search

yanyachen / rBayesianOptimization

Bayesian Optimization of Hyperparameters
81 stars 21 forks source link

Elapsed time doesn't match real stopwatch #16

Open eternalmothra opened 7 years ago

eternalmothra commented 7 years ago

After running 20 random initialization rounds, starting at about the 20th round of the algorithm, the elapsed time (printed on the screen) does not match the time I get if I time the code with stop watch. Elapsed will say ~30 seconds while my stopwatch will say 6 minutes.

I'm using this code on an XGBoost model, with acquisition function UCB and my objective function is AUC.

yanyachen commented 7 years ago

I'm guessing that you are optimizing at least 4 hyper-parameters. The elapsed time is logging for your CV running. It doesn't count the time for hyper-parameters computing. Since I'm limited by the speed of the GP backend. I suggest you try to decrease the number of hyper-parameters you are optimizing.

MichaelChirico commented 6 years ago

It might be worth adding a progress meter for the GP fitting, or otherwise give the user some clue that GP is what's causing the slowdown