Interesting article by Instacart and how they use Keras to sort shopping lists (you can ignore the emoji part since it is just for cutesy graphics in the article):
I’m only on lesson 5 so far, but will there be any discussion about their type of architecture and in particular ‘masked-soft max’ for considering items on the customer’s list in their first iteration?
It sounds like it just drops all items not on the customers list to 0 before it goes to the next stage. All items that are on the list are then given a confidence (or probability?) it should be chosen next. Is that correct?
When would this make sense from a performance reason?