def main():
# Settings
device = int(sys.argv[1]) if len(sys.argv) > 1 else None
batch_size = 128
n_l_train_data = 100
n_train_data = 60000
n_cls = 10
learning_rate = 1. * 1e-3
n_epoch = 50
act = F.relu
iter_epoch = n_train_data / batch_size
n_iter = n_epoch * iter_epoch
# Separate dataset
home = os.environ.get("HOME")
fpath = os.path.join(home, "datasets/mnist/train.npz")
separator = Separator(n_l_train_data)
separator.separate_then_save(fpath)
l_train_path = os.path.join(home, "datasets/mnist/l_train.npz")
u_train_path = os.path.join(home, "datasets/mnist/train.npz")
test_path = os.path.join(home, "datasets/mnist/test.npz")
# DataReader, Model, Optimizer, Losses
data_reader = MNISTDataReader(l_train_path, u_train_path, test_path,
batch_size=batch_size,
n_cls=n_cls,
da=True,
shape=True)
exp = Experiment014(
device,
learning_rate,
act,
)
# Training loop
print("# Training loop")
epoch = 1
st = time.time()
for i in range(n_iter):
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_l_train_batch()]
x_u, _ = [Variable(to_device(x, device)) \
for x in data_reader.get_u_train_batch()]
# Train
exp.train(x_l, y_l, x_u)
# Eval
if (i+1) % iter_epoch == 0:
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_test_batch()]
accs = [float(acc.data) for acc in exp.test(x_l, y_l)]
msg = "Epoch:{},ElapsedTime:{},Acc:{}".format(
epoch,
time.time() - st,
"|".join(map(str, accs)))
print(msg)
st = time.time()
epoch +=1
def main():
# Settings
device = int(sys.argv[1]) if len(sys.argv) > 1 else None
batch_size = 128
n_l_train_data = 100
n_train_data = 60000
n_cls = 10
learning_rate = 1. * 1e-3
n_epoch = 50
act = F.relu
iter_epoch = n_train_data / batch_size
n_iter = n_epoch * iter_epoch
# Separate dataset
home = os.environ.get("HOME")
fpath = os.path.join(home, "datasets/mnist/train.npz")
separator = Separator(n_l_train_data)
separator.separate_then_save(fpath)
l_train_path = os.path.join(home, "datasets/mnist/l_train.npz")
u_train_path = os.path.join(home, "datasets/mnist/train.npz")
test_path = os.path.join(home, "datasets/mnist/test.npz")
# DataReader, Model, Optimizer, Losses
data_reader = MNISTDataReader(l_train_path, u_train_path, test_path,
batch_size=batch_size,
n_cls=n_cls,
da=True,
shape=True)
exp = Experiment008(
device,
learning_rate,
act,
)
# Training loop
print("# Training loop")
epoch = 1
st = time.time()
acc_prev = 0.0
for i in range(n_iter):
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_l_train_batch()]
x_u, _ = [Variable(to_device(x, device)) \
for x in data_reader.get_u_train_batch()]
# Train
exp.train(x_l, y_l, x_u)
# Eval
if (i+1) % iter_epoch == 0:
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_test_batch()]
bs = 100
accs = []
for i in xrange(x_l.shape[0] / bs ):
accs_ = []
s = slice(i*bs, (i+1)*bs)
acc_for_each_y = exp.test(x_l[s, ], y_l[s, ])
for acc_for_each_y_ in acc_for_each_y:
accs_.append(float(acc_for_each_y_.data))
accs.append(accs_)
acc_mean = np.mean(accs, axis=0)
msg = "Epoch:{},ElapsedTime:{},Acc:{}".format(
epoch,
time.time() - st,
"|".join(map(str, acc_mean)))
print(msg)
if acc_mean[-1] < acc_prev:
exp.optimizer.alpha = 0.1 * exp.optimizer.alpha
print("Learning rate decays")
acc_prev = acc_mean[-1]
st = time.time()
epoch +=1
def main():
# Settings
device = int(sys.argv[1]) if len(sys.argv) > 1 else None
batch_size = 128
n_l_train_data = 4000
n_train_data = 50000
n_cls = 10
iter_epoch_labeled = n_l_train_data / batch_size
learning_rate = 1. * 1e-3
n_epoch = 50
act = F.relu
iter_epoch = n_train_data / batch_size
n_iter = n_epoch * iter_epoch
# Separate dataset
home = os.environ.get("HOME")
fpath = os.path.join(home, "datasets/cifar10/cifar-10.npz")
separator = Separator(n_l_train_data)
separator.separate_then_save(fpath)
l_train_path = os.path.join(home, "datasets/cifar10/l_cifar-10.npz")
u_train_path = os.path.join(home, "datasets/cifar10/cifar-10.npz")
test_path = os.path.join(home, "datasets/cifar10/cifar-10.npz")
# DataReader, Model, Optimizer, Losses
data_reader = Cifar10DataReader(l_train_path, u_train_path, test_path,
batch_size=batch_size,
n_cls=n_cls,
da=True,
shape=True)
exp = Experiment010(
device,
learning_rate,
act,
)
# Training loop
print("# Training loop")
epoch = 1
st = time.time()
acc_prev = 0.
