Giters

JasperSnoek / spearmint

Spearmint is a package to perform Bayesian optimization according to the algorithms outlined in the paper: Practical Bayesian Optimization of Machine Learning Algorithms. Jasper Snoek, Hugo Larochelle and Ryan P. Adams. Advances in Neural Information Processing Systems, 2012

Home Page:http://people.seas.harvard.edu/~jsnoek/software.html

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

specifying the number of initial points in Spearmint

ntienvu opened this issue · comments

commented

Hi There,
I am Vu Nguyen from Deakin University, Australia.
I want to use Spearmint as a baseline for my paper. I want to set the number of initial points (e.g., at the beginning of the Bayesian Optimization procedure). I assume this can be done in the config.json. However, I can specify the "max_finished_jobs", but not the # initial points.
Please help me.
Regards,
Vu

commented

Hey @ntienvu, this code is a bit dated now and probably isn't the best baseline to compare to (e.g. https://github.com/HIPS/Spearmint is more up to date - particularly the PESC branch). However, the flag --grid-size will do this for you (https://github.com/JasperSnoek/spearmint/blob/master/spearmint/spearmint/main.py#L83).

commented

Sorry and you can use spearmint-lite to initialize the experiment with whatever results you would like (just edit the flat file).

commented

Thanks JasperSnoeak for your answer. I will try PESC branch.
Cheers,