AttributeError: Caught AttributeError in DataLoader worker process 0

[ssf (sun shuo feng)]

I learned to use fastai in the kaggle race and had problems making predictions at the end.

Here’s a link to the contest and a screenshot of the data format:

kaggle

Here is my code:

tfms=get_transforms()

test_images=ImageList.from_folder(data_dir/‘test’)

data=(ImageList.from_folder(data_dir/‘train’)
.split_by_rand_pct()
.label_from_func(get_label)
.add_test(test_images)
.transform(tfms,size=299)
.databunch(bs=8)
.normalize(imagenet_stats))

model=models.resnet50(pretrained=True)
learn=Learner(data,model,metrics=[error_rate,accuracy])

learn.fit_one_cycle(10)
learn.export()

predict_learn=load_learner(data_dir/‘train’)
preds,_=predict_learn.get_preds()
print(preds.shape)

or

pred,_=predict_learn.predict(test_images)

All of them make mistakes like this:

in addition, this error occurs when using predict(test_images)：

how should i solve that two error?

[ssf (sun shuo feng)]

I find that if I use the trained learn directly to get_preds, there is no error, but if I use the learn obtained by load_learner for prediction, an error will be reported

[joatom (Johannes Tomasoni)]

Could be to different fastai versions (even minor version) in both environments. Check fastai.version in the exporting and also in the loading environment.

[jss367 (Julius)]

This was the issue for me.