train_set = torchvision.datasets.MNIST(root=data_folder,
train=True,
download=False,
transform=transform)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=batch_size,
shuffle=True,
num_workers=0)
# testset = torchvision.datasets.MNIST(root=data_folder, train=False,
# download=True, transform=transform)
# testloader = torch.utils.data.DataLoader(testset, batch_size=4,
# shuffle=False, num_workers=2)
featsModel = featuresModel(pretrained=pretrained)
distModel = distanceModel(pretrained=pretrained)
if trainstep <= 1:
if not pretrained:
model_weights_init(featsModel)
model_weights_init(distModel)
else:
featsModelfilename = '%s/featsModel%s_Iter%i.state' % (model_folder, name,
trainstep - 1)
distModelfilename = '%s/distModel%s_Iter%i.state' % (model_folder, name,
trainstep - 1)
featsModelfile = torch.load(featsModelfilename)
distModelfile = torch.load(distModelfilename)
featsModel.load_state_dict(featsModelfile)
distModel.load_state_dict(distModelfile)
sys.path.insert(0, '../../trainModels')
import lenet
trainstep = 1
delta = 100
lamda = 1
batch_size = 500
learningRate = 1e-3
modelName = "LearnDistanceNoPretrainDistAlexNetAugmentationDelta%iLamda%iBatch%iLR%f" % (
delta, lamda, batch_size, learningRate)
modelfolder = "trainedModels"
modelfilename = '%s/featsModel%s_Iter%i' % (modelfolder, modelName, trainstep)
modelfile = torch.load(modelfilename + ".state")
model = featuresModel()
model.load_state_dict(modelfile)
Nsamples = 1000
Niter = 1
transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))])
if torch.cuda.is_available():
datafolder = "/var/tmp/ioannis/data"
else:
datafolder = "../../data"
trainset = torchvision.datasets.MNIST(root=datafolder,
train_set = torchvision.datasets.MNIST(root=data_folder,
train=True,
download=False,
transform=transform)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=batch_size,
shuffle=True,
num_workers=0)
# testset = torchvision.datasets.MNIST(root=data_folder, train=False,
# download=True, transform=transform)
# testloader = torch.utils.data.DataLoader(testset, batch_size=4,
# shuffle=False, num_workers=2)
featsModel = featuresModel(pretrained=True)
distModel = distanceModel(pretrained=False)
if trainstep <= 1:
if not pretrained:
model_weights_init(featsModel)
model_weights_init(distModel)
else:
featsModelfilename = '%s/featsModel%s_Iter%i.state' % (model_folder, name,
trainstep - 1)
distModelfilename = '%s/distModel%s_Iter%i.state' % (model_folder, name,
trainstep - 1)
featsModelfile = torch.load(featsModelfilename)
distModelfile = torch.load(distModelfilename)
featsModel.load_state_dict(featsModelfile)
distModel.load_state_dict(distModelfile)