Resume after break

[bentoayr]

Hello,

I am optimizing a very complicated function whose evaluation requires calling multiple programs via a connection to the cloud, etc.

Sometimes, connection breaks, which forces me to start all over.

Does rbfopt allow me to save the state of solving an optimization problem and resume after a fail?

Do you know of any black box optimizer that allows me to do that?

Thanks !

[gnannicini]

Yes, you can save the state.
You have to create your own script to create a RbfoptBlackBox and an RbfoptAlgorithm. After that, you can execute one iteration at a time "manually" by repeatedly running

alg.optimize(pause_after_iters=1)

where alg is your RbfoptAlgorithm object. The state can be saved with

alg.save_to_file(filename)

and there is a classmethod load_from_file(filename) if you want to restore.

[bentoayr]

Also, can the RbfoptAlgorithm be one of the existing algorithms provided by the package?

[gnannicini]

I don't understand this question. RbfoptAlgorithm is the algorithm provided by the package.

[bentoayr]

OK. This is very useful ! I'm new to python, is there some simple example with these ideas already implemented ? Thanks , J

…

On Sat, Oct 16, 2021 at 10:41 PM Giacomo Nannicini ***@***.***> wrote: Yes, you can save the state. You have to create your own script to create a RbfoptBlackBox and an RbfoptAlgorithm. After that, you can execute one iteration at a time "manually" by repeatedly running alg.optimize(pause_after_iters=1) where alg is your RbfoptAlgorithm object. The state can be saved with alg.save_to_file(filename) and there is a classmethod load_from_file(filename) if you want to restore. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#39 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AINNYVWFIZ6YFEOQDFWEZDLUHIZVRANCNFSM5GEFAXRQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

[gnannicini]

I'm not aware of an example specifically for this, sorry.
There is an example of loading the state from file here: https://github.com/coin-or/rbfopt/blob/master/bin/rbfopt_cl_interface.py

[bentoayr]

Hello again,

I tried to follow your instructions. I can save the state and load the state.
However, if I stop the optimization midway, and then I load the saved file using

alg = rbfopt.RbfoptAlgorithm.load_from_file("./saved_rbf_opt_state")

to try to resume the optimization, when I proceed to iterate using

val, x, itercount, evalcount, fast_evalcount = alg.optimize(pause_after_iters = 1)

the rbf just finishes without making any more iterations.

To load the previous state, should I also call again the rbfopt.RbfoptUserBlackBox and call again the rbfopt.RbfoptSettings ?

[gnannicini]

When you resume, the algorithm object alg already stores the settings and black-box function.

I am not sure why you are seeing that behavior. It is supposed to work. I encourage you to try to debug a little bit to see if the black box is called at all, and if you can't get it to work, then please open an issue, providing more information and possibly a way to replicate the problem so I can look into it.

[bentoayr]

Fixed ! It was my lack of python knowledge that caused then problem. fyi: I've checked several black-box solvers outthere, yours seems to be the best so far. Congrats and thanks for then good work !

…

On Mon, Oct 25, 2021 at 12:22 PM Giacomo Nannicini ***@***.***> wrote: When you resume, the algorithm object alg already stores the settings and black-box function. I am not sure why you are seeing that behavior. It is supposed to work. I encourage you to try to debug a little bit to see if the black box is called at all, and if you can't get it to work, then please open an issue, providing more information and possibly a way to replicate the problem so I can look into it. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#39 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AINNYVV6I4G4U5KZOSRHPZ3UIV75RANCNFSM5GEFAXRQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.