def run_omniglot():
omniglot_database = OmniglotDatabase(
random_seed=47,
num_train_classes=1200,
num_val_classes=100,
)
maml = ModelAgnosticMetaLearningModel(
database=omniglot_database,
network_cls=SimpleModel,
n=20,
k_ml=1,
k_val_ml=5,
k_val=1,
k_val_val=15,
k_test=5,
k_val_test=15,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.4,
num_steps_validation=5,
save_after_iterations=1000,
meta_learning_rate=0.001,
report_validation_frequency=50,
log_train_images_after_iteration=200,
num_tasks_val=100,
clip_gradients=False,
experiment_name='omniglot',
val_seed=42,
val_test_batch_norm_momentum=0.0
)
# maml.train(iterations=5000)
maml.evaluate(iterations=50, num_tasks=1000, use_val_batch_statistics=True, seed=42)
def run_voxceleb():
voxceleb_database = VoxCelebDatabase()
maml = ModelAgnosticMetaLearningModel(
database=voxceleb_database,
network_cls=VoxCelebModel,
n=5,
k_ml=1,
k_val_ml=5,
k_val=1,
k_val_val=15,
k_test=1,
k_val_test=15,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=15000,
meta_learning_rate=0.001,
report_validation_frequency=1000,
log_train_images_after_iteration=-1,
num_tasks_val=100,
clip_gradients=True,
experiment_name='voxceleb3',
val_seed=42,
val_test_batch_norm_momentum=0.0,
)
# maml.train(iterations=60040)
maml.evaluate(50, num_tasks=1000, seed=42, use_val_batch_statistics=True)
def run_cub():
cub_database = CUBDatabase()
maml = ModelAgnosticMetaLearningModel(
database=cub_database,
test_database=MiniImagenetDatabase(),
network_cls=MiniImagenetModel,
n=5,
k_ml=1,
k_val_ml=5,
k_val=1,
k_val_val=15,
k_test=5,
k_val_test=15,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=15000,
meta_learning_rate=0.001,
report_validation_frequency=1000,
log_train_images_after_iteration=1000,
num_tasks_val=100,
clip_gradients=True,
experiment_name='cub',
val_seed=42,
val_test_batch_norm_momentum=0.0,
)
maml.train(iterations=60000)
maml.evaluate(50, num_tasks=1000, seed=42, use_val_batch_statistics=True)
maml.evaluate(50, num_tasks=1000, seed=42, use_val_batch_statistics=False)
def run_airplane():
airplane_database = AirplaneDatabase()
maml = ModelAgnosticMetaLearningModel(
database=airplane_database,
# test_database=Omniglot84x84Database(random_seed=47, num_train_classes=1200, num_val_classes=100),
test_database=MiniImagenetDatabase(),
network_cls=MiniImagenetModel,
n=5,
k_ml=1,
k_val_ml=5,
k_val=1,
k_val_val=15,
k_test=5,
k_val_test=15,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=15000,
meta_learning_rate=0.001,
report_validation_frequency=1000,
log_train_images_after_iteration=1000,
num_tasks_val=100,
clip_gradients=True,
experiment_name='airplane',
val_seed=42,
val_test_batch_norm_momentum=0.0,
)
# maml.train(iterations=60000)
maml.evaluate(50, num_tasks=1000, seed=42, use_val_batch_statistics=True)
maml.evaluate(50, num_tasks=1000, seed=42, use_val_batch_statistics=False)
def run_traffic_sign():
mscoco_database = MSCOCODatabase()
maml = ModelAgnosticMetaLearningModel(
database=mscoco_database,
network_cls=MiniImagenetModel,
n=5,
k_ml=1,
k_val_ml=5,
k_val=1,
k_val_val=15,
k_test=1,
k_val_test=15,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=15000,
meta_learning_rate=0.001,
report_validation_frequency=1000,
log_train_images_after_iteration=1000,
num_tasks_val=100,
clip_gradients=True,
experiment_name='mscoco',
val_seed=42,
val_test_batch_norm_momentum=0.0,
)
# This dataset is only for evaluation
maml.evaluate(50, num_tasks=1000, seed=42, use_val_batch_statistics=True)
maml.evaluate(50, num_tasks=1000, seed=42, use_val_batch_statistics=False)
def run_celeba():
celeba_database = CelebADatabase()
maml = ModelAgnosticMetaLearningModel(
database=celeba_database,
network_cls=MiniImagenetModel,
n=5,
k_ml=1,
k_val_ml=5,
k_val=1,
k_val_val=15,
k_test=1,
k_val_test=15,
meta_batch_size=4,
num_steps_ml=1,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=5000,
meta_learning_rate=0.0001,
report_validation_frequency=250,
log_train_images_after_iteration=1000,
num_tasks_val=100,
clip_gradients=True,
experiment_name='celeba'
)
maml.train(iterations=60000)
maml.evaluate(50, num_tasks=1000, seed=42)