InstanceSplitter does not return empty target as suggested by Extended Forecasting Tutorial.

[hericks]

I recently went over the chapter on Transformations in the extended Extended Forecasting Tutorial. It uses a transformation chain including an InstanceSplitter.

After applying this chain to construct a test dataset, the tutorial mentions (after cell 39)

The future target is empty but not the features - we always assume that we know the future features!

However, when I inspected the array test_tf_entry['future_target'] it does not appear empty. In fact, I've looked into the implementation of the InstanceSplitter and it's flatmap_transform method does not use the is_train argument in any way. Am I misunderstanding something here or should the future target returned by the transformation chain be indeed empty?