def main():
# Training settings
parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
parser.add_argument('--batch-size',
type=int,
default=64,
metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test-batch-size',
type=int,
default=1000,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=3,
metavar='N',
help='number of epochs to train (default: 14)')
parser.add_argument('--lr',
type=float,
default=1,
metavar='LR',
help='learning rate (default: 1.0)')
parser.add_argument('--gamma',
type=float,
default=0.7,
metavar='M',
help='Learning rate step gamma (default: 0.7)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument('--dry-run',
action='store_true',
default=False,
help='quickly check a single pass')
parser.add_argument('--seed',
type=int,
default=1,
metavar='S',
help='random seed (default: 1)')
parser.add_argument(
'--log-interval',
type=int,
default=10,
metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--save-model',
action='store_true',
default=False,
help='For Saving the current Model')
parser.add_argument('--T',
type=int,
default=300,
metavar='N',
help='SNN time window')
parser.add_argument('--resume',
type=str,
default=None,
metavar='RESUME',
help='Resume model from checkpoint')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'batch_size': args.batch_size}
if use_cuda:
kwargs.update({
'num_workers': 1,
'pin_memory': True,
'shuffle': True
}, )
transform = transforms.Compose([
transforms.ToTensor(),
#transforms.Normalize((0.1307,), (0.3081,))
])
dataset1 = datasets.MNIST('../data',
train=True,
download=True,
transform=transform)
dataset2 = datasets.MNIST('../data', train=False, transform=transform)
snn_dataset = SpikeDataset(dataset2, T=args.T)
#print(type(dataset1[0][0]))
train_loader = torch.utils.data.DataLoader(dataset1, **kwargs)
test_loader = torch.utils.data.DataLoader(dataset2, **kwargs)
#print(test_loader[0])
snn_loader = torch.utils.data.DataLoader(snn_dataset, **kwargs)
model = Net().to(device)
snn_model = CatNet(args.T).to(device)
if args.resume != None:
load_model(torch.load(args.resume), model)
for param_tensor in snn_model.state_dict():
print(param_tensor, "\t", snn_model.state_dict()[param_tensor].size())
optimizer = optim.Adadelta(model.parameters(), lr=args.lr)
scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
for epoch in range(1, args.epochs + 1):
train(args, model, device, train_loader, optimizer, epoch)
test(model, device, test_loader)
scheduler.step()
fuse_module(model)
transfer_model(model, snn_model)
test(snn_model, device, snn_loader)
def main():
# Training settings
parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
parser.add_argument('--batch-size',
type=int,
default=2,
metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test-batch-size',
type=int,
default=1000,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=3,
metavar='N',
help='number of epochs to train (default: 14)')
parser.add_argument('--lr',
type=float,
default=1,
metavar='LR',
help='learning rate (default: 1.0)')
parser.add_argument('--gamma',
type=float,
default=0.7,
metavar='M',
help='Learning rate step gamma (default: 0.7)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument('--dry-run',
action='store_true',
default=False,
help='quickly check a single pass')
parser.add_argument('--seed',
type=int,
default=1,
metavar='S',
help='random seed (default: 1)')
parser.add_argument(
'--log-interval',
type=int,
default=10,
metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--save-model',
action='store_true',
default=False,
help='For Saving the current Model')
parser.add_argument('--T',
type=int,
default=160,
metavar='N',
help='SNN time window')
parser.add_argument('--resume',
type=str,
default=None,
metavar='RESUME',
help='Resume model from checkpoint')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
f = np.load('dataeeg/eeg1201_r57_b_hamming_5s_2_c.npz')
X_train_ = f['X_train_']
y_train_ = f['y_train_']
X_test = f['X_test']
y_test = f['y_test']
X_train_ = torch.FloatTensor(X_train_)
y_train_ = torch.FloatTensor(y_train_)
X_test = torch.FloatTensor(X_test)
