I want to try ULMFit to fine-tune an lm for my dataset.
What’s the recommended way to feed data to a DataBunch for training. I see a number of factory functions to read from a df or csv or a folder, but what if my data is coming from a database with a potentially large query-set?
Should I make my queries and return generators and then implement getItem in DataSet? Every time DataLoader wants more data, it presumably calls getItem?