Split_by_folder with datablock API

Here is my folder structure.

train folder contains all files.Estu.csv contains all file names in train folder. valid folder contains a subset of files from train folder.

Is this how it is supposed to be?

I get an error as below.


For split_by_folder, your “train” subfolder should only contain your training set, and your “valid” subfolder should only contain your validation set - they should not leak into / duplicate each other.

Please have a look at the docs.

Similar to the docs linked above, you should be able to create the ItemList using from_folder on your “banana” folder. Then use split_by_folder on “train” and “valid” (remember, they should not contain overlapping data). In this scenario, you wouldn’t need to use your Estu.csv that contains all filenames.