AEDRNNNetwork

class AEDRNNNetwork(latent_space_dim=128, temporal_latent_space=False, n_layers_encoder=3, n_layers_decoder=1, dilation_rate_encoder=None, dilation_rate_decoder=None, activation_encoder='relu', activation_decoder=None, n_units_encoder=None, n_units_decoder=None)

Auto-Encoder based Dilated Recurrent Neural Networks (DRNN).

Parameters:

latent_space_dimint, default = 128

Dimensionality of the latent space.

temporal_latent_spacebool, default = False

Flag to choose whether the latent space is an MTS or Euclidean space.

n_layers_encoderint, default = 3

Number of GRU layers in the encoder.

n_layers_decoderint, default = 1

Number of GRU layers in the decoder.

dilation_rate_encoderUnion[int, List[int]], default = None

List of dilation rates for each layer of the encoder. If None, default = powers of 2 up to n_stacked.

dilation_rate_decoderUnion[int, List[int]], default = None

List of dilation rates for each layer of the decoder. If None, default to a list of ones.

activation_encoderUnion[str, List[str]], default=”relu”

Activation function to use in the GRU layers.

activation_decoderUnion[str, List[str]], default=None

Activation function of the single GRU layer in the decoder. If None, defaults to relu.

n_units_encoderList[int], default=”None”

Number of units in each GRU layer of the encoder, by default None. If None, default to [100, 50, 50].

n_units_decoderList[int], default=”None”

Number of units in each GRU layer of the decoder, by default None. If None, default to two times sum of units of the encoder.

Methods

build_network(input_shape, **kwargs)

Build the encoder and decoder networks.

build_network(input_shape, **kwargs)

Build the encoder and decoder networks.

Parameters:

input_shapetuple of shape = (n_timepoints (m), n_channels (d))

The shape of the data fed into the input layer.

**kwargsdict

Additional keyword arguments for building the network.

Returns:

encodertf.keras.Model

The encoder model.

decodertf.keras.Model

The decoder model.