def test_create_model(self):
# Check output of encoder has shape specified in paper
encoder = get_few_shot_encoder(num_input_channels=1).float()
omniglot = OmniglotDataset('background')
self.assertEqual(
encoder(omniglot[0][0].unsqueeze(0).float()).shape[1], 64,
'Encoder network should produce 64 dimensional embeddings on Omniglot dataset.'
)
encoder = get_few_shot_encoder(num_input_channels=3).float()
omniglot = MiniImageNet('background')
self.assertEqual(
encoder(omniglot[0][0].unsqueeze(0).float()).shape[1], 1600,
'Encoder network should produce 1600 dimensional embeddings on miniImageNet dataset.'
)
def __init__(self, dataset: str, num_tasks: int, n_shot: int, k_way: int,
q_queries: int, distance_metric: str,
open_world_testing: bool):
self.dataset_name = dataset
self.num_tasks = num_tasks
self.n_shot = n_shot
self.k_way = k_way
self.q_queries = q_queries
self.episodes_per_epoch = 10
self.distance_metric = 'l2'
self.open_world_testing = open_world_testing
self.prepare_batch = prepare_nshot_task(self.n_shot, self.k_way,
self.q_queries)
self.num_different_models = 0
if dataset == 'whoas':
self.evaluation_dataset = Whoas('evaluation')
#self.evaluation_dataset = Whoas('background')
elif dataset == 'kaggle':
self.evaluation_dataset = Kaggle('evaluation')
elif dataset == 'miniImageNet':
self.evaluation_dataset = MiniImageNet('evaluation')
else:
raise (ValueError, 'Unsupported dataset')
self.batch_sampler = NShotCustomTaskSampler(self.evaluation_dataset,
self.episodes_per_epoch,
n_shot, k_way, q_queries,
num_tasks, None,
open_world_testing)
self.evaluation_taskloader = DataLoader(
self.evaluation_dataset, batch_sampler=self.batch_sampler)
assert torch.cuda.is_available()
self.device = torch.device('cuda')
torch.backends.cudnn.benchmark = True
self.model = get_few_shot_encoder(
self.evaluation_dataset.num_input_channels)
self.model.to(self.device, dtype=torch.double)
self.optimiser = Adam(self.model.parameters(), lr=1e-3)
self.loss_fn = torch.nn.NLLLoss().cuda()
batch_sampler=NShotTaskSampler(background, episodes_per_epoch, args.n_train, args.k_train, args.q_train),
num_workers=4
)
# evaluation = dataset_class('images_evaluation')
evaluation = dataset_class('datasets/test')
evaluation_taskloader = DataLoader(
evaluation,
batch_sampler=NShotTaskSampler(evaluation, episodes_per_epoch, args.n_test, args.k_test, args.q_test),
num_workers=4
)
#########
# Model #
#########
model = get_few_shot_encoder(num_input_channels)
if args.retrain:
model.load_state_dict(torch.load(PATH + f'/models/proto_nets/{param_str}.pth'))
model.to(device, dtype=torch.double)
############
# Training #
############
print(f'Training Prototypical network on {args.dataset}...')
