def ae_build(_data, encoding, optimizer, loss, _type):
if (_data.name == 'mnist'):
train = _data.X_train
test = _data.x_test
else:
train = _data.X_train_std
test = _data.x_test_std
if (_type == 'linear'):
name = 'aelinear_' + _data.name
else:
name = 'aenon-linear_' + _data.name
ae = make_ae(train.shape[1], [encoding], encoding, [train.shape[1]],
[_type], [PAR.af_adec])
ae.autoencoder.compile(optimizer=optimizer, loss=loss)
hist = ae.autoencoder.fit(train,
train,
epochs=PAR.ae_epoch,
batch_size=PAR.ae_batch,
shuffle=True,
validation_data=(test, test))
plot_ae(name, hist)
# encoded_imgs = ae.encoder.predict(x_test)
# decoded_imgs = ae.decoder.predict(encoded_imgs)
reduced_train = ae.encoder.predict(train)
reduced_test = ae.encoder.predict(test)
ae_data = DS(reduced_train, _data.Y_train, reduced_test, _data.y_test,
name)
return ae_data, ae
def pca_linear(_data, n_components):
train = _data.X_train
test = _data.x_test
name = 'pca_' + _data.name
pca = PCA(n_components=n_components)
reduced_train = pca.fit_transform(train)
reduced_test = pca.transform(test)
pca_data = DS(reduced_train, _data.Y_train, reduced_test, _data.y_test,
name)
return pca_data
torch.backends.cudnn.benchmark = True
if not os.path.exists(args.save_dir):
os.makedirs('{:s}/ckpt'.format(args.save_dir))
writer = SummaryWriter()
size = (args.image_size, args.image_size)
train_tf = transforms.Compose([
transforms.Resize(size),
transforms.RandomHorizontalFlip(),
transforms.RandomGrayscale(),
transforms.ToTensor(),
])
train_set = DS(args.root, train_tf)
iterator_train = iter(data.DataLoader(
train_set,
batch_size=args.batch_size,
sampler=InfiniteSampler(len(train_set)),
num_workers=args.n_threads))
print(len(train_set))
g_model = InpaintNet().to(device)
fd_model = FeaturePatchDiscriminator().to(device)
pd_model = PatchDiscriminator().to(device)
l1 = nn.L1Loss().to(device)
cons = ConsistencyLoss().to(device)
start_iter = 0
g_optimizer = torch.optim.Adam(
device = torch.device('cuda')
if not os.path.exists(args.save_dir):
os.makedirs('{:s}/images'.format(args.save_dir))
os.makedirs('{:s}/ckpt'.format(args.save_dir))
#writer = SummaryWriter(logdir=args.log_dir)
size = (args.image_size, args.image_size)
img_tf = transforms.Compose([
transforms.Resize(size=size),
transforms.RandomHorizontalFlip(),
transforms.ToTensor()
])
dataset = DS(args.root, img_tf)
iterator_train = iter(
data.DataLoader(dataset,
batch_size=args.batch_size,
sampler=InfiniteSampler(len(dataset)),
num_workers=args.n_threads))
print(len(dataset))
model = DFNet().to(device)
lr = args.lr
start_iter = 0
optimizer = torch.optim.Adam(model.parameters(), lr=lr)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.1)
criterion = InpaintingLoss().to(device)
def read_data():
#Read and seperate the mnist dataset.
(X_train, Y_train), (x_test, y_test) = mnist.load_data()
#Flatten the data, MLP doesn't use the 2D structure of the data. 784 = 28*28
X_train = X_train.reshape(60000, 784)
x_test = x_test.reshape(10000, 784)
#Make the value floats in [0;1] instead of int in [0;255]
tr = X_train.astype('float32')
te = x_test.astype('float32')
tr = tr / 255
te = te / 255
trl = np.asarray(Y_train, dtype=np.int32)
tel = np.asarray(y_test, dtype=np.int32)
print(tr[1, :])
mnist_data = DS(tr, trl, te, tel, 'mnist')
print(tr)
print(trl)
#Read and seperate the HAR dataset.
tr1 = pd.read_csv(os.path.join(path, 'UCI HAR Dataset', 'train',
'X_train.txt'),
delim_whitespace=True,
error_bad_lines=False,
warn_bad_lines=False,
header=None)
tr1 = tr1.to_numpy(dtype='float32', copy=True)
trl1 = pd.read_csv(os.path.join(path, 'UCI HAR Dataset', 'train',
'y_train.txt'),
error_bad_lines=False,
warn_bad_lines=False,
header=None)
te1 = pd.read_csv(os.path.join(path, 'UCI HAR Dataset', 'test',
'X_test.txt'),
delim_whitespace=True,
error_bad_lines=False,
warn_bad_lines=False,
header=None)
te1 = te1.to_numpy(dtype='float32', copy=True)
tel1 = pd.read_csv(os.path.join(path, 'UCI HAR Dataset', 'test',
'y_test.txt'),
error_bad_lines=False,
warn_bad_lines=False,
header=None)
har_data = DS(tr1, trl1.values.T[0], te1, tel1.values.T[0], 'har')
#Read and seperate the pulsar dataset.
alldata = pd.read_csv(os.path.join(path, 'HTRU2', 'HTRU_2.csv'),
sep=',',
error_bad_lines=False,
warn_bad_lines=False,
header=None)
labels = pd.DataFrame(alldata[alldata.columns[8]])
alldata.drop(alldata.columns[8], axis=1, inplace=True)
tr2, te2, trl2, tel2 = train_test_split(alldata,
labels,
test_size=0.20,
random_state=42)
tr2 = tr2.to_numpy(dtype='float32', copy=True)
te2 = te2.to_numpy(dtype='float32', copy=True)
pulsar_data = DS(tr2, trl2.values.T[0], te2, tel2.values.T[0], 'pulsar')
return mnist_data, har_data, pulsar_data