Each consecutive pair in dims becomes one (linear, relu) step, matching
the trunk-building loop shared by the MADE, MDN and embedding networks.
Usage
mlp_layers(dims, masks = NULL)
Arguments
- dims
Layer sizes in order, e.g. c(input, hidden1, hidden2, ...).
- masks
One weight mask per pair (as returned by
made_masks()$hidden) for an autoregressive trunk; NULL for a plain
MLP.
Value
A list of torch layers, ready for do.call(torch::nn_sequential, .).
