fix: update fallback criterion (#1210)

* fix: update fallback criterion

* fix: ensure evo_fb is initialized and used correctly in fallback logic

* refactor: rename use_new_evo to should_use_new_evo for clarity
This commit is contained in:
you-n-g
2025-08-29 17:07:41 +08:00
committed by GitHub
parent f82de4a380
commit dbbe374ac8
4 changed files with 15 additions and 7 deletions
+8 -4
View File
@@ -77,7 +77,7 @@ class CoSTEER(Developer[Experiment]):
assert isinstance(fb, CoSTEERMultiFeedback), "feedback must be of type CoSTEERMultiFeedback"
return fb
def compare_and_pick_fb(self, base_fb: CoSTEERMultiFeedback | None, new_fb: CoSTEERMultiFeedback | None) -> bool:
def should_use_new_evo(self, base_fb: CoSTEERMultiFeedback | None, new_fb: CoSTEERMultiFeedback) -> bool:
"""
Compare new feedback with the fallback feedback.
@@ -110,14 +110,16 @@ class CoSTEER(Developer[Experiment]):
fallback_evo_exp = None
fallback_evo_fb = None
reached_max_seconds = False
evo_fb = None
for evo_exp in self.evolve_agent.multistep_evolve(evo_exp, self.evaluator):
assert isinstance(evo_exp, Experiment) # multiple inheritance
evo_fb = self._get_last_fb()
fallback_decision = self.compare_and_pick_fb(
update_fallback = self.should_use_new_evo(
base_fb=fallback_evo_fb,
new_fb=evo_fb,
)
if fallback_decision:
if update_fallback:
fallback_evo_exp = deepcopy(evo_exp)
fallback_evo_fb = deepcopy(evo_fb)
fallback_evo_exp.create_ws_ckp() # NOTE: creating checkpoints for saving files in the workspace to prevent inplace mutation.
@@ -139,7 +141,9 @@ class CoSTEER(Developer[Experiment]):
logger.info("Fallback to the fallback solution.")
evo_exp = fallback_evo_exp
evo_exp.recover_ws_ckp()
evo_exp = self._exp_postprocess_by_feedback(evo_exp, self._get_last_fb())
evo_fb = fallback_evo_fb
assert evo_fb is not None # multistep_evolve should run at least once
evo_exp = self._exp_postprocess_by_feedback(evo_exp, evo_fb)
except CoderError as e:
e.caused_by_timeout = reached_max_seconds
raise e