I am looking at the Embedding layers in the Rossmann notebook. I noticed that the sizes of the inputs to the Embeddings are one bigger than I expected. Here are the embeddings and associated names:
{('Assortment', Embedding(4, 3)),
('CloudCover_na', Embedding(3, 3)),
('CompetitionDistance_na', Embedding(3, 3)),
('CompetitionMonthsOpen', Embedding(26, 10)),
('CompetitionOpenSinceYear', Embedding(24, 9)),
('Day', Embedding(32, 11)),
('DayOfWeek', Embedding(8, 5)),
('Events', Embedding(22, 9)),
('Month', Embedding(13, 7)),
('Promo2SinceYear', Embedding(9, 5)),
('Promo2Weeks', Embedding(27, 10)),
('PromoInterval', Embedding(4, 3)),
('Promo_bw', Embedding(7, 5)),
('Promo_fw', Embedding(7, 5)),
('SchoolHoliday_bw', Embedding(9, 5)),
('SchoolHoliday_fw', Embedding(9, 5)),
('State', Embedding(13, 7)),
('StateHoliday', Embedding(3, 3)),
('StateHoliday_bw', Embedding(4, 3)),
('StateHoliday_fw', Embedding(4, 3)),
('Store', Embedding(1116, 81)),
('StoreType', Embedding(5, 4)),
('Week', Embedding(53, 15)),
('Year', Embedding(4, 3))}
You see that Week input is 53, month is 13, DayOfWeek is 8.
Am I missing something here? Is there a bias term?