def test_5_shot_5_way(self):
config = {
'data.dataset': 'mini-imagenet',
'data.dataset_path': 'data/mini-imagenet',
'data.split': 'vinyals',
'data.train_way': 5,
'data.batch': 10,
'data.train_support': 5,
'data.train_query': 5,
'data.test_way': 5,
'data.test_support': 5,
'data.test_query': 5,
'data.episodes': 1,
'data.cuda': cuda_on,
'data.gpu': 0,
'model.x_dim': '84,84,3',
'model.lstm_size': 32,
'model.save_dir': './miniimagenet_test',
'train.epochs': 1,
'train.optim_method': 'Adam',
'train.lr': 0.001,
'train.patience': 100,
'train.restore': 0,
'train.log_dir': 'tests/logs'
}
train(config)
def test_2_way_batch_4(self):
config = {
'data.dataset_path':
'/home/igor/dl/siamese-networks-tf/data/omniglot',
'data.dataset': 'omniglot',
'data.train_way': 2,
'data.test_way': 2,
'data.split': 'vinyals',
'data.batch': 4,
'data.episodes': 2,
'data.cuda': 1,
'data.gpu': gpu_num,
'train.epochs': 1,
'train.lr': 0.001,
'train.patience': 100,
'train.tb_dir': 'results/logs/gradient_tape/',
'train.log_dir': 'results/logs',
'train.restore': 0,
'model.x_dim': '105,105,1',
'model.save_dir': 'results/models/omniglot'
}
train(config)
config['train.restore'] = 1
train(config)
def test_1_shot_1_way(self):
config = {
"data.dataset": "omniglot",
"data.split": "vinyals",
"data.train_way": 1,
"data.train_support": 1,
"data.train_query": 1,
"data.test_way": 1,
"data.test_support": 1,
"data.test_query": 1,
"data.episodes": 10,
"data.cuda": cuda_on,
"data.gpu": 0,
"model.x_dim": "28,28,1",
"model.z_dim": 64,
"train.epochs": 2,
'train.optim_method': "Adam",
"train.lr": 0.001,
"train.patience": 5,
"model.save_path": 'test_omniglot.h5'
}
train(config)
os.remove('test_omniglot.h5')
num_workers=1, pin_memory=True, drop_last=True)
resume = args.resume
if args.resume:
assert(bool(args.exp_dir))
if args.reuse_old:
with open("%s/args.pkl" % args.exp_dir, "rb") as f:
args = pickle.load(f)
args.resume = resume
print(args)
# get number of features of input
n_features = train_loader.dataset[0][0].shape[1]
model = models.FinanceModel(input_dim=n_features, output_dim=output_dim,
dropout_p=args.dropout, binary=args.direction)
if not bool(args.exp_dir):
print("exp_dir not specified, automatically creating one...")
args.exp_dir = "exp/Data-%s/Optim-%s_LR-%s_Epochs-%s" % (
os.path.basename(args.data_file), args.optim, args.lr, args.n_epochs)
if not args.resume:
print("\nexp_dir: %s" % args.exp_dir)
os.makedirs("%s/models" % args.exp_dir)
with open("%s/args.pkl" % args.exp_dir, "wb") as f:
pickle.dump(args, f)
train(model, train_loader, val_loader, args)