if args.retrain:
param_str = f'{args.dataset}_nt={args.n_train}_kt={args.k_train}_qt={args.q_train}_' \
f'nv={args.n_test}_kv={args.k_test}_qv={args.q_test}_experiment={args.experiment_name}_retrain=True'
optimiser = Adam(model.parameters(), lr=1e-3)
loss_fn = torch.nn.NLLLoss().cuda()
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--model_path",
type=str,
default=
"./models/proto_nets/miniImageNet_nt=5_kt=5_qt=10_nv=5_kv=5_qv=10_dist=l2_sampling_method=True_is_diverisity=True.pth",
help="model path")
parser.add_argument(
"--result_path",
type=str,
default="./results/proto_nets/5shot_training_5shot_diverisity.csv",
help="Directory for evaluation report result (for experiments)")
parser.add_argument('--dataset', type=str, required=True)
parser.add_argument('--distance', default='cosine')
parser.add_argument('--n_train', default=1, type=int)
parser.add_argument('--n_test', default=1, type=int)
parser.add_argument('--k_train', default=5, type=int)
parser.add_argument('--k_test', default=5, type=int)
parser.add_argument('--q_train', default=15, type=int)
parser.add_argument('--q_test', default=15, type=int)
parser.add_argument(
"--debug",
action="store_true",
help="set logging level DEBUG",
)
args = parser.parse_args()
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.DEBUG if args.debug else logging.INFO,
)
###################
# Create datasets #
###################
episodes_per_epoch = 600
if args.dataset == 'miniImageNet':
n_epochs = 5
dataset_class = MiniImageNet
num_input_channels = 3
else:
raise (ValueError('need to make other datasets module'))
test_dataset = dataset_class('test')
test_dataset_taskloader = DataLoader(
test_dataset,
batch_sampler=NShotTaskSampler(test_dataset, episodes_per_epoch,
args.n_test, args.k_test, args.q_test),
num_workers=4)
#########
# Model #
#########
model = get_few_shot_encoder(num_input_channels).to(device,
dtype=torch.double)
model.load_state_dict(torch.load(args.model_path), strict=False)
model.eval()
#############
# Inference #
#############
logger.info("***** Epochs = %d *****", n_epochs)
logger.info("***** Num episodes per epoch = %d *****", episodes_per_epoch)
result_writer = ResultWriter(args.result_path)
# just argument (function: proto_net_episode)
prepare_batch = prepare_nshot_task(args.n_test, args.k_test, args.q_test)
optimiser = Adam(model.parameters(), lr=1e-3)
loss_fn = torch.nn.NLLLoss().cuda()
train_iterator = trange(
0,
int(n_epochs),
desc="Epoch",
)
for i_epoch in train_iterator:
epoch_iterator = tqdm(
test_dataset_taskloader,
desc="Iteration",
)
seen = 0
metric_name = f'test_{args.n_test}-shot_{args.k_test}-way_acc'
metric = {metric_name: 0.0}
for _, batch in enumerate(epoch_iterator):
x, y = prepare_batch(batch)
loss, y_pred = proto_net_episode(model,
optimiser,
loss_fn,
x,
y,
n_shot=args.n_test,
k_way=args.k_test,
q_queries=args.q_test,
train=False,
distance=args.distance)
seen += y_pred.shape[0]
metric[metric_name] += categorical_accuracy(
y, y_pred) * y_pred.shape[0]
metric[metric_name] = metric[metric_name] / seen
logger.info("epoch: {}, categorical_accuracy: {}".format(
i_epoch, metric[metric_name]))
result_writer.update(**metric)
def few_shot_training(datadir=DATA_PATH,
dataset='fashion',
num_input_channels=3,
drop_lr_every=20,
validation_episodes=200,
evaluation_episodes=1000,
episodes_per_epoch=100,
n_epochs=80,
small_dataset=False,
n_train=1,
n_test=1,
k_train=30,
k_test=5,
q_train=5,
q_test=1,
distance='l2',
pretrained=False,
monitor_validation=False,
n_val_classes=10,
architecture='resnet18',
gpu=None):
setup_dirs()
if dataset == 'fashion':
dataset_class = FashionProductImagesSmall if small_dataset \
else FashionProductImages
else:
raise (ValueError, 'Unsupported dataset')
param_str = f'{dataset}_nt={n_train}_kt={k_train}_qt={q_train}_' \
