[Feature]: Training and testing splitting.

[VittoriaOssanna]

Feature Request

Good morning everyone,

I am new to PySR, so apologies if the questions I place are naiive.
I was wondering if there is any way to have an automatic "training and testing" split, and for training I mean the value for fitting, while for testing unseen data on which the loss can be calculated.

I am trying to implement this feature for my project, and I see that model.predict() evaluates the input based on the equation chosen by model_selection. Is there any way of evaluating the expression also on other candidate functions?

Thank you for your help!

[MilesCranmer]

You can use scikit learn’s train test split: https://scikit-learn.org/1.5/modules/generated/sklearn.model_selection.train_test_split.html, and then feed in the train split to .fit and the test split to .predict

Note that .predict also has a keyword for the equation index so you can test with other ones too. See the docs for more info: https://ai.damtp.cam.ac.uk/pysr/examples/

hope this helps!
Best,
Miles

[VittoriaOssanna]

Thank you for your answer!
I was also wondering at this point if there is a default method to calculate the loss on the test set, instead of predict.

Thank you again!