run)