[QST] Projecting inputs of NextItemPredictionTask to'64' As weight tying requires the input dimension '320' to be equal to the item-id embedding dimension '64'

[ralgond]

Projecting inputs of NextItemPredictionTask to'64' As weight tying requires the input dimension '320' to be equal to the item-id embedding dimension '64'. Should I care the warning?

Details

I found a warning on this example page:

https://nvidia-merlin.github.io/Transformers4Rec/stable/examples/end-to-end-session-based/02-End-to-end-session-based-with-Yoochoose-PyT.html

Should I care the warning?

Thank you.

[rnyak]

@ralgond you can ignore the warning, it is telling that the it applies projection and it will be projected to the item_id embedding dimension which is 64 by default. Note that you can also change this embedding dimension if you want.

[ralgond]

So how can I change the item embedding dimension?

[rnyak]

you can add it here: inputs = tr.TabularSequenceFeatures.from_schema(...) as an argument of embedding_dims.

Please see this ticket as well.

[rnyak]

closing this due to low activity.