autoencode: Automatically compute an encoding of a data matrix

In ruta: Implementation of Unsupervised Neural Architectures

Automatically compute an encoding of a data matrix

Description

Trains an autoencoder adapted to the data and extracts its encoding for the same data matrix.

Usage

autoencode(data, dim, type = "basic", activation = "linear", epochs = 20)

Arguments

data	Numeric matrix to be encoded
dim	Number of variables to be used in the encoding
type	Type of autoencoder to use: "basic", "sparse", "contractive", "denoising", "robust" or "variational"
activation	Activation type to be used in the encoding layer. Some available activations are "tanh", "sigmoid", "relu", "elu" and "selu"
epochs	Number of times the data will traverse the autoencoder to update its weights

Value

Matrix containing the encodings

See Also

\link{autoencoder}

Examples

inputs <- as.matrix(iris[, 1:4])


if (interactive() && keras::is_keras_available()) {
  # Train a basic autoencoder and generate a 2-variable encoding
  encoded <- autoencode(inputs, 2)

  # Train a contractive autoencoder with tanh activation
  encoded <- autoencode(inputs, 2, type = "contractive", activation = "tanh")
}

ruta documentation built on Jan. 9, 2023, 1:20 a.m.