f'nv={n_test}_kv={k_test}_qv={q_test}_small={small_dataset}_' \
f'pretrained={pretrained}_validate={monitor_validation}'
print(param_str)
###################
# Create datasets #
###################
# ADAPTED: data transforms including augmentation
resize = (80, 60) if small_dataset else (400, 300)
background_transform = transforms.Compose([
transforms.RandomResizedCrop(resize, scale=(0.8, 1.0)),
# transforms.RandomGrayscale(),
transforms.RandomPerspective(),
transforms.RandomHorizontalFlip(),
# transforms.Resize(resize),
transforms.ToTensor(),
# transforms.Normalize(mean=[0.485, 0.456, 0.406],
# std=[0.229, 0.224, 0.225])
])
evaluation_transform = transforms.Compose([
transforms.Resize(resize),
# transforms.CenterCrop(224),
transforms.ToTensor(),
# transforms.Normalize(mean=[0.485, 0.456, 0.406],
# std=[0.229, 0.224, 0.225])
])
if monitor_validation:
if not n_val_classes >= k_test:
n_val_classes = k_test
print("Warning: `n_val_classes` < `k_test`. Take a larger number"
" of validation classes next time. Increased to `k_test`"
" classes")
# class structure for background (training), validation (validation),
# evaluation (test): take a random subset of background classes
validation_classes = list(
np.random.choice(dataset_class.background_classes, n_val_classes))
background_classes = list(
set(dataset_class.background_classes).difference(
set(validation_classes)))
# use keyword for evaluation classes
evaluation_classes = 'evaluation'
# Meta-validation set
validation = dataset_class(datadir,
split='all',
classes=validation_classes,
transform=evaluation_transform)
# ADAPTED: in the original code, `episodes_per_epoch` was provided to
# `NShotTaskSampler` instead of `validation_episodes`.
validation_sampler = NShotTaskSampler(validation, validation_episodes,
n_test, k_test, q_test)
validation_taskloader = DataLoader(validation,
batch_sampler=validation_sampler,
num_workers=4)
else:
# use keyword for both background and evaluation classes
background_classes = 'background'
evaluation_classes = 'evaluation'
# Meta-training set
background = dataset_class(datadir,
split='all',
classes=background_classes,
transform=background_transform)
background_sampler = NShotTaskSampler(background, episodes_per_epoch,
n_train, k_train, q_train)
background_taskloader = DataLoader(background,
batch_sampler=background_sampler,
num_workers=4)
# Meta-test set
evaluation = dataset_class(datadir,
split='all',
classes=evaluation_classes,
transform=evaluation_transform)
# ADAPTED: in the original code, `episodes_per_epoch` was provided to
# `NShotTaskSampler` instead of `evaluation_episodes`.
evaluation_sampler = NShotTaskSampler(evaluation, evaluation_episodes,
n_test, k_test, q_test)
evaluation_taskloader = DataLoader(evaluation,
batch_sampler=evaluation_sampler,
num_workers=4)
#########
# Model #
#########
if torch.cuda.is_available():
if gpu is not None:
device = torch.device('cuda', gpu)
else:
device = torch.device('cuda')
torch.backends.cudnn.benchmark = True
else:
device = torch.device('cpu')
if not pretrained:
model = get_few_shot_encoder(num_input_channels)
# ADAPTED
model.to(device)
# BEFORE
# model.to(device, dtype=torch.double)
else:
assert torch.cuda.is_available()
model = models.__dict__[architecture](pretrained=True)
model.fc = Identity()
if gpu is not None:
model = model.cuda(gpu)
else:
model = model.cuda()
# TODO this is too risky: I'm not sure that this can work, since in
# the few-shot github repo the batch axis is actually split into
# support and query samples
# model = torch.nn.DataParallel(model).cuda()
def lr_schedule(epoch, lr):
# Drop lr every 2000 episodes
if epoch % drop_lr_every == 0:
return lr / 2
else:
return lr
############
# Training #
############
print(f'Training Prototypical network on {dataset}...')
