Gaussian process surrogate models for the CMA-ES

c-bata / goptuna

A hyperparameter optimization framework, inspired by Optuna.

https://pkg.go.dev/github.com/c-bata/goptuna

MIT License

258 stars 22 forks source link

Open c-bata opened 4 years ago

c-bata commented 4 years ago

c-bata commented 4 years ago

The following paper describes the details of uncertainty criterion.

There are 4 types of uncertainty criteria.

The most simple approach which just uses the variance of the first model performs the best.

This and all the following experiments use the criterion Cs^2 which performed best in the first set of experiments

c-bata commented 4 years ago

In this paper, they use IPOP-CMA-ES. I implemented it at #131.