Error trying to create an image classifier for the RSNA Screening Mammography competition on Kaggle

I would appreciate any help anybody can offer with the issue descibed here

https://www.kaggle.com/code/julianmacnamara/rsna-screening-with-error

Many thanks in advance

Julian

1 Like

I think you need to save a version of your notebook before other people can look at it, right now if I visit the page it gives a message to that effect.

I have had quite a few issues with getting errors for this competition so I hope I can help.

1 Like

Hello Andrew

Thanks for getting back to me - much appreciated.

Please could you let me know what error message you’re getting because I can see the notebook if I click on the link and the notebook is public

The approach I’m now taking is to move all the images into a single directory. This means I can now create a dataloader but am now getting an assertion error when I run

learn = vision_learner(dls, resnet18, metrics=error_rate, pretrained=False)
learn.fit_one_cycle(1)

which I think is happening with “fit_one_cycle”

All the best

Julian

I can see the notebook, whatever you did works now :slight_smile:

The notebook version I am looking at though still has the errors with building the dataloaders, did you move past that? You might have to save another version so I can see the most up to date issues you are running into.

Hi Andrew

Basically I abandoned this approach and moved all the images into a single directory. I now have a working learner but the accuracy leaves a great deal to be desired :slight_smile:

Do you know if it’s possible to close this post or mark it as resolved?

Many thanks again for your help

All the best

Julian

I am in the same boat with respect to accuracy.

Not sure about marking it as resolved, I am relatively new here still.