search

yanyachen / rBayesianOptimization

Bayesian Optimization of Hyperparameters
84 stars 21 forks source link

BayesOptimization function progressively slows down #43

Open asheetal opened 3 years ago

asheetal commented 3 years ago

Wondering if anyone else has observed this. I have 128GB RAM computer and trying to optimize ~6 parameters. Initially the function call triggers the child function and next set of parameters within a few seconds, but progressively the 300th iteration spaces out at 4-5 hours. So what I find is that bulk of time is spent in generating the next param set and hardly any time is spent in child process. Is an awful waste of GPU time that is mostly sitting idle waiting for the next set of params. Is the computation within the BayesianOptimization really so CPU-intensive? Can it be improved?

samFarrellDay commented 2 years ago

I have run into this. It is a limitation with the Gaussian Process fitting package, GPfit. I never found a way around it.