Bounded normal parameters

[Artur Luis]

Hello,

I'm running Hyperopt (with the TPE method) on a function with multiple parameters defined in the [0, 1] range. The function is not defined outside of this range. I would like to define a gaussian distribution for these parameters using the hp.normal parameter type, but still keep the parameter bounded (i.e. avoid values below 0 or above 1). Is there a way to achieve this in Hyperopt?

Thanks,

[Uvnik]

Hello Artur,

Unfortunately there is no way in Hyperopt to externally constrain a search function. This is the reason I moved from Hyperopt to Optuna. I have written an article comparing Hyperopt and Optuna. You may want to refer it : https://www.linkedin.com/pulse/hyper-parameter-tuning-hyperopt-optuna-uvnik-gupta/?trackingId=ZEruQGrfKXiYHiBREw7vUA%3D%3D

Cheers