Problem in the predict function of Tabular Model

[binga]

I am trying the Tabular learner and during predict phase, even if I input the same training data, the following learn.predict function throws up an error.

test = df.sample(n=100).reset_index(drop=True)
test_dl = learn.dls.test_dl(test.drop(['salary'], axis=1))
learn.predict(test_dl)

The error trace gist is here: https://gist.github.com/binga/2fabdc620a3a97b9d218b366df41b94c.

Am I missing any params or configuration settings?

P.S: This is on a Win10 machine.

[sgugger]

Predict is not intended to be used on a DataLoader but one row of a dataframe. Did you mean to use learn.get_preds(dl=test_dl)?

[binga]

Ouch. My bad. Yes, I'd like to get the predictions of all the test rows and now when I switch to get_preds, I encounter the following error on this gist - https://gist.github.com/binga/2fabdc620a3a97b9d218b366df41b94c

[sgugger]

Your are not building your DataLoaders with the fastai methods, this is why it fails. At the very leasy, use TabularDataLoaders instead of DataLoaders, which should help, but I'm not 100% confident it will work since you are not using TabularPandas.dataloaders.

[binga]

Oh yes. When I wrap my dataloaders with TabularDataLoaders, it works!!

[binga]

I was trying to have a different batch size in train and test and hence I tried the following:

trn_dl = TabDataLoader(to.train, bs=256, shuffle=True, drop_last=True)
val_dl = TabDataLoader(to.valid, bs=512)

dls = DataLoaders(trn_dl, val_dl)

Changing this to

trn_dl = TabDataLoader(to.train, bs=256, shuffle=True, drop_last=True)
val_dl = TabDataLoader(to.valid, bs=512)

dls = TabularDataLoaders(trn_dl, val_dl)

fixes the problem.

This works perfectly fine in CPU mode. However, when I add device='cuda' in TabularPandas, the model fails when I call learn.summary() or learn.lr_find() when I use your suggestion.

I can paste the gist in a few min.

Edit: Updated gist is here - https://gist.github.com/binga/2fabdc620a3a97b9d218b366df41b94c

It works if I do learn.model.to('cuda') Looks like the user has to explicitly put the model on the GPU. Am I missing anything here?

[sgugger]

Please use the forum for debugging your code, we keep the issues for bugs only and the wide community is there, not here.