I’m attempting to train a language model like in IMDB example, using all the data I have (both the 'training set; (labelled) and ‘test set’ (unlabelled)). The provided training and test sets are in separate CSV files (I’m doing the ‘real or not’ competition from Kaggle). How can I use all the text from both CSVs to train the model?
I tried concatening the two CSVs into one dataframe, but then when I try to make the databunch using TextLMDatabunch, it looks like I have to provide it separate files for train_dl and valid_dl… There’s no option for split_by_random_pct…
so what’s the correct way to do this?