I am recreating dataloaders on a dataframe that includes new data not seen by the model.
These work fine when I have the dep_var for the new data, however in reality I will never have the dep_var when I am trying to make predictions.
I cannot fill these values with nan as:
dl = learn.dls.test_dl(predict_df, bs=64) will not accept nan values in y:
“nan values in y but not in setup training set”
I’ve tried substituting with 0 but this produces bogus results since there are no zero values in my dep_vars.
I know preds, _ = learn.get_preds(dl=dl) does not require the dep_var in order to make predictions, but i cannot recreate the dataloaders correctly to get to that stage.
dl = learn.dls.test_dl(test_df) is still asking for the ‘y’ column!! (See below post)
Any suggestions would be greatly appreciated.
Thanks
Edit: fixed… apparently you can’t include the dep var in the cat/cont variables when creating the dls