4 ms·
I know that's too much to ask, but if you implement simple into a "chooser" in https://github.com/avantoss/loop/tree/master/lib/choosers https://github.com/avan
by kirillseva 9y ago
I know that's too much to ask, but if you implement simple into a "chooser" in https://github.com/avantoss/loop/tree/master/lib/choosers https://github.com/avantoss/loop/tree/master/lib/choosers you'll have at least one company using this method!
We use a solution inspired by Jasper Snoek's Spearmint, which allows to somewhat parallelize bayesian search through a concept called "fantasies". If you've never read his paper I encourage you to find it on arxiv. The most gentle and thorough introduction to this topic that I've seen.
- chroem- 9y agoHi there, Unfortunately I have quite a backlog of other projects that I need to clear out, though Simple is AGPL so anyone is free to fork it. However I am also looking for employment opportunities. Simple is already highly parallelizeable due to the fact that it relies on dynamic programming. Adding more threads of execution is just a matter of popping off another candidate test point from the priority queue while the objective function is already being evaluated at another point. I was planning to include this feature in a more mature release.
- cvigoe 9y agoAny chance you could link the specific paper you’re talking about? This seems like a area I’d love to explore and the older you mentioned seems like a good starting point - I looked up the author on arxiv and there seems to be a few papers you could be referring to
- kirillseva 9y agoThis is the spearmint paper, describes how bayesian search algorithms can be parallelized in practice. There was also a very gentle introduction to bayesian optimization, I think it's one of the referenced papers. Cannot seem to find it right now. https://arxiv.org/pdf/1206.2944.pdf https://arxiv.org/pdf/1206.2944.pdf