Integrating Symbolic Regression to get GP Mean Function, in Active Learning. #90
Replies: 3 comments 4 replies
-
Hi @utkarshp1161 - it seems to be somewhat similar to this work, where the mean functions for sGPs in active hypothesis learning came from SISSO analysis (symbolic regression + compressed sensing). Here's the associated paper. |
Beta Was this translation helpful? Give feedback.
-
Yes, that is correct
Ah, got it. Yes, that would be a new thing. |
Beta Was this translation helpful? Give feedback.
-
Closing this, as was able to parse the discovered functions as mean function to structured GP. Still trying to think how to go about assigning rewards in active learning. |
Beta Was this translation helpful? Give feedback.
-
I was playing with https://github.com/ziatdinovmax/gpax/blob/main/examples/GP_sGP.ipynb notebook and I tried one experiment where I fit the mean function using https://github.com/MilesCranmer/PySR and then used that prior in the GP to get the predictive mean and uncertainty[see last section of the notebook: https://file.io/DtTc01chwZ4Q].
I am not sure how useful this will be:
Can we accomodate this in active learning? As we are exploring points we get a new estimation of mean f;n which we use in fitting the gp?
Beta Was this translation helpful? Give feedback.
All reactions