Skip to content
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

Allow save checkpoints without validation data #73

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Allow save checkpoints without validation data #73

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion machine/trainer/supervised_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -221,7 +221,7 @@ def get_optim(optim_name):

callbacks = CallbackContainer(self,
[Logger(),
ModelCheckpoint(top_k=top_k),
ModelCheckpoint(top_k=top_k, save_last=dev_data is None),
History()] + custom_callbacks)

logs = self._train_epoches(data, num_epochs,
Expand Down
25 changes: 17 additions & 8 deletions machine/util/callbacks/model_checkpoint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,14 +11,19 @@ class ModelCheckpoint(Callback):
Model checkpoint to save weights during training.
This callback is automatically applied for every model that
is trained with the SupervisedTrainer.

Args:
save_last (optional, bool): if True, save last top_k models
instead of the best top_k models
"""

def __init__(self, top_k=5, monitor='val',
save_best_only=True):
save_last=False):
super(ModelCheckpoint, self).__init__()
self.top_k = top_k
self.monitor = monitor
self.save_best_only = save_best_only
self.save_last = save_last
self.next_index = 1

def set_trainer(self, trainer):
self.trainer = trainer
Expand All @@ -43,14 +48,18 @@ def on_batch_end(self, batch, info=None):

max_eval_loss = max(self.loss_best)

if total_loss < max_eval_loss:
index_max = self.loss_best.index(max_eval_loss)
if total_loss < max_eval_loss or self.save_last:
if self.save_last:
index_to_overwrite = self.next_index
self.next_index = (self.next_index + 1) % self.top_k
else:
index_to_overwrite = self.loss_best.index(max_eval_loss)
# rm prev model
if self.best_checkpoints[index_max] is not None:
if self.best_checkpoints[index_to_overwrite] is not None:
shutil.rmtree(os.path.join(
self.expt_dir, self.best_checkpoints[index_max]))
self.best_checkpoints[index_max] = model_name
self.loss_best[index_max] = total_loss
self.expt_dir, self.best_checkpoints[index_to_overwrite]))
self.best_checkpoints[index_to_overwrite] = model_name
self.loss_best[index_to_overwrite] = total_loss

# save model
Checkpoint(model=self.trainer.model,
Expand Down