Using Fastai on Kaggle Kernels

I am really glad that Kaggle has released new Docker images which allows us to use fastai v1 and PyTorch v1 in Kaggle kernels.

But there is an issue since some parts of the library require files to present in a particular order(I am talking about file hierarchy like some csv file show be present in train folder when using ImageItemList.from_csv). How can I use fastai on Kaggle Kernels where data is provided through a read-only folder?

One option is to copy files outside …/input since it’s read-only. You can create directory structure that fits your scenario.

@AbuFadi I tried doing that but failed. It would be very helpful if you could provide the link of the kernels.

See this kernel