Why not using existing item and item_gender sparse feature to create the sequence features in run DIN example?

[pancodia]

https://github.com/shenweichen/DeepCTR-Torch/blob/master/examples/run_din.py#L17-L18

In the run DIN example, why don't we use the existing item and item_gender sparse features when creating the sequence features? If we follow the example, the interest learned from sequences is not based on the embedding learned for items and item_genders, right?