def load_data(mode, train=True):
if mode == 's2m':
return get_mnist(train), get_usps(train)
elif mode == 'u2m':
return get_mnist(train), get_usps(train)
elif mode == 'm2u':
return get_mnist(train), get_usps(train)
def run(dtrain, dtest, epochs=10, verbose=False):
args = AttrDict(**dict(
save_dir="_models",
iters=epochs,
epochs=epochs,
bootstrap_epochs=1,
ncams=1,
verbose=verbose
))
mnist = Collection('multiinput_edge_dropout_mpcc_{}'.format(args.ncams), args.save_dir, nepochs=args.epochs, verbose=args.verbose)
ncams = args.ncams
mnist.set_model_family(MultiInputEdgeDropoutFamily, ninputs=ncams, resume=False,
merge_function="max_pool_concat", drop_comm_train=dtrain,
drop_comm_test=dtest, input_dims=1, output_dims=10)
train, test = get_mnist()
mnist.add_trainset(train)
mnist.add_testset(test)
mnist.set_searchspace(
nfilters_embeded=[3],
nlayers_embeded=[2],
nfilters_cloud=[3],
nlayers_cloud=[2],
lr=[1e-3],
branchweight=[.1],
ent_T=[100]
)
# currently optimize based on the validation accuracy of the main model
traces = mnist.train(niters=args.iters, bootstrap_nepochs=args.bootstrap_epochs)
return traces[-1]['y']
def get_data_loader(name):
'''Get data loader by name'''
if name == 'mnist':
return get_mnist()
elif name == 'celeba':
return get_celeba()
else:
assert False, '[*] dataset not implement!'
def get_data_loader(name, path, train=True):
"""Get data loader by name."""
if name == "MNIST":
return get_mnist(path, train)
elif name == "USPS":
return get_usps(path, train)
elif name == "SVHN":
return get_svhn(path, train)
def get_loader(name, split, batch_size=50):
if name == "mnist":
return get_mnist(split,batch_size)
if name == "usps":
return get_usps(split,batch_size)
if name == "mnistBig":
return mnistBig.get_mnist(split,batch_size)
if name == "uspsBig":
return uspsBig.get_usps(split,batch_size)
def get_data( name, start_date, end_date):
entitys = None
if name == 'mnist' or name == 'default':
nb_classes, input_shape, x_train, \
x_test, y_train, y_test = get_mnist()
entitys = Entitys(input_shape, nb_classes, x_train, y_train, x_test, y_test)
elif name == 'cifar10':
nb_classes, input_shape, x_train, \
x_test, y_train, y_test = get_cifar10()
entitys = Entitys(input_shape, nb_classes, x_train, y_train, x_test, y_test)
elif name == 'stock':
nb_classes, input_shape, x_train, \
x_test, y_train, y_test = get_stock(start_date, end_date)
entitys = Entitys(input_shape, nb_classes, x_train, y_train, x_test, y_test)
return entitys
def get_loader(name, split, batch_size=50):
if name == "mnist":
return get_mnist(split, batch_size)
elif name == "usps":
return get_usps(split, batch_size)
elif name == "mnistBig":
return mnistBig.get_mnist(split, batch_size)
elif name == "uspsBig":
return uspsBig.get_usps(split, batch_size)
elif name == "coxs2v":
return coxs2v.get_coxs2v(split, batch_size)
else:
raise Exception("Dataset name {} not supported".format(name))
def get_data_loader(name, train=True):
"""Get data loader by name."""
if name == "MNIST":
return get_mnist(train)
elif name == "USPS":
return get_usps(train)
def get_data_iter(name, train):
if name == 'MNIST':
return get_inf_iterator(mnist.get_mnist(train=True))
else:
return get_inf_iterator(usps.get_usps(train=True))
if __name__ == '__main__':
data_itr_tgt = get_data_iter("USPS", train=True)
image_tgt, label_tgt = next(data_itr_tgt)
image_tgt = image_tgt[0:2]
print(image_tgt.shape)
print(image_tgt)
new_tgt = make_larger_size(image_tgt)
print(new_tgt.shape)
print(new_tgt)
plt.imshow(new_tgt[0].numpy().reshape(36, 36), cmap="gray")
plt.show()
print(label_tgt[0])
exit(0)
#options = { 'dir' : 'data' , 'name' : 'MNIST' , 'batch_size' : 64 , 'dataset_mean' : (0.5,0.5,0.5) , 'dataset_std' : (0.5,0.5,0.5)}
mnist_loader = mnist.get_mnist(train=True)
usps_loader = usps.get_usps(train=True)
im = Image.open("snapshots/Figure_2.png")
im = im.convert("L")
#im = im.resize((image_width, image_height))
# im.show()
data = im.getdata()
data = np.matrix(data)
print(data.shape)
for tgt_img, tgt_label in usps_loader:
#print(type(tgt_img))
#print(tgt_img[0])
plt.imshow(tgt_img[0].numpy().reshape(28, 28), cmap="gray")
plt.show()
print(tgt_img[0].numpy())