Fastai tabular learner - continue training on new dataframe

Part 1 (2019)
1

We have trained a model on one large csv file and save the model learner.export("my_model.pth").

THen a month later we get a new csv file and want to continue training on it.

I have tried:

learner = load_learner('my_model.pth')
my_new_df = pd.read_csv('new_file.csv")

learner.dls.train = my_new_df

learner.fit_one_cycle(1)

returns an error
string indices must be integers

I thought that dls object inside the learner is handling the preprocessing - e.g. normalizing etc.

2

You replaced the dataloader with a pandas dataframe which I think it can not work. Try to create a new dataloader from your new .csv file then replace the dataloader in learner with the new one.

Take a look here how to create a dataloader for tabular data: Tabular training | fastai