I’m trying to apply a similar method from the movielens lecture to
I have separated the different features but im having a bit of a trouble deciding how to embed. Do I need to create an embedding for every single feature that I plan to use and then feed those as the input, or there’s a different approach when dealing with more than 2 relationships (ie instead of movie-user, we now have bathrooms, bedrooms, price, size, etc, etc all affecting the target category (desired, undersired, highly desired).
In general curious what’s the recommended approach of feeding multiple features like that to a model and is batch-norm enough or typically they need to be manually rescaled before feeding to the network.