-
Notifications
You must be signed in to change notification settings - Fork 737
Closed
Labels
bugSomething isn't workingSomething isn't working
Description
Assessing whether a program performs better than an other program, is currently governed by a hard-coded threshold of 0.02 in controller.py:
openevolve/openevolve/controller.py
Lines 368 to 372 in dd2e3e4
| # If the combined_score of this program is significantly better, use it instead | |
| if ( | |
| best_by_combined.metrics["combined_score"] | |
| > best_program.metrics["combined_score"] + 0.02 | |
| ): |
What "significantly" means, depends a lot on the nature of the combined_score being used. Therefore, hard-coding 0.02 does not always make sense.
My proposal is to make it a config parameter that can also be put to zero if needed.
Metadata
Metadata
Assignees
Labels
bugSomething isn't workingSomething isn't working