Rolling test windows in Multivariate Time Series post

🤗Transformers
1

I was reading the blog post on multivariate time-series, and was confused by the snippet of code pasted below:

from gluonts.dataset.multivariate_grouper import MultivariateGrouper

num_of_variates = len(train_dataset)

train_grouper = MultivariateGrouper(max_target_dim=num_of_variates)
test_grouper = MultivariateGrouper(
    max_target_dim=num_of_variates,
    num_test_dates=len(test_dataset) // num_of_variates, # number of rolling test windows
)

multi_variate_train_dataset = train_grouper(train_dataset)
multi_variate_test_dataset = test_grouper(test_dataset)

Why is the num_test_dates = len(test_dataset) // num_of_variates? The length of the test_dataset and num_variates are both the same, they correspond to the number of entries in the dataset, not the number of entries in the target series. In this case num_test_dates=1 which I think is correct because the validation and test sets are one prediction_length longer than the train and validation sets, respectively, but I don’t see the connection to the number of variates.

Thanks for any help!