Is there a way to avoid running the same experiment twice?

[yuvval]

I am migrating my experimental framework (with basic scheduling) to sacred.

My framework scheduler might call same experiment multiple times from different machines. In case an experiment is COMPLETED or RUNNING, then a pre-hook exits the experiment (skipping the execution of the main function). Otherwise (when the experiment is DEAD or does not exists), the experiment is launched . Or in other words, it avoids running the same experiment twice.

Is there a way to have a similar functionality in sacred?

Thanks!

[theRealSuperMario]

having the same question.

[yuvval]

I have published an example project for integrating SACRED and HyperOpt distributed training. I also implemented there the feature that allows to avoid running the same experiment twice. You can check it out here: https://github.com/yuvalatzmon/SACRED_HyperOpt_v2

…

On Tue, Feb 26, 2019 at 2:12 PM theRealSuperMario ***@***.***> wrote: having the same question. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#351 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AFo0e3mPzeGNV-V2B5tFo8GqB7ujEFbNks5vRSSpgaJpZM4WJz8i> .

[JarnoRFB]

@yuvval nice project! I think a functionality like this might be worth directly integrating into sacred. In addition to checking for the same config, I think one would also need to check for the same source, dependencies and options. Then one could have a new option to check for existing experiments like --check_duplicates.