n_iter: Adding multiple results for one experiment evaluation

[ibayer]

Hi, one of my hyperparameter is the number of iterations n_iter. If an experiment with n_iter = 50 is executed I would like that the results for iteration 10, 20, 30, 40 also influence the selection of the next experiment. I guess I'm looking for a way to add multiple result for one experiment evaluation to trails. Is there a recommended way to do this? Best, Immanuel

PS Would be okay if it only works for the serial search

[jaberg]

This was supported at one time, try searching the code for "inject" or "inject jobs". It is possible for a job to use the "ctrl" object it receives to inject trial results directly into the global Trials database.

I'm afraid I also remember there being some slightly dodgy side-effects of this procedure, but not precisely what they were (whether it was something wrong with my particular function or the mechanism in general). I suggest exploring this with a fast-running function that you can use to test the mechanism. It's a powerful technique, so thanks for asking about it!

• JB

On Mon, Mar 9, 2015 at 12:48 PM, ibayer [email protected] wrote:

Hi, one of my hyperparameter is the number of iterations n_iter. If an experiment with n_iter = 50 is executed I would like that the results for iteration 10, 20, 30, 40 also influence the selection of the next experiment. I guess I'm looking for a way to add multiple result for one experiment evaluation to trails. Is there a recommended way to do this? Best, Immanuel

— Reply to this email directly or view it on GitHub https://github.com/hyperopt/hyperopt/issues/222.

[ibayer]

thx

[github-actions[bot]]

This issue has been marked as stale because it has been open 120 days with no activity. Remove the stale label or comment or this will be closed in 30 days.