autoencoder_keras: autoencoder_keras
In jdestefani/ExtendedDFML: Extended Dynamic Factor Machine Learner

Wrapper function for all the different keras autoencoder implementations

autoencoder_keras(
  X,
  method = AUTOENCODER_METHODS,
  latent_dim = 3,
  time_window = 5,
  epochs = 200,
  batch_size = 32,
  optimizer_params = list(loss = "mean_squared_error", optimizer = "adam"),
  embedding_params = list(padNA = F, shift = time_window)
)

`X`	nxN matrix containing the N time series as columns, each one of length n time steps - Numeric
`method`	Autoencoder type (from AUTOENCODER_METHODS list) - String
`latent_dim`	Number of latent dimensions - Numeric
`time_window`	Size of time window (in time steps) - Numeric
`epochs`	Number of epochs required for training - Numeric
`batch_size`	Batch size required for training - Numeric
`optimizer_params`	Optimizer parameters for keras fit function `loss`: Loss function used for optimization (among those defined by Keras) - String `optimizer`: Optimizer function used for optimization (among those defined by Keras) - String
`embedding_params`	Paramaters to control the embedding process `padNA`: Value used to indicate whether incomplete tensors should be dropped or filled with NAs - Boolean `shift`: Number of time step that be skipped in order to determine the start of the following tensor - Numeric

List containing:

autoencoder: Fitted autoencoder model (full model) - Keras object
encoder: Encoder part of the autoencoder model (Original space -> Latent Dimensions) - Keras object
decoder: Decoder part of the autoencoder model (Latent Dimensions -> Original space) - Keras object
time_window: Time window used for the embedding (to fit the autoencoder) - Numeric scalar
train_history: Object containing statistics on the training history - Keras object