How to use torchmetrics.FBetaScore with weights?

1

Sorry about posting my question here, because I can NOT found the forum of torchmetrics.
If any one would pls tell me where is it, I’ll appreciate.

I need to use torchmetrics.FBetaScore with weights, but I can not understand the usage.
I’m resolving a binary classification problem, and the positive class should be more considered when validating.

Even though the function can accept a parameter “average” with “weighted” as real value, I need more detail example to mimic, and there is no sample code about it everywhere.

According to the docs literally, it looks like that weighted is of string type, but the type that I want to pass should be two floats.

How to pass two weights into the function?

Thanks!

2

Would using 'none' as the average argument work, as it would return the “raw” metric which you could then process in your custom weighting approach?

3

Thank you for your help!
I calculated them by myself, and found that they are same with the opponents in torchmetics.

The reason why I can’t understand the docs is that I mistake that the function accepts some weights as parameters.
In fact, “FBetaScore” only accepts a parameter named “average”, that only can be a string value such as ‘none’/‘weights’ and so on.
There is no opportunity we can specify some weights, they always accords to a fixed formula, if my comprehension is right.

Thanks!