Training a global LSTM with multiple time series

In the previous recipe, we learned how to prepare datasets with multiple time series for supervised learning with a global forecasting model. In this recipe, we continue this topic and describe how to train a global LSTM neural network for forecasting.

Getting ready

We’ll continue with the same data module we used in the previous recipe:

N_LAGS = 7
HORIZON = 7
from gluonts.dataset.repository.datasets import get_dataset, dataset_names
dataset = get_dataset('nn5_daily_without_missing', regenerate=False)
datamodule = GlobalDataModule(data=dataset,
    n_lags=N_LAGS,
    horizon=HORIZON,
    batch_size=32,
    test_size=0.3)

Let’s see how to create an LSTM module to handle a data module with multiple time series.

How to do it…

We create a LightningModule class that contains the implementation of the LSTM. First...