parser.add_argument('--dcn_lr', type=float, default=0.01, metavar='N',
help='learning rate for training (default: 0.001)')
args = parser.parse_args()
log_dir = 'logs/dec-' + datasetname
if os.path.exists(log_dir) == False:
os.makedirs(log_dir)
for i in range(1, repeat+1):
sdae_savepath = ("model/sdae-run-"+datasetname+"-%d.pt" % i)
if os.path.exists(sdae_savepath)==False:
print("Experiment #%d" % i)
write_log("Experiment #%d" % i,log_dir)
train_loader=None
test_loader=None
if datasetname=='mnist':
train_loader = torch.utils.data.DataLoader(
MNIST('./dataset/mnist', train=True, download=True),
batch_size=batch_size, shuffle=True, num_workers=0)
# test_loader = torch.utils.data.DataLoader(
# MNIST('./dataset/mnist', train=False),
# batch_size=batch_size, shuffle=False, num_workers=0)
elif datasetname=='cifar':
transform = transforms.Compose(
[transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
trainset = datasets.CIFAR10(
root='./dataset/cifar', train=True, download=False, transform=transform) # download=True会通过官方渠道下载
train_loader = torch.utils.data.DataLoader(
trainset, batch_size=batch_size, shuffle=True, num_workers=2)
testset = datasets.CIFAR10(
root='./dataset/cifar', train=False, download=False, transform=transform)
test_loader = torch.utils.data.DataLoader(
testset, batch_size=batch_size, shuffle=False, num_workers=2)
help='input batch size for training (default: 256)')
parser.add_argument('--pretrainepochs',
type=int,
default=300,
metavar='N',
help='number of epochs to train (default: 300)')
parser.add_argument('--epochs',
type=int,
default=500,
metavar='N',
help='number of epochs to train (default: 500)')
args = parser.parse_args()
# Load data for pre-training
train_loader = torch.utils.data.DataLoader(MNIST('./dataset/mnist',
train=True,
download=True),
batch_size=args.batch_size,
shuffle=True,
num_workers=0)
test_loader = torch.utils.data.DataLoader(MNIST('./dataset/mnist',
train=False),
batch_size=args.batch_size,
shuffle=False,
num_workers=0)
sdae = StackedDAE(input_dim=784,
z_dim=10,
binary=False,
encodeLayer=[500, 500, 2000],
decodeLayer=[2000, 500, 500],
help='number of epochs to train (default: 200)')
parser.add_argument('--pretrain',
type=str,
default="../model/mnist_sdae_weights.pt",
metavar='N',
help='directory for pre-trained weights')
parser.add_argument('--data',
type=str,
default="MNIST",
metavar='N',
help='dataset(MNIST, Fashion)')
parser.add_argument('--use_pretrain', type=bool, default=True)
args = parser.parse_args()
# Load data
mnist_train = MNIST('./dataset/mnist', train=True, download=True)
mnist_test = MNIST('./dataset/mnist', train=False)
X = mnist_train.train_data
y = mnist_train.train_labels
test_X = mnist_test.test_data
test_y = mnist_test.test_labels
# Set parameters
ml_penalty, cl_penalty = 0.1, 1
if args.data == "Fashion":
fashionmnist_train = FashionMNIST('./dataset/fashion_mnist',
train=True,
download=True)
fashionmnist_test = FashionMNIST('./dataset/fashion_mnist',
train=False)
X = fashionmnist_train.train_data
LEARNING_RATE = 0.001
TRAIN_DIR = None
# LOG_DIR = 'data/summaries/mnist_slic_spatial'
LOG_DIR = 'data/summaries/mnist_quickshift_spatial'
SAVE_STEP = 250
AUGMENT_TRAIN_EXAMPLES = False
DROPOUT = 0.5
BATCH_SIZE = 64
MAX_STEPS = 15000
DISPLAY_STEP = 10
# FORM_FEATURES = SLIC_FEATURES
FORM_FEATURES = QUICKSHIFT_FEATURES
NUM_FEATURES = len(FORM_FEATURES) + 1
data = Data(DATA_DIR)
# segmentation_algorithm = slic_fixed(
# num_segments=100, compactness=5, max_iterations=10, sigma=0)
segmentation_algorithm = quickshift_fixed(ratio=1,
kernel_size=2,
max_dist=2,
sigma=0)
feature_extraction_algorithm = extract_features_fixed(FORM_FEATURES)
def preprocess_spatial_fixed(segmentation_algorithm,
feature_extraction_algorithm, node_size,
node_stride, delta, neighborhood_size,
connectivity):
from lib.models import UNet
from lib.datasets import MNIST
from lib.process import Trainer, Evaluator
import matplotlib.pyplot as plt
import torch
# CONSTANST
MODEL_PATH = './u-net-mnist.pth'
EPOCHS = 1
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
dataset = MNIST()
model = UNet(n_channels=1, n_classes=1)
model = model.to(device)
trainer = Trainer(model=model, dataset=dataset, batch_size=16, device=device)
trainer.load_model(model, MODEL_PATH)
evaluator = Evaluator(dataset=dataset, batch_size=16, device=device)
def train():
for _ in range(EPOCHS):
loss = trainer.train_epoch(progress_bar=False)
print('loss', loss)
with torch.no_grad():
score = evaluator.DCM(model=model)
print('DCM score:', score)
print('end of training')
trainer.save_model(MODEL_PATH)
def eval():
model.eval()
EPOCHS = args.epochs
MODEL_PATH = './{}-ds{}.pth'.format(args.net, args.dataset)
EPOCHS = args.epochs
BATCH = args.batch
DEEPVESSEL = False
if args.pre_transform:
if args.dataset.startswith('G'):
pre_transform = Crop(30, 150, 256, 256)
else:
pre_transform = CropVessel12(30, 150, 256, 256)
else:
pre_transform = None
if args.dataset == 'MNIST':
dataset = MNIST(background=args.background)
elif args.dataset == 'GMNIST':
dataset = GMNIST(background=args.background)
elif args.dataset == 'VESSEL12':
dataset = VESSEL12(data_dir=args.vesseldir, pre_transform=pre_transform)
elif args.dataset == 'GVESSEL12':
dataset = GVESSEL12(data_dir=args.vesseldir, pre_transform=pre_transform)
elif args.dataset == 'SVESSEL':
dataset = SVESSEL(data_dir=args.svesseldir)
elif args.dataset == 'GSVESSEL':
dataset = GSVESSEL(data_dir=args.svesseldir)
else:
dataset = MNIST()
if args.net == 'GFCN':
