ImageDataBunch.single_from_classes says training and valid sets are empty

pshkrh · January 16, 2019, 9:33am

I was following along with the v3 course videos, and this part about inference has been troubling me for a while.

This is the line:

data2 = ImageDataBunch.single_from_classes(path, classes, tfms=get_transforms(), size=224).normalize(imagenet_stats)

The path is the same path which I used for the first DataBunch which was from_folder. It works without issues there. The classes, transforms, size and normalization is the same as the original data.

Not sure what I’m doing wrong here.

This is the error:

/home/ubuntu/anaconda3/lib/python3.6/site-packages/fastai/data_block.py:388: UserWarning: Your training set is empty. Is this is by design, pass ignore_empty=True to remove this warning.
warn(“Your training set is empty. Is this is by design, pass ignore_empty=True to remove this warning.”)
/home/ubuntu/anaconda3/lib/python3.6/site-packages/fastai/data_block.py:391: UserWarning: Your validation set is empty. Is this is by design, use no_split()
or pass ignore_empty=True when labelling to remove this warning.
or pass ignore_empty=True when labelling to remove this warning.""")

Any help would be appreciated.

sgugger · January 16, 2019, 2:28pm

I just pushed a fix for this in master. Note that there is an easier way now, using learn.export() and load_learner.

pshkrh · January 17, 2019, 3:42pm

Thanks for the fix.

However the new method you mentioned learn.export doesn’t work on my paperspace machine. I checked by doing git pull and everything seems to be updated.

It gives me this error:

AttributeError Traceback (most recent call last)
in
----> 1 learner.export()
AttributeError: ‘Learner’ object has no attribute ‘export’

sgugger · January 17, 2019, 3:46pm

You need a developer install as it’s only in master for now, or wait a little bit more for v1.0.40 (normally later today).

pshkrh · January 17, 2019, 3:48pm

Ah I see. I’ll wait for the update.

amritv · January 29, 2019, 2:18am

Hi @sgugger, has this been updated?

v1.0.40 (normally later today)

I am still getting the same error with non dev install. Thanks

sgugger · January 29, 2019, 3:27am

Yes, the latest is 1.0.41

thornaments · February 19, 2019, 9:52am

I’m also getting this error running on Crestle today. Everything seems to be up to date. Any idea what I should do?

sgugger · February 19, 2019, 2:10pm

It isn’t up to date. You can check your version with

from fastai.utils.show_install import *
show_install()

aaaaamp · March 3, 2019, 10:05pm

ModuleNotFoundError: No module named ‘fastai.utils.show_install’