Generate architectures for Deep Neural Network

Generate architectures for Deep Neural Network

Usage

generate_dnn_architecture(
  number_of_features = 7,
  number_of_outputs = 1,
  number_of_hidden_layers = 2,
  hidden_layers_size = c(14, 7),
  batch_norm = TRUE,
  dropout = 0,
  verbose = FALSE
)

Arguments

number_of_features

numeric. Value that specifies the number of features in the dataset.

number_of_outputs

numeric. Value that specifies the number of outputs.

number_of_hidden_layers

numeric. Number of hidden layers in the neural network. Default 2.

hidden_layers_size

numeric vector. Size of each hidden layer in the neural network. Default c(14, 7).

batch_norm

logical. Whether to include batch normalization layers. Default TRUE.

dropout

logical. Specifies whether dropout is included in the architecture. Default FALSE.

verbose

logical. Whether to print the architecture. Default FALSE.

Value

A list containing:

• net: a instantiated torch neural net.

• arch: a string with a R expression to instantiate the neural network.

• arch_dict: a list with a matrix describing the architecture structure.

See also

generate_arch_list, generate_cnn_architecture

Examples

if (FALSE) {
# Generate a Deep Neural Network with:
dnn_arch <- generate_dnn_architecture(
  number_of_features = 8, # eight input variables
  number_of_outputs = 1, # one output
  number_of_hidden_layers = 5, # five layers between input and output
  hidden_layers_size = c(8, 16, 32, 16, 8), # of this size, respectively
  batch_norm = TRUE, # with batch normalization
  dropout = 0, # without dropout
)

dnn_arch$net() # a torch net
dnn_arch$arch %>% cat() # the torch code to create it
dnn_arch$arch_dict # and a quick description of its structure
}