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

Fix OFA valid/test #50

Merged
merged 3 commits into from
Sep 27, 2022
Merged

Fix OFA valid/test #50

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3c15c43
Unify valid and test geotypes code
yuikosakuma1 Sep 21, 2022
edde90b
Remove one_epoch variable
yuikosakuma1 Sep 22, 2022
724094e
Merge branch 'master' into feature/20220913-fix-ofa-valid and resolve…
hyingho Sep 27, 2022
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
63 changes: 27 additions & 36 deletions nnabla_nas/runner/searcher/ofa.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,23 +71,7 @@ def run(self):

# Test for init parameters
if self.hparams['task'] != 'fullnet':
OFAResize.IS_TRAINING = False
for genotype in self.hparams['valid_genotypes']:
for img_size in self.hparams['valid_image_size_list']:
self.monitor.reset()
OFAResize.ACTIVE_SIZE = img_size
self.model.set_valid_arch(genotype)
self.reset_running_statistics()
for i in tqdm(range(self.one_epoch_test), desc='Test for init parameters'):
self.update_graph('test')
self.valid_on_batch(is_test=True)
clear_memory_cache()
self.monitor.info(f'img_size={img_size}, genotype={genotype} \n')
self.callback_on_epoch_end(is_test=True)

self.loss.zero()
for k in self.metrics:
self.metrics[k].zero()
self.valid_genotypes(mode='test')

# training
for self.cur_epoch in range(self.cur_epoch, self.hparams['epoch']):
Expand All @@ -106,25 +90,8 @@ def run(self):
self.monitor.display(i, key=train_keys)
clear_memory_cache()
if self.cur_epoch % self.hparams["validation_frequency"] == 0:
OFAResize.IS_TRAINING = False
for genotype in self.hparams['valid_genotypes']:
for img_size in self.hparams['valid_image_size_list']:
self.monitor.reset()
OFAResize.ACTIVE_SIZE = img_size
self.model.set_valid_arch(genotype)
self.reset_running_statistics()
for i in tqdm(range(self.one_epoch_valid),
desc=f'Valid [{self.cur_epoch}/{self.hparams["epoch"]}]'):
self.update_graph('valid')
self.valid_on_batch(is_test=False)
clear_memory_cache()
self.monitor.info(f'img_size={img_size}, genotype={genotype} \n')
self.callback_on_epoch_end(is_test=False)
self.monitor.write(self.cur_epoch)

self.loss.zero()
for k in self.metrics:
self.metrics[k].zero()
self.valid_genotypes(mode='valid')

return self

def callback_on_start(self):
Expand Down Expand Up @@ -205,6 +172,30 @@ def valid_on_batch(self, is_test=False):
[self.loss] + list(self.metrics.values()), division=True, inplace=False)
self.event.add_default_stream_event()

def valid_genotypes(self, mode='valid'):
assert mode in ['valid', 'test']
is_test = True if mode == 'test' else False

OFAResize.IS_TRAINING = False
for genotype in self.hparams['valid_genotypes']:
for img_size in self.hparams['valid_image_size_list']:
self.monitor.reset()
OFAResize.ACTIVE_SIZE = img_size
self.model.set_valid_arch(genotype)
self.reset_running_statistics()
for _ in tqdm(range(self.one_epoch_valid if mode == 'valid' else self.one_epoch_test),
desc=f'{mode} [{self.cur_epoch}/{self.hparams["epoch"]}]'):
self.update_graph(mode)
self.valid_on_batch(is_test=is_test)
clear_memory_cache()
self.monitor.info(f'img_size={img_size}, genotype={genotype} \n')
self.callback_on_epoch_end(is_test=is_test)
self.monitor.write(self.cur_epoch)

self.loss.zero()
for k in self.metrics:
self.metrics[k].zero()

def callback_on_epoch_end(self, epoch=None, is_test=False, info=None):
if is_test:
num_of_samples = self.one_epoch_test * self.accum_test * self.mbs_test
Expand Down