I’m currently working on a Recommender System using Collaborative Filtering and the MovieLens dataset. The goal is to export a trained collab_learner to later import and use it in an application so that you can later also upload your own movie-data as a CSV, retrain the model and get your recommendations:
I’m exporting the learner with learner.export(export.pkl) after training it for a longer time and now import the trained model in a seperate application with learner = load_learner(path=learn.path). I would now like to add a new user with its specific ratings to the model and retrain it for a few epochs.
learner.data shows that the data is empty, but learner.data.train_ds.x.classes tells that the user and item names are there. Is it possible to attach the new data here and retrain? Or would I still need the original data in a seperate file?
Of course it works when I add the new data to the old beforehand and train a new learner from scratch, but that takes too much time for an actual application.
The used Fast AI version is 1.0.50.post1