How Can I force points to be explored?

52 views
Skip to first unread message

DevShark

unread,
May 12, 2016, 11:49:46 AM5/12/16
to BayesOpt discussion

The library is fantastic. However, how can I force the optimization process to explore certain points that I know are good?

Alternatively, how can I enter a set of points and their value?

Thanks

Ruben Martinez-Cantin

unread,
May 12, 2016, 12:01:08 PM5/12/16
to DevShark, BayesOpt discussion
Right now you can add initial points by hand but it's a little bit tricky.

You can use the save and restore functionality, start the optimization
and before doing anything, close it. That would have generated a save
file with the parameters and the format. Then you can add as many
points as you want in mX and mY as soon as you follow the format of
the file (the parser is not robust to file inconsistencies).

Best,

Ruben
> --
> You received this message because you are subscribed to the Google Groups
> "BayesOpt discussion" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to bayesopt-discus...@googlegroups.com.
> To post to this group, send email to bayesopt-...@googlegroups.com.
> Visit this group at https://groups.google.com/group/bayesopt-discussion.

DevShark

unread,
May 12, 2016, 1:15:32 PM5/12/16
to BayesOpt discussion, sebastien....@gmail.com, rmca...@unizar.es
Thanks. The problem when doing that is the following. Say I use mParameters.n_iterations=30 and mParameters.n_init_samples=10.
If it fails straight away, the last 2 lines of the file will look like this:
 mY=[0]()
mX=[10,4] ...

I can modify mX to include my starting points. However, when starting the process again from this starting point, it will only evaluate the function 10 times, and not the full 30 times. What can I do to make sure the optimization runs fully?

DevShark

unread,
May 12, 2016, 1:19:14 PM5/12/16
to BayesOpt discussion, sebastien....@gmail.com, rmca...@unizar.es
Actually, I just saw that it was because mCurrentIter was set at a very large number. I can now see that I need to let the process go through at least one iteration.
What are mCounterStuck and mYPrev?

Thanks

Ruben Martinez-Cantin

unread,
May 13, 2016, 4:29:07 AM5/13/16
to DevShark, BayesOpt discussion
Set both to 0.

Ruben

DevShark

unread,
May 13, 2016, 8:33:56 AM5/13/16
to BayesOpt discussion, sebastien....@gmail.com, rmca...@unizar.es
Thanks for your response. There is some precision issue. If I run an optimization, in mode "save file", then remove the mY values, set the current iteration at 0, it will go through almost the same points to explore, but not quite. I think it's because when you load the file, the resulting points are not quite the same as when you create them directly, because of precision issue. Is there a way to solve this?

Thanks 
Reply all
Reply to author
Forward
0 new messages