def _05_05():
res = []
for label in [NEGATIVE_NETWORK, STREAKS, GLOBULES]:
for lr in [1e-5]:
res.append(
RunConfig(lr=lr,
label=label,
epochs=100,
when_present=True,
tags=['train_new'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2),
balanced=True))
res.append(
RunConfig(lr=lr,
label=label,
epochs=100,
when_present=True,
tags=['train_new'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='none',
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2),
balanced=True))
return res
def _02_05():
res = []
for label in LABELS:
for lr in [1e-4, 1e-5]:
res.append(
RunConfig(lr=lr,
label=label,
epochs=100,
when_present=True,
tags=['train_new'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2)))
res.append(
RunConfig(lr=lr,
label=label,
epochs=100,
when_present=True,
tags=['train_new'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='none',
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2)))
return res
def _14_05_neg():
res = []
for label in [GLOBULES]:
for lr in [1e-4]:
for ae_loss in ['none', 'mse']:
for neg_p in [80, 60, 50]:
res.append(
RunConfig(lr=lr,
label=label,
epochs=100,
when_present=True,
tags=['train_new'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion=ae_loss,
device='cuda:1',
seed=numpy.random.randint(low=1,
high=2**32 - 2),
no_scheduler=True,
balanced=False,
attention_loss_lambda=1,
no_rand_dataloader=False,
train_on_test=False,
neg_percent=neg_p))
return res
def _09_05_2():
res = []
for label in [GLOBULES]:
for ae_loss in ['none', 'mse']:
res.append(
RunConfig(lr=1e-5,
label=label,
epochs=100,
when_present=True,
tags=['train_new', 'represent'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion=ae_loss,
device='cuda:1',
no_scheduler=True,
seed=3521001796,
balanced=False,
train_on_test=True,
no_rand_dataloader=False,
grouping_label='manual-seed'))
res.append(
RunConfig(lr=1e-5,
label=label,
epochs=100,
when_present=True,
tags=['train_new', 'represent'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion=ae_loss,
device='cuda:1',
seed=numpy.random.randint(low=1, high=2**32 - 2),
no_scheduler=True,
balanced=False,
train_on_test=True,
no_rand_dataloader=True,
grouping_label='not-random-dataloader'))
return res
def _04_05():
res = []
for label in LABELS:
for lr in [1e-4, 1e-5]:
res.append(
RunConfig(lr=lr,
label=label,
epochs=100,
when_present=True,
tags=['train_new'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2),
balanced=True))
res.append(
RunConfig(lr=lr,
label=label,
epochs=100,
when_present=True,
tags=['train_new'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='none',
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2),
balanced=True))
res = [
x for x in res if x.label != PIGMENT_NETWORK and (
x.label != MILIA_LIKE_CYST and x.lr != 1e-5)
]
return res
def check_03_05():
return [
RunConfig(lr=1e-4,
label=PIGMENT_NETWORK,
epochs=2,
when_present=True,
tags=['checking-run'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2),
balanced=True)
]
def _08_04_2():
res = []
for label in [GLOBULES]:
for ae_loss in ['none', 'mse']:
res.append(
RunConfig(lr=1e-5,
label=label,
epochs=100,
when_present=True,
tags=['train_new', 'correct-globules-lr'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion=ae_loss,
device='cuda:1',
no_scheduler=True,
seed=numpy.random.randint(low=1, high=2**32 - 2),
balanced=False,
train_on_test=True))
return res
def _09_05():
res = []
for label in [GLOBULES]:
for ae_loss in ['none', 'mse']:
res.append(
RunConfig(lr=1e-5,
label=label,
epochs=100,
when_present=True,
tags=['train_new', 'manual-seed'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion=ae_loss,
device='cuda:1',
no_scheduler=True,
seed=3521001796,
balanced=False,
train_on_test=True,
no_rand_dataloader=True))
return res
def _11_05():
return [
RunConfig(
lr=1e-4,
label=GLOBULES,
epochs=100,
when_present=True,
tags=[
'train_new', '11_05_balanced', '11_05_bce_ae', '11_05_lambda'
],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
seed=numpy.random.randint(low=1, high=2**32 - 2),
no_scheduler=True,
balanced=False,
),
RunConfig(
lr=1e-4,
label=GLOBULES,
epochs=100,
when_present=True,
tags=['train_new', '11_05_bce_ae'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='bce',
device='cuda:1',
seed=numpy.random.randint(low=1, high=2**32 - 2),
no_scheduler=True,
balanced=False,
),
RunConfig(
lr=1e-4,
label=GLOBULES,
epochs=100,
when_present=True,
tags=['train_new', '11_05_balanced'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='none',
device='cuda:1',
seed=numpy.random.randint(low=1, high=2**32 - 2),
no_scheduler=True,
balanced=False,
),
RunConfig(lr=1e-4,
label=GLOBULES,
epochs=100,
when_present=True,
tags=['train_new', '11_05_lambda'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
seed=numpy.random.randint(low=1, high=2**32 - 2),
no_scheduler=True,
balanced=False,
attention_loss_lambda=10),
RunConfig(lr=1e-4,
label=GLOBULES,
epochs=100,
when_present=True,
tags=['train_new', '11_05_lambda'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
seed=numpy.random.randint(low=1, high=2**32 - 2),
no_scheduler=True,
balanced=False,
attention_loss_lambda=100),
RunConfig(lr=1e-4,
label=GLOBULES,
epochs=100,
when_present=True,
tags=['train_new', '11_05_lambda'],
images_size=512,
cl_criterion='weight-bce',
attention_criterion='mse',
device='cuda:1',
seed=numpy.random.randint(low=1, high=2**32 - 2),
no_scheduler=True,
balanced=False,
attention_loss_lambda=1000),
]