Spaces:

BVRA

/

SnakeCLEF2024

like 11

Running

Hi Mithilss,

Indeed, there is no metadata in the "SnakeCLEF2024-TestMetadata.csv".
What kind of metadata are you looking for?

Best,
Lukas

The same that is present in Train Metadata ie - country and endemic columns . From my experiments in local validation they improve they score by a good margin

Hello again,

Thank you for your description. Please see the explanation of why it is not present in the metadata file below :)

If a species is endemic, it means that it lives only within a specific country. Therefore, endemicity is unique for each species, and if we provided it, you could easily get the prediction. The information about the country of origin is removed on purpose. The test set originates from just a few countries, and if you know it, you could focus just on those countries. Which is something we do not want.

See last year's overview paper, where we explain last year's scenarios. This year, it work very similarly.

Was it helpful?

LP

Thanks for the help !

Hi @Mithilss ,

I have been thinking about your request, and I changed my mind a bit. I believe we should provide you with the information for the evaluation.
It is a little bit tricky as we do not want you to overfit to those "regions".
However, we are considering multiple ways of providing information about the country in evaluation.

Best,
Lukas

Where did this land, ultimately? Currently the test metadata in the example submission code only has the observation id and filename. Presumably this year there's no country codes, etc?