y_test = torch.FloatTensor(y_test)
X_train_ = X_train_ * 3000 + (0.001) * torch.randn(
len(X_train_), len(X_train_[0]), len(X_train_[0][0]),
len(X_train_[0][0][0]))
X_train_ = np.clip(X_train_, 0, 1)
X_test = X_test * 3000
X_test = np.clip(X_test, 0, 1)
for i in range(4):
X_train_ = torch.cat([X_train_, X_train_], axis=0)
y_train_ = torch.cat([y_train_, y_train_], axis=0)
torch_dataset_train = torch.utils.data.TensorDataset(X_train_, y_train_)
torch_dataset_test = torch.utils.data.TensorDataset(X_test, y_test)
snn_dataset = SpikeDataset(torch_dataset_test, T=args.T)
train_loader = torch.utils.data.DataLoader(torch_dataset_train,
shuffle=True,
batch_size=512)
test_loader = torch.utils.data.DataLoader(torch_dataset_test,
shuffle=False,
batch_size=64)
snn_loader = torch.utils.data.DataLoader(snn_dataset,
shuffle=False,
batch_size=16)
model = Net().to(device)
snn_model = CatNet(args.T).to(device)
if args.resume != None:
load_model(torch.load(args.resume), model)
optimizer = optim.Adadelta(model.parameters(), lr=args.lr)
scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
Acc = 0
for epoch in range(1, args.epochs + 1):
train(args, model, device, train_loader, optimizer, epoch)
test(model, device, train_loader)
Acc_ = test(model, device, test_loader)
if Acc_ > Acc:
Acc = Acc_
fuse_module(model)
#torch.save(model.state_dict(), "eeg_1201_3_layers_final_1.pt")
scheduler.step()
print(Acc)
fuse_module(model)
test(model, device, test_loader)
transfer_model(model, snn_model)
test(snn_model, device, snn_loader)
def main():
# Training settings
parser = argparse.ArgumentParser(description='PyTorch Cifar10 LeNet Example')
parser.add_argument('--batch-size', type=int, default=64, metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test-batch-size', type=int, default=1000, metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs', type=int, default=14, metavar='N',
help='number of epochs to train (default: 14)')
parser.add_argument('--lr', type=float, default=1e-5, metavar='LR',
help='learning rate (default: 1)')
parser.add_argument('--gamma', type=float, default=0.7, metavar='M',
help='Learning rate step gamma (default: 0.7)')
parser.add_argument('--no-cuda', action='store_true', default=False,
help='disables CUDA training')
parser.add_argument('--dry-run', action='store_true', default=False,
help='quickly check a single pass')
parser.add_argument('--seed', type=int, default=1, metavar='S',
help='random seed (default: 1)')
parser.add_argument('--log-interval', type=int, default=10, metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--save-model', action='store_true', default=False,
help='For Saving the current Model')
parser.add_argument('--resume', type=str, default=None, metavar='RESUME',
help='Resume model from checkpoint')
parser.add_argument('--T', type=int, default=60, metavar='N',
help='SNN time window')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'batch_size': args.batch_size}
if use_cuda:
kwargs.update({'num_workers': 1,
'pin_memory': True,
'shuffle': True},
)
mean = [0.4913997551666284, 0.48215855929893703, 0.4465309133731618]
std = [0.24703225141799082, 0.24348516474564, 0.26158783926049628]
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=6),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
AddGaussianNoise(std=0.01)
])
im_aug = transforms.Compose([
#transforms.ColorJitter(brightness=0.5, contrast=0.5, hue=0.5),
transforms.RandomRotation(10),
transforms.RandomCrop(32, padding = 6),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
AddGaussianNoise(std=0.01)
])
transform_test = transforms.Compose([
transforms.ToTensor()
#transforms.Normalize(mean, std)
])
trainset = datasets.CIFAR10(
root='./data', train=True, download=True, transform=transform_train)
for i in range(100):
trainset = trainset + datasets.CIFAR10(root='./data', train=True, download=True, transform=im_aug)
train_loader = torch.utils.data.DataLoader(
trainset, batch_size=128, shuffle=True)
testset = datasets.CIFAR10(
root='./data', train=False, download=True, transform=transform_test)
