Confusion about Normalisation of Dataset. #452
-
Hello! PySR has turned out to be a hidden gem for us recently on certain problems. I might be drawing a parallel to NN here but do we need to normalise our data for Symbolic regression in general? or both of the approaches are equivalent? |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
Hi @anubhavkamal, You don’t need to normalize data in PySR, as it can find constants of any magnitude, including scales that are very large or small. However, if you are okay with the extra complexity of normalization constants, then it can definitely help the search if you normalize. The reason for this is that new constants in the equations are initialized with a call to cheers, |
Beta Was this translation helpful? Give feedback.
-
Thank you for the answer! |
Beta Was this translation helpful? Give feedback.
Hi @anubhavkamal,
You don’t need to normalize data in PySR, as it can find constants of any magnitude, including scales that are very large or small. However, if you are okay with the extra complexity of normalization constants, then it can definitely help the search if you normalize. The reason for this is that new constants in the equations are initialized with a call to
randn()
so it can be preferable for the initial constants to be similar magnitude to the data. But it’s highly problem dependent!cheers,
Miles