Runs in parrallel with bayesian optimisation

W&B Help
,
1

Hi everyone,

I started to use sweeps with bayesian optimisation. The idea is that the selected criterion in the futur runs will be impacted by the past runs.

But if a run starts when the last run isn’t over yet, will the last run results (which is not completed yet) have an impact over the selected criterion on the run which is up to start or will the selection of the parameters for the new run be only based on the runs that are already over ?

Thanks for your time.

2

Hi @felix_quinton, when running a parallel sweeps, we use both intermediate results in addition to results from jobs that have finished for selection of the new parameters. Our implementation is based on, reference1, and reference2, reference3

3

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.