Open leightonvg opened 1 year ago
Thanks for the report! I was under the impression that this issue had been resolved.
The error is originally raised here.My memory is a bit fuzzy on this, but I think the deadline was a workaround because sometimes the stopit
module would not stop runtime evaluation in time, so I added a thin layer of manual checks (the deadline
stuff). Ideally I'd like to fix the core issue by either using a different module to accomplish the same thing as stopit
in a more reliable manner, or figure out why stopit
wasn't reliable in the first place. But a quick fix to the deadline
logic to patch things over would also be OK for now. I am a bit busy right now, but if that sounds like something you'd like to work on I am open to PRs :)
@leightonvg Any advancements in this direction?
I was running GAMA multiple times and sometimes it happens that the search phase does not end within the max_total_time. I have to interrupt GAMA to stop the search. The logs show the following behavior:
![image](https://user-images.githubusercontent.com/18681138/221818285-40b43089-5329-4739-9ac3-cb4c167c8aad.png)
I notice that the errors are raised in
Causing timeout issues in ![image](https://user-images.githubusercontent.com/18681138/221820468-2cc26256-5ecb-40b9-8e84-a25c8489d8f9.png)
gama.genetic_programming.operator_set
. I suppose that withinGama._search_phase()
it happens thatgama.genetic_programming.compilers.scikitlearn.evaluate_individual
is called beyond the deadline.gama.genetic_programming.compilers.scikitlearn.evaluate_individual
because the timeout is negative.I am not sure whether this is the cause of the errors and the non-stopping search space execution, and why it happens exactly. But do note that in the logs the timeout values are negative, and seem to correspond to when I would have expected the deadline based on a 120 second run. Therefore I believe sometimes individuals are evaluated after the deadline, causing GAMA to go on and eventually crash with memory issues.