search

SciML / DataDrivenDiffEq.jl

Data driven modeling and automated discovery of dynamical systems for the SciML Scientific Machine Learning organization
https://docs.sciml.ai/DataDrivenDiffEq/stable/
MIT License
404 stars 57 forks source link

State saving feature for SymbolicRegression.jl #329

Open MilesCranmer opened 2 years ago

MilesCranmer commented 2 years ago

I recently introduced a state saving feature in SymbolicRegression >=0.6.19. This lets you resume from previous searches which is useful for quick tweaks of hyperparameters for long-running searches.

If you set Options(..., stateReturn=true), then EquationSearch will return (saved_state, hallOfFame) instead of just hallOfFame. You can then call EquationSearch(..., saved_state=saved_state) and it will restart from your previous search.

There are some requirements for restarts: the operators (identical ordering as well) and the dataset should be identical. Otherwise you should restart a new search.

Let me know if you need any help in adding this. Cheers, Miles

AlCap23 commented 2 years ago

Oh, nice! I'll think it should just run given that I implemented the API just with kwargs... for any of the keywords. For the result, this could can be handled internally. I'll let you know if I need any help!