Dear organizers,
I just submitted a pytorch version of my submission to CodaLab.
This submission unfortunately failed on CodaLab.
I suspect, this is due to the fact that torch is not installed by default in your codebase (as it builds upon tensorflow).
Do you have a suggestion to work around that issue?
Thanks a lot in advance for your efforts, best,
jgucci
Hi jgucci!
There are some teams that are using PyTorch so I don't think it is a problem of installation. If you want to know more about how to convert the TensorFlow data generator in PyTorch, this link might be useful :
https://github.com/google-research/meta-dataset/blob/master/Intro_to_Metadataset.ipynb
It redirects to the Meta-Dataset introduction notebook and its fourth part discuss a potential Pytorch converter. Let me know if it helps!
Also, if you provide the submission's error logs (available directly on CodaLab), I could make sure it is indeed a data generator problem.
Kind regards,
Adrian
On behalf of the organizing team.