optimiser = Adam(model.parameters(), lr=1e-3)
loss_fn = torch.nn.NLLLoss().to(device)
callbacks = [
# ADAPTED: this is the test monitoring now - and is only done at the
# end of training.
EvaluateFewShot(
eval_fn=proto_net_episode,
num_tasks=evaluation_episodes, # THIS IS NOT USED
n_shot=n_test,
k_way=k_test,
q_queries=q_test,
taskloader=evaluation_taskloader,
prepare_batch=prepare_nshot_task(n_test,
k_test,
q_test,
device=device),
distance=distance,
on_epoch_end=False,
on_train_end=True,
prefix='test_')
]
if monitor_validation:
callbacks.append(
# ADAPTED: this is the validation monitoring now - computed
# after every epoch.
EvaluateFewShot(
eval_fn=proto_net_episode,
num_tasks=evaluation_episodes, # THIS IS NOT USED
n_shot=n_test,
k_way=k_test,
q_queries=q_test,
# BEFORE taskloader=evaluation_taskloader,
taskloader=validation_taskloader, # ADAPTED
prepare_batch=prepare_nshot_task(n_test,
k_test,
q_test,
device=device),
distance=distance,
on_epoch_end=True, # ADAPTED
on_train_end=False, # ADAPTED
prefix='val_'))
callbacks.extend([
ModelCheckpoint(
filepath=PATH + f'/models/proto_nets/{param_str}.pth',
monitor=f'val_{n_test}-shot_{k_test}-way_acc',
verbose=1, # ADAPTED
save_best_only=monitor_validation # ADAPTED
),
LearningRateScheduler(schedule=lr_schedule),
CSVLogger(PATH + f'/logs/proto_nets/{param_str}.csv'),
])
fit(
model,
optimiser,
loss_fn,
epochs=n_epochs,
dataloader=background_taskloader,
prepare_batch=prepare_nshot_task(n_train,
k_train,
q_train,
device=device),
callbacks=callbacks,
metrics=['categorical_accuracy'],
fit_function=proto_net_episode,
fit_function_kwargs={
'n_shot': n_train,
'k_way': k_train,
'q_queries': q_train,
'train': True,
'distance': distance
},
)
def run():
episodes_per_epoch = 600
'''
###### LearningRateScheduler ######
drop_lr_every = 20
def lr_schedule(epoch, lr):
# Drop lr every 2000 episodes
if epoch % drop_lr_every == 0:
return lr / 2
else:
return lr
# callbacks add: LearningRateScheduler(schedule=lr_schedule)
'''
if args.dataset == 'miniImageNet':
n_epochs = 500
dataset_class = MiniImageNet
num_input_channels = 3
else:
raise (ValueError('need to make other datasets module'))
param_str = f'{args.dataset}_nt={args.n_train}_kt={args.k_train}_qt={args.q_train}_' \
f'nv={args.n_test}_kv={args.k_test}_qv={args.q_test}_' \
f'dist={args.distance}_sampling_method={args.sampling_method}_is_diverisity={args.is_diversity}'
print(param_str)
#########
# Model #
#########
model = get_few_shot_encoder(num_input_channels)
model.to(device, dtype=torch.double)
###################
# Create datasets #
###################
train_dataset = dataset_class('train')
eval_dataset = dataset_class('eval')
# Original sampling
if not args.sampling_method:
train_dataset_taskloader = DataLoader(
train_dataset,
batch_sampler=NShotTaskSampler(train_dataset, episodes_per_epoch,
args.n_train, args.k_train,
args.q_train),
num_workers=4)
eval_dataset_taskloader = DataLoader(
eval_dataset,
batch_sampler=NShotTaskSampler(eval_dataset, episodes_per_epoch,
args.n_test, args.k_test,
args.q_test),
num_workers=4)
# Importance sampling
else:
# ImportanceSampler: Latent space of model
train_dataset_taskloader = DataLoader(
train_dataset,
batch_sampler=ImportanceSampler(
train_dataset, model, episodes_per_epoch, n_epochs,
args.n_train, args.k_train, args.q_train,
args.num_s_candidates, args.init_temperature,
args.is_diversity),
num_workers=4)
eval_dataset_taskloader = DataLoader(
eval_dataset,
batch_sampler=NShotTaskSampler(eval_dataset, episodes_per_epoch,
args.n_test, args.k_test,
args.q_test),
num_workers=4)
############
# Training #
############
print(f'Training Prototypical network on {args.dataset}...')
