Hi, let say I’ve trained a model using my data. Now I want that as soon I have new data the model keeps up to date.
Training a CF model is pretty fast, so I think a possible solution is throw away the whole thing and retrain with the data (old data + new data) from scratch.
But I wonder if there is a way to keep all embeddings already trained? I mean, I would to like to load a dataloader with only new data to the model, but I have to be sure that users and items are encoded the same way. Then I could have new users and new items, so I have to “expand” Embeddings of the model.