Get the score/loss associated with ground truth equations.

[mihirp1998]

Hi,

Thanks for the great work! I was wondering if there is a way to get the loss/score value associated with a known ground truth equation. For example i ran search using PySr however i found that it didn't find the correct equation, so i wanted to see the score value given to the ground truth equation that i have access to. This might be useful as i would want to know if the search couldn't find a equation with higher score or my complexity definition were bad which resulted in lower score for the ground truth equation.

[MilesCranmer]

So the loss is specified with elementwise_loss. By default this is just MSE. So if you have y and X you could potentially compute ypredicted directly in Python, and then compute np.sum(np.square(y - ypredicted)) / y.shape[0]. Does that work for you?

Note the "score" is just a metric to display users, it doesn't actually affect the search itself. The search is all based on loss, which by default is MSE.

See https://astroautomata.com/PySR/api/#the-objective for details

[mihirp1998]

Thanks for the reply. Although i thought score affects the best equation u pick and the best equation u pick depends on complexity of that equation? So i just wanted to see if the correct equation had a higher score (i.e a good ratio between complexity and loss)?

Plz let me know if i'm getting something wrong here