[GENERAL SUPPORT]: slow compared to other bayes opt package

[Yoonho-Na]

Question

Hi I've been using the bayesian optimization python package below. https://github.com/bayesian-optimization/BayesianOptimization

for the same experiment, the above package took about 45 mins (initital points: 20, iteration number: 250) but when I use Ax, it takes like 2 hours(max_trials=1, trial_number=250).

why is that? Is there any way to speed up the optimizing process in Ax?

Please provide any relevant code snippet if applicable.

Code of Conduct

• [x] I agree to follow this Ax's Code of Conduct

[VMLC-PV]

You could try BoTorch if you want speed. Ax is built on it anyway. Depending on the type of optimization you are doing, I would suggest trying the TURBO optimization, it is quick and works like a charm, at least for me

[Balandat]

@Yoonho-Na can you provide a bit more detail on the specific problem you're running (how many parameters and what kind, is this single-objective or multi-objective, are there any constraints either on parameters or black box outcomes, etc.)?

In general, Ax focuses on quality over speed and 250 iterations are quite a bit for typical black box optimization setting, so I'm not too surprised that under the default settings it is slower than the other library you mention. There are ways to speed up things, but it would be good to understand the specific problem setting first.