Models¶
WaveNet¶
Implements WaveNet architecture for time series forecasting.
- References:
-
class
models.WaveNet(input_channels: int, output_channels: int, horizon: int, hidden_channels=64, skip_channels=64, dense_units=128, n_layers=7, n_blocks=1, dilation=2)¶ Parameters: - input_channels – Number of covariates in input time series.
- output_channels – Number of covariates in target time series.
- horizon – Number of time steps for forecast.
- hidden_channels – Number of channels in convolutional hidden layers.
- skip_channels – Number of channels in convolutional layers for skip connections.
- dense_units – Number of hidden units in final dense layer.
- n_layers – Number of layers per Wavenet block (determines receptive field size).
- n_blocks – Number of Wavenet blocks.
- dilation – Dilation factor for temporal convolution.
-
decode(inputs)¶ Decoder part of the architecture.
-
encode(inputs)¶ Encoder part of the architecture.
-
forward(inputs)¶ Returns the parameters for a Gaussian distribution.
-
n_parameters¶ Return the number of parameters of model.
-
receptive_field_size¶ Return the length of the receptive field.