OmarMAmin
OmarMAmin
If this is true, this will explain a weird behavior that I used to see while decreasing the batch size, i get a better model in general (the reason might...
I understand that, my understanding is that we end up with a square matrix with logits, the example you used for the accidental hits has 4 candidates and 3 queries,...
This is what i got when executing your code
I guess it's currently working correctly right? it makes the softmax for this value almost 0
I believe this need to be fixed as you mentioned, I tried a different batching (supposedly should work better) but was performing worse and one explanation would be that the...
Can you please check this @maciejkula ? I tried to create batches from similar products (to make it harder for the model), and these batches will have many duplicate classes,...
I checked the test case for that layer, and It didn't make sense to me I tried to run the test script here ```python shape = (2, 4) rng =...
Hi @chiragdaryani Are you providing the examples one by one? wondering if ranking can be improving things if the catalogue is only 20 items
I guess hashing have two main benefits, one that new items gets mapped to different hashing buckets (so not all new items will be treated the same), and having a...
For the user_ids, You can represent the user id by the item_ids the user is consuming, to avoid retraining the model for each new user_id, if the cataloge is more...