Why learn.tta(dl=test_dl) could give Error: 'NoneType' object is not callable?

Part 1 (2020)
1

I’m trying to use learn.tta(dl=test_dl) and getting this error: ‘NoneType’ object is not callable. This happens in this piece of fast.ai code:
aug_preds.append(self.get_preds(dl=dl, inner=True)[0][None])

But using just learn.get_preds(dl=test_dl) works well. Any ideas why this could happen?

2

Need more info. Post the context code and error with full strack trace.

3

You can find code here (Kaggle Notebook)

4

I was not abole to run your code because of missing path input/temp-inference-model-resnet18.

5

In general you should post a full stack trace along with your issue report. Just before the tta can you do: print(len(test_dl)) and tell us what this returns.

6

I’ve updated the link above and as you can see it traces ‘1’ for every of 3 folds. I suppose it is because of 1 test image.

7

I’ve updated a bit the code and added comments for better understanding the problem.
https://www.kaggle.com/volcanoflash/inference-test-tta

8

I was able to run this code properly on local Paperspace machine, so the problem is somehow connected to Kaggle itself or it’s environment. Getting mad with this bug…

9

It could be because of the version of fastai package. did you check the installed versions in both platforms?

run this in the notebook and compare:

!pip list | grep fast

also what do you mean by ‘local Paperspace machine’? isn’t Paperspace cloud notebook server? :thinking: