Ticket #4222 (new task)
|Reported by:||wehart||Owned by:||jdsiiro|
I'd like to finalize the semantics for 'maximum number of evaluations' for the COLIN 3.0 release.
In COLIN 2.0, there is a distinction between (1) the total number of allowed evaluations, and (2) the number of evaluations allowed by the current execution of an optimizer.
(2) is probably what most users think of when they think of a termination rule for optimizers like GAs.
(1) is a termination rule that makes sense for simple hybrid methods that share a single optimization application.
I'm not sure that (1) makes sense in the broader context that we're defining for COLIN 3.0. If we don't know who we're sharing an optimization application with, does this termination rule make sense?
I _think_ we should redefine max_neval to only refer to context (2). But, how would we define (1) for solvers like ABO? Certainly, it makes sense to have global-level control of this sort of hybridization management...
- Priority changed from blocker to major
- Milestone changed from Colin 3.0 to Colin 3.1