test_loader = torch.utils.data.DataLoader(
testset, batch_size=100, shuffle=False)
snn_dataset = SpikeDataset(testset, T = args.T)
snn_loader = torch.utils.data.DataLoader(snn_dataset, batch_size=10, shuffle=False)
from models.vgg import VGG, CatVGG
model = VGG('VGG19', clamp_max=1, quantize_bit=32).to(device)
snn_model = CatVGG('VGG19', args.T).to(device)
if args.resume != None:
model.load_state_dict(torch.load(args.resume), strict=False)
load_model(torch.load(args.resume), model)
load_model(torch.load(args.resume), snn_model)
optimizer = optim.Adam(model.parameters(), lr=args.lr)
scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
for epoch in range(1, args.epochs + 1):
train(args, model, device, train_loader, optimizer, epoch)
#test(model, device, train_loader)
test(model, device, test_loader)
#transfer_model(model, snn_model)
#test(snn_model, device, snn_loader)
if args.save_model:
torch.save(model.state_dict(), "cifar_cnn_19.pt")
scheduler.step()
#test(model, device, train_loader)
test(model, device, test_loader)
transfer_model(model, snn_model)
with torch.no_grad():
normalize_weight(snn_model.features, quantize_bit=8)
test(snn_model, device, snn_loader)
if args.save_model:
torch.save(model.state_dict(), "cifar_cnn_19.pt")
def main():
# Training settings
parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
parser.add_argument('--batch-size',
type=int,
default=64,
metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test-batch-size',
type=int,
default=1000,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=14,
metavar='N',
help='number of epochs to train (default: 14)')
parser.add_argument('--lr',
type=float,
default=0.001,
metavar='LR',
help='learning rate (default: 1.0)')
parser.add_argument('--gamma',
type=float,
default=0.7,
metavar='M',
help='Learning rate step gamma (default: 0.7)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument('--dry-run',
action='store_true',
default=False,
help='quickly check a single pass')
parser.add_argument('--seed',
type=int,
default=3,
metavar='S',
help='random seed (default: 1)')
parser.add_argument(
'--log-interval',
type=int,
default=10,
metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--save-model',
action='store_true',
default=False,
help='For Saving the current Model')
parser.add_argument('--resume',
type=str,
default=None,
metavar='RESUME',
help='Resume model from checkpoint')
parser.add_argument('--T',
type=int,
default=40,
metavar='N',
help='SNN time window')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
#torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'batch_size': args.batch_size}
if use_cuda:
kwargs.update({
'num_workers': 1,
'pin_memory': True,
'shuffle': True
}, )
path = './ecg_data_normalize_smoke_2c.npz'
#path = './all_0810_paper_smoke_normalize_01.npz'
f = np.load(path)
train_x, train_y = f['x_train'], f['y_train']
test_x, test_y = f['x_test'], f['y_test']
"""
for i in range(len(train_x)):
train_x[i] = minmaxscaler (train_x[i])
for i in range(len(test_x)):
test_x[i] = minmaxscaler (test_x[i])
"""
y_test_ = test_y
X_train_ = torch.FloatTensor(train_x)
print(X_train_.shape)
X_train_ = X_train_.reshape(-1, 1, 180, 1)
X_train_ = torch.clamp(X_train_, min=0, max=1)
X_train_ = torch.div(torch.ceil(torch.mul(X_train_, 8)), 8)
y_train_ = torch.FloatTensor(train_y)
X_test = torch.FloatTensor(test_x)
X_test = X_test.reshape(-1, 1, 180, 1)
X_test = torch.clamp(X_test, min=0, max=1)
X_test = torch.div(torch.ceil(torch.mul(X_test, 4)), 4)
y_test = torch.FloatTensor(test_y)
torch_dataset_train = torch.utils.data.TensorDataset(X_train_, y_train_)
torch_dataset_test = torch.utils.data.TensorDataset(X_test, y_test)
snn_dataset = SpikeDataset(torch_dataset_test, T=args.T)
train_loader = torch.utils.data.DataLoader(torch_dataset_train,
shuffle=True,
batch_size=256 * 3)
test_loader = torch.utils.data.DataLoader(torch_dataset_test,
shuffle=False,
batch_size=64)
snn_loader = torch.utils.data.DataLoader(snn_dataset,
shuffle=False,
batch_size=1)
model = Net().to(device)
snn_model = CatNet(args.T).to(device)
if args.resume != None:
