Sam Daulton comments

Results 64 comments of


                                            Sam Daulton

Repeated trials without convergence

Hi @HBSaddiq, Thanks for raising the issue. A few questions: 1) What is are your objective thresholds (i.e. reference point)? This can significantly influence the optimization behavior. 2) What is...

Setting up robust optimization experiment with MVaR gives error

MVaR is not differentiable, so gradient issues not terribly surprising. To get unblocked on this, a recommended alternative is to use MARS (https://proceedings.mlr.press/v162/daulton22a.html) which is way faster and differentiable than...

Setting up robust optimization experiment with MVaR gives error

Glad that unblocked you! MARS optimizes MVaR by optimizing the VaR of random Chebyshev scalarizations. Since it scalarizes the problem, it uses a single-objective acquisition function.

Setting up robust optimization experiment with MVaR gives error

@saitcakmak, did the differentiable MVaR version resolve the NaN issue?

evaluate_acquisition_function does not work correctly

Hi @lsassen, Can you share your code including setting up AxClient and the GenerationStrategy for repro? Thanks!

evaluate_acquisition_function does not work correctly

Closing due to lack of activity. Feel free to reopen @lsassen if the fix did not work.

Repeated trials without convergence

Yes, that sounds quite plausible. qLogNEHVI (https://arxiv.org/abs/2310.20708) ---which is now the default in Ax for multi-objective optimization---should work quite a bit better when the AF surface is very flat.

Feature Request: surrogate models of the objectives

That sounds great! Random Forests make sense (perhaps on the average value from multiple measurements if noise is a concern). Thanks!

Chemistry and Material Science Datasets

Any update on this?

Extending Models.THOMPSON with an extra parameter

Hi @seabull, Feel free to open a pull request that adds this new parameter to `Models.THOMPSON`! Contributions are always welcome!