I have a dataset of 3900 users and 920 unique items with around 550k interactions between them. If, for example, I trained a model on a graph that has 3000 users and 920 items. Then, I expanded the graph to include the 3900 users.
Can I load the state dict of the model with the 3000 nodes, freeze the updates for them, and only update the new 900 users?
I am trying to simulate new incoming users to the graph and generate recommendations for them without having to retrain the model on the whole graph again.
