create_model_lstm_cnn: Creates LSTM/CNN network

In hiddengenome/deepG: deepG

Description

Creates a netwotk consisting of an arbitrary number of LSTM layers (>0) and an optional CNN layer at the beginning. Last layer is a dense layer with softmax activation.

Usage

create_model_lstm_cnn(
  maxlen = 50,
  dropout = 0,
  recurrent_dropout = 0,
  layer.size = 128,
  layers.lstm = 2,
  solver = "adam",
  use.codon.cnn = FALSE,
  learning.rate = 0.001,
  use.cudnn = TRUE,
  use.multiple.gpus = FALSE,
  merge.on.cpu = TRUE,
  gpu.num = 2,
  num_targets = 4,
  vocabulary.size = 4,
  bidirectional = FALSE,
  compile = TRUE
)

Arguments

maxlen	Length of predictor sequence.
dropout	Fraction of the units to drop for inputs.
recurrent_dropout	Fraction of the units to drop for recurrent state.
layer.size	Number of cells per network layer.
layers.lstm	Number of LSTM layers.
solver	Optimization method, options are "adam", "adagrad", "rmsprop" or "sgd".
use.codon.cnn	First layer is a CNN layer with size of 3 to mimic codons (experimental).
learning.rate	Learning rate for optimizer.
use.cudnn	If true, using layer_cudnn_lstm() instead of layer_lstm() which is if GPU supports cudnn.
use.multiple.gpus	If true, multi_gpu_model() will be used based on gpu_num.
merge.on.cpu	True on default, false recommend if the server supports NVlink, only relevant if use.multiple.gpu is true.
gpu.num	Number of GPUs to be used, only relevant if multiple_gpu is true.
num_targets	Number of possible predictions. Determines number of neurons in dense layer.
vocabulary.size	Number of unique character in vocabulary.
bidirectional	Use bidirectional wrapper for lstm layers.

hiddengenome/deepG documentation built on April 16, 2020, 1:38 a.m.