rnn.graph.unroll

Description

Unroll representation of RNN running on non CUDA device

Usage

rnn.graph.unroll(

  num_rnn_layer,

  seq_len,

  input_size = NULL,

  num_embed = NULL,

  num_hidden,

  num_decode,

  dropout = 0,

  ignore_label = -1,

  loss_output = NULL,

  init.state = NULL,

  config,

  cell_type = "lstm",

  masking = F,

  output_last_state = F,

  prefix = "",

  data_name = "data",

  label_name = "label"

)

Arguments

Argument

Description

num_rnn_layer

int, number of stacked layers

seq_len

int, number of time steps to unroll

input_size

int, number of levels in the data - only used for embedding

num_embed

int, default = NULL - no embedding. Dimension of the embedding vectors

num_hidden

int, size of the state in each RNN layer

num_decode

int, number of output variables in the decoding layer

dropout

config

Either seq-to-one or one-to-one

cell_type

Type of RNN cell: either gru or lstm