[QST] How to change categorical embeddings size?

[antklen]

Hi! I can't find out how to control size of embeddings for categorical features. I create embeddings with TabularSequenceFeatures.from_schema and get embeddings with size 64. How to change this value?

[rnyak]

@antklen you can change the args here:

https://github.com/NVIDIA-Merlin/Transformers4Rec/blob/main/transformers4rec/torch/features/embedding.py#L105

the reason you get 64 is because the default embedding_dim_default is set to 64. You can change that value, In addition, you can set the embedding_dims: Optional[Dict[str, int]] as a dictionary (keys are feature names values are emb dims).

let us know if you have further questions.

[antklen]

@rnyak thank you, this is complete answer to my question)