daltoolboxdp (version 1.2.737)

autoenc_denoise_ed: Denoising Autoencoder - Encode-Decode

Description

Creates a deep learning denoising autoencoder (DAE) that encodes and decodes sequences, learning robustness to input noise. Wraps a PyTorch implementation.

Usage

autoenc_denoise_ed(
  input_size,
  encoding_size,
  batch_size = 32,
  num_epochs = 1000,
  learning_rate = 0.001,
  noise_factor = 0.3
)

Value

A autoenc_denoise_ed object.

Arguments

input_size

input size

encoding_size

encoding size

batch_size

size for batch learning

num_epochs

number of epochs for training

learning_rate

learning rate

noise_factor

Numeric. Standard deviation (scale) of the noise added during training.

References

Vincent, P. et al. (2008). Extracting and Composing Robust Features with Denoising Autoencoders.

Examples

if (FALSE) {
# 1) Prepare data
X <- matrix(rnorm(1000), nrow = 50, ncol = 20)

# 2) Fit denoising autoencoder (encode-decode)
ae <- autoenc_denoise_ed(input_size = 20, encoding_size = 5, noise_factor = 0.2, num_epochs = 50)
ae <- daltoolbox::fit(ae, X)

# 3) Reconstruct inputs and compute error
X_hat <- daltoolbox::transform(ae, X)
mean((X - X_hat)^2)
}

# More examples:
# https://github.com/cefet-rj-dal/daltoolbox/blob/main/autoencoder/autoenc_denoise_ed.md