search

dflemin3 / approxposterior

A Python package for approximate Bayesian inference and optimization using Gaussian processes
https://dflemin3.github.io/approxposterior/
MIT License
41 stars 9 forks source link

Optimize the GP less #28

Closed dflemin3 closed 5 years ago

dflemin3 commented 5 years ago

Currently, I re-optimize the GP hyperparameters every time I add a new design point. This can get computationally expensive, so it may be a good idea to add the option to re-optimize the GP even nGPOpt times AP adds a new design point.

dflemin3 commented 5 years ago

Added tons of new kwargs that allow user to tweak how GP hyperparameters are optimized and how utility functions are optimizes when selecting new design points for the GP in the 0.2 release candidate on the dev branch. I also added means to restart the optimization several times to hopefully mitigate the impact of local extrema that can hamper the aforementioned optimization.