create.mlp: Initialize a Neural Network.

In Greg-Hallenbeck/easy.mlp: Easy-To-Use Multilayer Perceptron

Initialize a Neural Network.

Description

Create a new neural network to fitting tabular data. The network is initialized randomized weights and should be trained with the 'train()' function.

Usage

create.mlp(
  formula,
  data = NULL,
  hidden = c(),
  type = NULL,
  activation = "leaky.relu",
  valid.split = 0.25,
  randomize = TRUE,
  learning.rate = 0.001,
  momentum.beta = 0.9,
  rmsprop.beta = 0.999,
  batch.size = 64
)

Arguments

formula	An object of class 'formula' describing the model to be fit.
data	A 'data.frame' object containing the training and validation data.
hidden	A vector indicating how many layers and hidden notes the network will have. Passing c(4,5,2) creates a network with 3 hidden layers with 4, 5, and 2 nodes, respectively.
type	One of "regression", "logistic", or "classification".
activation	The name of a function to be used for activations in the hidden layers. "leaky.relu", "relu", "tanh", and "sigmoid" are included in the package.
valid.split	The fraction of the data to be used for validation.
randomize	If TRUE, the data frame will be randomly sampled before train/validation split. If not, the final rows are the validation data.
learning.rate	The learning rate of the network.
momentum.beta	Beta parameter for gradient descent with momentum.
batch.size	Minibatch size. Choosing 1 is equivalent to stochastic gradient descent.
rmpsprop.beta	Beta parameter for RMSprop gradient descent.

Value

An object of class 'mlp'

Examples

 create.mlp(Species ~ ., data=iris, hidden=c(5,5,5), type="classification")

Greg-Hallenbeck/easy.mlp documentation built on March 10, 2023, 6:31 a.m.