Hi,
I am dealing with a very large train and test datasets that I cannot fit to RAM.
I was thinking about first training the model, saving it, purging train_df from the memory and then using the trained model to pass test_df by pieces, concatenating the results.
However, I cannot figure out the process to do so. The learn object contains old data, so even when I pass a new test_df piece, I am still getting the answer generated for the very first (‘old’) piece.