for i in range(n_iter):
# Train with labeled samples
for j in range(iter_epoch_labeled):
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_l_train_batch()]
exp.train_with_labeled(x_l, y_l)
# Train with unlabeled samples
x_u, _ = [Variable(to_device(x, device)) \
for x in data_reader.get_u_train_batch()]
exp.train_with_unlabeled(x_u)
# Eval
if (i+1) % iter_epoch == 0:
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_test_batch()]
bs = 100
accs = []
for i in range(0, x_l.shape[0], bs):
acc = exp.test(x_l[i:i+bs, ], y_l[i:i+bs, ])
accs.append(cuda.to_cpu(acc.data))
acc_mean = np.mean(accs)
msg = "Epoch:{},ElapsedTime:{},Acc:{}".format(
epoch,
time.time() - st,
acc_mean)
print(msg)
if acc_prev > acc_mean:
exp.lambda_ *= 0.5
print("lambda decay")
acc_prev = acc_mean
st = time.time()
epoch +=1
def main():
# Settings
device = int(sys.argv[1]) if len(sys.argv) > 1 else None
batch_size = 128
n_l_train_data = 100
n_train_data = 60000
n_cls = 10
learning_rate = 1. * 1e-3
n_epoch = 50
act = F.relu
iter_epoch = n_train_data / batch_size
n_iter = n_epoch * iter_epoch
# Separate dataset
home = os.environ.get("HOME")
fpath = os.path.join(home, "datasets/mnist/train.npz")
separator = Separator(n_l_train_data)
separator.separate_then_save(fpath)
l_train_path = os.path.join(home, "datasets/mnist/l_train.npz")
u_train_path = os.path.join(home, "datasets/mnist/train.npz")
test_path = os.path.join(home, "datasets/mnist/test.npz")
# DataReader, Model, Optimizer, Losses
data_reader = MNISTDataReader(l_train_path,
u_train_path,
test_path,
batch_size=batch_size,
n_cls=n_cls,
da=True,
shape=True)
exp = Experiment014(
device,
learning_rate,
act,
)
# Training loop
print("# Training loop")
epoch = 1
st = time.time()
for i in range(n_iter):
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_l_train_batch()]
x_u, _ = [Variable(to_device(x, device)) \
for x in data_reader.get_u_train_batch()]
# Train
exp.train(x_l, y_l, x_u)
# Eval
if (i + 1) % iter_epoch == 0:
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_test_batch()]
accs = [float(acc.data) for acc in exp.test(x_l, y_l)]
msg = "Epoch:{},ElapsedTime:{},Acc:{}".format(
epoch,
time.time() - st, "|".join(map(str, accs)))
print(msg)
st = time.time()
epoch += 1
def main():
# Settings
device = int(sys.argv[1]) if len(sys.argv) > 1 else None
batch_size = 128
n_l_train_data = 1000
n_train_data = 73257
n_cls = 10
learning_rate = 1. * 1e-3
n_epoch = 50
act = F.relu
iter_epoch = n_train_data / batch_size
n_iter = n_epoch * iter_epoch
# Separate dataset
home = os.environ.get("HOME")
fpath = os.path.join(home, "datasets/svhn/train.mat")
separator = Separator(n_l_train_data)
separator.separate_then_save(fpath)
l_train_path = os.path.join(home, "datasets/svhn/l_train.mat")
u_train_path = os.path.join(home, "datasets/svhn/u_train.mat")
test_path = os.path.join(home, "datasets/svhn/test.mat")
# DataReader, Model, Optimizer, Losses
data_reader = SVHNDataReader(l_train_path,
u_train_path,
test_path,
batch_size=batch_size,
n_cls=n_cls,
da=True,
shape=True)
exp = Experiment002(
device,
learning_rate,
act,
)
# Training loop
print("# Training loop")
epoch = 1
st = time.time()
acc_prev = 0.
for i in range(n_iter):
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_l_train_batch()]
x_u, _ = [Variable(to_device(x, device)) \
for x in data_reader.get_u_train_batch()]
# Train
exp.train(x_l, y_l, x_u)
# Eval
if (i + 1) % iter_epoch == 0:
# Get data
x_l, y_l = [Variable(to_device(x, device)) \
for x in data_reader.get_test_batch()]
bs = 100
accs = []
for i in range(0, x_l.shape[0], bs):
acc = exp.test(x_l[i:i + bs, ], y_l[i:i + bs, ])
accs.append(cuda.to_cpu(acc.data))
acc_mean = np.mean(accs)
msg = "Epoch:{},ElapsedTime:{},Acc:{}".format(
epoch,
time.time() - st, acc_mean)
print(msg)
if acc_prev > acc_mean:
exp.lambda_ *= 0.5
print("lambda decay")
acc_prev = acc_mean
st = time.time()
epoch += 1
