Skip to content

Added patience parameter to BestCheckpointer hook #5429

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Added patience parameter to BestCheckpointer hook #5429

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9819cf1
Added early stopping & patience parameter, needs cleanup
palanjian Jan 23, 2025
3ccb191
Code cleanup
palanjian Jan 23, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 21 additions & 0 deletions detectron2/engine/hooks.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -222,6 +222,7 @@ def __init__(
val_metric: str,
mode: str = "max",
file_prefix: str = "model_best",
patience: int = None,
) -> None:
"""
Args:
Expand All @@ -231,10 +232,12 @@ def __init__(
mode (str): one of {'max', 'min'}. controls whether the chosen val metric should be
maximized or minimized, e.g. for "bbox/AP50" it should be "max"
file_prefix (str): the prefix of checkpoint's filename, defaults to "model_best"
patience (int): the number of evaluation cycles without improvement before early stopping
"""
self._logger = logging.getLogger(__name__)
self._period = eval_period
self._val_metric = val_metric
self._patience = patience
assert mode in [
"max",
"min",
Expand Down Expand Up @@ -297,6 +300,24 @@ def after_step(self):
and next_iter != self.trainer.max_iter
):
self._best_checking()

if self._patience is None or self.best_iter is None:
return

iterations_without_improvement = (self.trainer.iter-self.best_iter) // self._period

if(iterations_without_improvement > self._patience):
self._logger.info(
f"Early stopping triggered at iteration {self.trainer.iter} due to lack of improvement "
f"after {iterations_without_improvement} cycles."
)
raise Exception("Early stopping triggered. Terminating training process.")

if(iterations_without_improvement > 0):
self._logger.info(
f"No improvement detected in the last {iterations_without_improvement} evaluation cycles. "
f"{self._patience - iterations_without_improvement} cycles remain before early stopping."
)

def after_train(self):
# same conditions as `EvalHook`
Expand Down