load_model(torch.load(args.resume), model)
optimizer = optim.Adadelta(model.parameters(), lr=args.lr)
scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
Acc = 0
for epoch in range(1, args.epochs + 1):
train(args, model, device, train_loader, optimizer, epoch)
test(model, device, train_loader)
Acc_ = test(model, device, test_loader)
if Acc_ > Acc:
Acc = Acc_
torch.save(model.state_dict(), "ecg2c_1.pt")
scheduler.step()
fuse_module(model)
transfer_model(model, snn_model)
test(snn_model, device, snn_loader)
def main():
# Training settings
parser = argparse.ArgumentParser(
description='PyTorch Cifar10 LeNet Example')
parser.add_argument('--batch-size',
type=int,
default=64,
metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test-batch-size',
type=int,
default=1000,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=50,
metavar='N',
help='number of epochs to train (default: 14)')
parser.add_argument('--lr',
type=float,
default=1e-3,
metavar='LR',
help='learning rate (default: 1)')
parser.add_argument('--gamma',
type=float,
default=0.7,
metavar='M',
help='Learning rate step gamma (default: 0.7)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument('--dry-run',
action='store_true',
default=False,
help='quickly check a single pass')
parser.add_argument('--seed',
type=int,
default=1,
metavar='S',
help='random seed (default: 1)')
parser.add_argument(
'--log-interval',
type=int,
default=10,
metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--save-model',
action='store_true',
default=False,
help='For Saving the current Model')
parser.add_argument('--resume',
type=str,
default=None,
metavar='RESUME',
help='Resume model from checkpoint')
parser.add_argument('--T',
type=int,
default=1000,
metavar='N',
help='SNN time window')
parser.add_argument('--k',
type=int,
default=100,
metavar='N',
help='Data augmentation')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'batch_size': args.batch_size}
if use_cuda:
kwargs.update({
'num_workers': 1,
'pin_memory': True,
'shuffle': True
}, )
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=6),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
AddGaussianNoise(std=0.01)
])
transform_test = transforms.Compose([transforms.ToTensor()])
trainset = datasets.CIFAR10(root='./data',
train=True,
download=True,
transform=transform_train)
train_loader_ = torch.utils.data.DataLoader(trainset,
batch_size=512,
shuffle=True)
for i in range(args.k):
im_aug = transforms.Compose([
transforms.RandomRotation(10),
transforms.RandomCrop(32, padding=6),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
AddGaussianNoise(std=0.01)
])
trainset = trainset + datasets.CIFAR10(
root='./data', train=True, download=True, transform=im_aug)
train_loader = torch.utils.data.DataLoader(trainset,
batch_size=256 + 512,
shuffle=True)
testset = datasets.CIFAR10(root='./data',
train=False,
download=False,
transform=transform_test)
test_loader = torch.utils.data.DataLoader(testset,
batch_size=100,
shuffle=False)
snn_dataset = SpikeDataset(testset, T=args.T)
snn_loader = torch.utils.data.DataLoader(snn_dataset,
batch_size=10,
shuffle=False)
from models.vgg import VGG, VGG_, CatVGG, CatVGG_
model = VGG('VGG16', clamp_max=1, quantize_bit=32, bias=False).to(device)
snn_model = CatVGG('VGG16', args.T, bias=True).to(device)
#Trainable pooling
#model = VGG_('VGG19_', clamp_max=1, quantize_bit=32,bias =True).to(device)
#snn_model = CatVGG_('VGG19_', args.T,bias =True).to(device)
if args.resume != None:
model.load_state_dict(torch.load(args.resume), strict=False)
optimizer = optim.Adam(model.parameters(), lr=args.lr)
scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
correct_ = 0
for epoch in range(1, args.epochs + 1):
train(args, model, device, train_loader, optimizer, epoch)
test(model, device, train_loader_)
correct = test(model, device, test_loader)
if correct > correct_:
correct_ = correct
scheduler.step()
model = fuse_bn_recursively(model)
transfer_model(model, snn_model)
with torch.no_grad():
normalize_weight(snn_model.features, quantize_bit=32)
test(snn_model, device, snn_loader)