optimiser = Adam(model.parameters(), lr=1e-3)
loss_fn = torch.nn.NLLLoss().cuda()
callbacks = [
EvaluateFewShot(eval_fn=proto_net_episode,
n_shot=args.n_test,
k_way=args.k_test,
q_queries=args.q_test,
taskloader=eval_dataset_taskloader,
prepare_batch=prepare_nshot_task(
args.n_test, args.k_test, args.q_test),
distance=args.distance),
ModelCheckpoint(
filepath=PATH + f'/models/proto_nets/{param_str}.pth',
monitor=f'val_{args.n_test}-shot_{args.k_test}-way_acc',
save_best_only=True,
),
ReduceLROnPlateau(
patience=40,
factor=0.5,
monitor=f'val_{args.n_test}-shot_{args.k_test}-way_acc'),
CSVLogger(PATH + f'/logs/proto_nets/{param_str}.csv'),
]
fit(
model,
optimiser,
loss_fn,
epochs=n_epochs,
dataloader=train_dataset_taskloader,
prepare_batch=prepare_nshot_task(args.n_train, args.k_train,
args.q_train),
callbacks=callbacks,
metrics=['categorical_accuracy'],
fit_function=proto_net_episode,
fit_function_kwargs={
'n_shot': args.n_train,
'k_way': args.k_train,
'q_queries': args.q_train,
'train': True,
'distance': args.distance
},
)
def evaluate_few_shot(state_dict,
n_shot,
k_way,
q_queries,
device,
architecture='resnet18',
pretrained=False,
small_dataset=False,
metric_name=None,
evaluation_episodes=1000,
num_input_channels=3,
distance='l2'):
if not pretrained:
model = get_few_shot_encoder(num_input_channels)
model.load_state_dict(state_dict)
else:
# assert torch.cuda.is_available()
model = models.__dict__[architecture](pretrained=True)
model.fc = Identity()
model.load_state_dict(state_dict)
dataset_class = FashionProductImagesSmall if small_dataset \
else FashionProductImages
# Meta-test set
resize = (80, 60) if small_dataset else (400, 300)
evaluation_transform = transforms.Compose([
transforms.Resize(resize),
transforms.ToTensor(),
])
evaluation = FashionProductImagesSmall(DATA_PATH,
split='all',
classes='evaluation',
transform=evaluation_transform)
sampler = NShotTaskSampler(evaluation, evaluation_episodes, n_shot, k_way,
q_queries)
taskloader = DataLoader(evaluation, batch_sampler=sampler, num_workers=4)
prepare_batch = prepare_nshot_task(n_shot, k_way, q_queries)
if metric_name is None:
metric_name = f'test_{n_shot}-shot_{k_way}-way_acc'
seen = 0
totals = {'loss': 0, metric_name: 0}
optimiser = torch.optim.Adam(model.parameters(), lr=1e-3)
loss_fn = torch.nn.NLLLoss().to(device)
for batch_index, batch in enumerate(taskloader):
x, y = prepare_batch(batch)
loss, y_pred = proto_net_episode(model,
optimiser,
loss_fn,
x,
y,
n_shot=n_shot,
k_way=k_way,
q_queries=q_queries,
train=False,
distance=distance)
seen += y_pred.shape[0]
totals['loss'] += loss.item() * y_pred.shape[0]
totals[metric_name] += categorical_accuracy(y, y_pred) * \
y_pred.shape[0]
totals['loss'] = totals['loss'] / seen
totals[metric_name] = totals[metric_name] / seen
return totals
