Open raybellwaves opened 3 years ago
I know you can't really compare a bandit approach with a Bayesian approach.
Combing Hyperband and Bayesian sampling is an open issue for Dask-ML: https://github.com/dask/dask-ml/issues/697.
I think that it would be simplest to include an isolated dask-optuna notebook into the examples. I think that that would have good value. It might be useful after that to have a comparison example?
On Tue, Oct 20, 2020 at 9:02 PM Scott Sievert notifications@github.com wrote:
I know you can't really compare a bandit approach with a Bayesian approach.
Combing Hyperband and Bayesian sampling is an open issue for Dask-ML: dask/dask-ml#697 https://github.com/dask/dask-ml/issues/697.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/jrbourbeau/dask-optuna/issues/18#issuecomment-713284709, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACKZTEOXZ2PFAWLT7KALWTSLZMNBANCNFSM4SZETFDQ .
@raybellwaves is this something that you would be interested in?
Wonder if this could fit into https://examples.dask.org/machine-learning/hyperparam-opt.html or a different notebook?
Could even apply to the setting for the Hyperband example? I would be interested in comparisons against Hyberband. However, I know you can't really compare a bandit approach with a Bayesian approach.