)
net = Net(
video_dim=args.feature_dim,
embd_dim=args.embd_dim,
we_dim=args.we_dim,
n_pair=args.n_pair,
max_words=args.max_words,
sentence_dim=args.sentence_dim,
)
net.train()
# Optimizers + Loss
loss_op = MaxMarginRankingLoss(
margin=args.margin,
negative_weighting=args.negative_weighting,
batch_size=args.batch_size,
n_pair=args.n_pair,
hard_negative_rate=args.hard_negative_rate,
)
net.cuda()
loss_op.cuda()
if args.pretrain_path != '':
net.load_checkpoint(args.pretrain_path)
optimizer = optim.Adam(net.parameters(), lr=args.lr)
if args.verbose:
print('Starting training loop ...')
def splits(sources, activities, train_persons, test_persons):
# TODO: Hyper-param tuning code here. Ensure this block is before utils.print_hyperparams() fn call TODO: in
# hyperparameter tuning code, make sure you change val of args.X so that correct values printed in tensorboard
# with print_hyperparams fn
train_dataset = BF(os.path.join(visual_feat_path, "train"),
os.path.join(text_path, "train"),
map_path,
sources,
activities,
train_persons,
rm_SIL=True)
train_dataloader = DataLoader(train_dataset,
batch_size=args.train_batch_size,
shuffle=True,
num_workers=0)
test_dataset = BF(os.path.join(visual_feat_path, "test"),
os.path.join(text_path, "test"),
map_path,
sources,
activities,
test_persons,
rm_SIL=True)
test_dataloader = DataLoader(test_dataset,
batch_size=args.test_batch_size,
shuffle=False,
num_workers=0)
# w2v of numbers or text?
train_uniq_w2v = train_dataset.labels_uniq["labels_num_w2v"]
test_uniq_w2v = test_dataset.labels_uniq["labels_num_w2v"]
print("Dataset Loaded")
train_dataset_size = len(train_dataset)
print("Size of training", train_dataset_size)
test_dataset_size = len(test_dataset)
print("Size of testing", len(test_dataset))
log_dir = str(int(time.time()))
writer = SummaryWriter(os.path.join(log_path, log_dir))
utils.print_hyperparams_tb(writer, args, train_dataset_size,
test_dataset_size)
# Setup
w2v_dim = 200
mod_dim = train_dataset.mod_dim
assert train_dataset.mod_dim == test_dataset.mod_dim
net = model.Net(mod_dim, w2v_dim, args.latent_feat_dim, device).to(device)
max_margin = MaxMarginRankingLoss(margin=args.margin).to(device)
if args.optimizer == 'adam':
optimizer = optim.Adam(net.parameters(), lr=args.lr)
elif args.optimizer == 'sgd':
optimizer = optim.SGD(net.parameters(),
lr=args.lr,
momentum=args.momentum)
# TODO: define scheduler here
# learning rate scheduler
best_train_accu, best_test_accu, best_epoch = -np.inf, -np.inf, -np.inf
for epoch in range(args.epochs):
train_accu = train(net, optimizer, max_margin, train_dataloader,
sources, train_uniq_w2v, train_dataset_size, epoch,
writer, train_dataset.stoi_map,
train_dataset.itos_map)
# print(list(net.parameters()))
test_accu = test(net, test_dataloader, sources, test_uniq_w2v,
test_dataset_size, epoch, writer,
test_dataset.stoi_map, test_dataset.itos_map)
if (test_accu - best_test_accu > 0.05) and (epoch < 155):
best_train_accu, best_test_accu, best_epoch = train_accu, test_accu, epoch + 1
print("Best Train Accuracy: %.7f Test Accuracy: %.7f (Epoch: %d)" %
(best_train_accu, best_test_accu, best_epoch))
face_ind_test = np.load(os.path.join(root_feat, 'no_face_ind_retrieval.npy'))
face_ind_test = 1 - face_ind_test
print('Done.')
# Model
video_modality_dim = {'face': (128, 128), 'audio': (128 * 16, 128),
'visual': (2048, 2048), 'motion': (1024, 1024)}
net = Net(video_modality_dim, 300,
audio_cluster=16, text_cluster=args.text_cluster_size)
net.train()
if args.GPU:
net.cuda()
# Optimizers + Loss
max_margin = MaxMarginRankingLoss(margin=args.margin)
if args.optimizer == 'adam':
optimizer = optim.Adam(net.parameters(), lr=args.lr)
elif args.optimizer == 'sgd':
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=args.momentum)
if args.GPU:
max_margin.cuda()
n_display = args.n_display
dataset_size = len(dataset)
lr_decay = args.lr_decay
print('Starting training loop ...')
face_ind_test = np.load(os.path.join(root_feat,'no_face_ind_retrieval.npy'))
face_ind_test = 1 - face_ind_test
print 'Done.'
# Model
video_modality_dim = {'face': (128,128), 'audio': (128*16,128),
'visual': (2048,2048), 'motion': (1024,1024)}
net = Net(video_modality_dim,300,
audio_cluster=16,text_cluster=args.text_cluster_size)
net.train()
if args.GPU:
net.cuda()
# Optimizers + Loss
max_margin = MaxMarginRankingLoss(margin=args.margin)
if args.optimizer == 'adam':
optimizer = optim.Adam(net.parameters(), lr=args.lr)
elif args.optimizer == 'sgd':
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=args.momentum)
if args.GPU:
max_margin.cuda()
n_display = args.n_display
dataset_size = len(dataset)
lr_decay = args.lr_decay
print 'Starting training loop ...'