Aggregating losses from different tasks

[mikhmed-nabiev]

Hi, thank you for all your work! It is amazing!!!

Does PySR support multitask learning? Specifically, I have several datasets for different tasks and I would want to compute losses relative the datasets and aggregate them in some way. I have looked up documentation and it seems like the only way to do that is just combine all datasets into one matrix X and call .fit on it. But going this direction makes it impoossible to distinguish the datasets. Does anybody know any other way of approaching it?

[MilesCranmer]

Hi, thanks!

If you are ok with having fixed weightings, I would just make a custom loss function and sum up the different losses.

I recommend using a TemplateExpressionSpec for this and just putting all the datasets into X; then you can call an expression multiple times with different features each time.