search

wujian16 / Cornell-MOE

A Python library for the state-of-the-art Bayesian optimization algorithms, with the core implemented in C++.
Other
259 stars 64 forks source link

Unknown Variance in Noisy Function Evalution #74

Closed tyler-abbot closed 5 years ago

tyler-abbot commented 5 years ago

In all of my applications I do not actually know the true function to be evaluated. In the objective I simply set evaluate and evaluate_true to be the same noisy function. However, what do I set _sample_var to? I would like for the GP regression model to estimate the sample variance on each step, but it is hard to tell how to set this option.

wujian16 commented 5 years ago

You can just set it to be 0 (this parameter is not important in the algorithm), but remember to set noisy as true in the main.py. Then the algorithm will automatically learn the noise structure.

tyler-abbot commented 5 years ago

Thanks for the quick reply! I just saw this when looking through the MOE documentation.