Dataset

Inherits from pytorch datasets to allow use with pytorch dataloader.

class time_series_dataset.TimeSeriesDataset(time_series: list, lookback: int, horizon: int, step=1, transform=None, static_covs=None)

Take a list of time series and provides access to windowed subseries for training.

Parameters:

• time_series – List of time series arrays.
• lookback – Length of time window used as input for forecasting.
• horizon – Number of time steps to forecast.
• step – Time step size between consecutive examples.
• dropout_regularizer – Generally needs to be set to 2 / N, where N is the number of training examples.
• init_range – Initial range for dropout probabilities.
• channel_wise – Determines if dropout is appplied accross all input or across channels .

Initialize variables.