def validate(data_loader, net, loss):
start_time = time.time()
net.eval()
metrics = []
for i, (data, target, coord) in enumerate(data_loader):
data = Variable(data.cuda(async = True), volatile = True)
target = Variable(target.cuda(async = True), volatile = True)
coord = Variable(coord.cuda(async = True), volatile = True)
output = net(data, coord)
loss_output = loss(output, target, train = False)
loss_output[0] = loss_output[0].data[0]
metrics.append(loss_output)
end_time = time.time()
metrics = np.asarray(metrics, np.float32)
print('Validation: tpr %3.2f, tnr %3.8f, total pos %d, total neg %d, time %3.2f' % (
100.0 * np.sum(metrics[:, 6]) / np.sum(metrics[:, 7]),
100.0 * np.sum(metrics[:, 8]) / np.sum(metrics[:, 9]),
np.sum(metrics[:, 7]),
np.sum(metrics[:, 9]),
end_time - start_time))
print('loss %2.4f, classify loss %2.4f, regress loss %2.4f, %2.4f, %2.4f, %2.4f' % (
np.mean(metrics[:, 0]),
np.mean(metrics[:, 1]),
np.mean(metrics[:, 2]),
np.mean(metrics[:, 3]),
np.mean(metrics[:, 4]),
np.mean(metrics[:, 5])))
print
print
def singletest(data,net,config,splitfun,combinefun,n_per_run,margin = 64,isfeat=False):
z, h, w = data.size(2), data.size(3), data.size(4)
print(data.size())
data = splitfun(data,config['max_stride'],margin)
data = Variable(data.cuda(async = True), volatile = True,requires_grad=False)
splitlist = range(0,args.split+1,n_per_run)
outputlist = []
featurelist = []
for i in range(len(splitlist)-1):
if isfeat:
output,feature = net(data[splitlist[i]:splitlist[i+1]])
featurelist.append(feature)
else:
output = net(data[splitlist[i]:splitlist[i+1]])
output = output.data.cpu().numpy()
outputlist.append(output)
output = np.concatenate(outputlist,0)
output = combinefun(output, z / config['stride'], h / config['stride'], w / config['stride'])
if isfeat:
feature = np.concatenate(featurelist,0).transpose([0,2,3,4,1])
feature = combinefun(feature, z / config['stride'], h / config['stride'], w / config['stride'])
return output,feature
else:
return output
def train(data_loader, net, loss, epoch, optimizer, get_lr, save_freq, save_dir):
start_time = time.time()
net.train()
lr = get_lr(epoch)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
metrics = []
for i, (data, target, coord) in enumerate(data_loader):
data = Variable(data.cuda(async = True))
target = Variable(target.cuda(async = True))
coord = Variable(coord.cuda(async = True))
output = net(data, coord)
loss_output = loss(output, target)
optimizer.zero_grad()
loss_output[0].backward()
optimizer.step()
loss_output[0] = loss_output[0].data[0]
metrics.append(loss_output)
if epoch % args.save_freq == 0:
state_dict = net.module.state_dict()
for key in state_dict.keys():
state_dict[key] = state_dict[key].cpu()
torch.save({
'epoch': epoch,
'save_dir': save_dir,
'state_dict': state_dict,
'args': args},
os.path.join(save_dir, '%03d.ckpt' % epoch))
end_time = time.time()
metrics = np.asarray(metrics, np.float32)
print('Epoch %03d (lr %.5f)' % (epoch, lr))
print('Train: tpr %3.2f, tnr %3.2f, total pos %d, total neg %d, time %3.2f' % (
100.0 * np.sum(metrics[:, 6]) / np.sum(metrics[:, 7]),
100.0 * np.sum(metrics[:, 8]) / np.sum(metrics[:, 9]),
np.sum(metrics[:, 7]),
np.sum(metrics[:, 9]),
end_time - start_time))
print('loss %2.4f, classify loss %2.4f, regress loss %2.4f, %2.4f, %2.4f, %2.4f' % (
np.mean(metrics[:, 0]),
np.mean(metrics[:, 1]),
np.mean(metrics[:, 2]),
np.mean(metrics[:, 3]),
np.mean(metrics[:, 4]),
np.mean(metrics[:, 5])))
print
def test():
global best_acc
model.eval()
test_loss = 0
correct = 0
bin_op.binarization()
for data, target in testloader:
data, target = Variable(data.cuda()), Variable(target.cuda())
output = model(data)
test_loss += criterion(output, target).item()
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
bin_op.restore()
acc = 100. * torch.tensor(correct, dtype=torch.float64) / len(
test_loader.dataset)
if acc > best_acc:
best_acc = acc
save_state(model, best_acc)
test_loss /= len(testloader.dataset)
writer.add_scalar('testing_loss', test_loss * args.batch_size, epoch)
writer.add_scalar('testing_accuracy',
100. * correct / len(testloader.dataset), epoch)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
test_loss * args.batch_size, correct, len(testloader.dataset),
100. * correct / len(testloader.dataset)))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return
def train(epoch, writer):
model.train()
for batch_idx, (data, target) in enumerate(trainloader):
# process the weights including binarization
bin_op.binarization()
# forwarding
data, target = Variable(data.cuda()), Variable(target.cuda())
optimizer.zero_grad()
output = model(data)
# backwarding
loss = criterion(output, target)
loss.backward()
# restore weights
bin_op.restore()
bin_op.updateBinaryGradWeight()
optimizer.step()
if batch_idx % 100 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\tLR: {}'.
format(epoch, batch_idx * len(data),
len(trainloader.dataset),
100. * batch_idx / len(trainloader), loss.item(),
optimizer.param_groups[0]['lr']))
#To-Do: Fix it
bin_op.binarization()
#writer.add_histogram('weights',model.state_dict()['bin_conv2.conv.weight'], epoch)
bin_op.restore()
return
def test():
global best_acc
model.eval()
test_loss = 0
correct = 0
testing = True
bin_params.binarize()
for data, target in testloader:
if not args.cpu:
data, target = data.cuda(), target.cuda()
data, target = Variable(data), Variable(target)
output = model(data, testing)
test_loss += criterion(output, target).data.item()
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
bin_params.restore()
acc = 100. * correct.item() / float(len(testloader.dataset))
# Save the model params if the accuracy is the highest yet
if acc > best_acc:
best_acc = acc
save_state(model, best_acc)
test_loss /= float(len(testloader.dataset))
print('\nTest Accuracy: {}/{} ({:.2f}%)'.format(
correct, len(testloader.dataset), 100. * correct.item() / float(len(testloader.dataset)))
)
return
def batchify(data, bsz):
nbatch = data.size(0) // bsz # Work out how cleanly we can divide the dataset into bsz parts.
data = data.narrow(0, 0, nbatch * bsz) # Trim off any extra elements that wouldn't cleanly fit (remainders).
data = data.view(bsz, -1).t().contiguous() # Evenly divide the data across the bsz batches.
if args.cuda: # If we can do this on the gpu
data = data.cuda() # Then move the data to the gpu
return data # Return the answer
def batchify(data, bsz):
nbatch = data.size(0) // bsz
data = data.narrow(0, 0, nbatch * bsz)
data = data.view(bsz, -1).t().contiguous()
if args.cuda:
data = data.cuda()
return data
def test2(evaluate=False):
global best_acc
model.eval()
test_loss = 0
correct = 0
bin_op.binarizationTest()
for data, target in test_loader:
data, target = data.cuda(), target.cuda()
data, target = Variable(data, volatile=True), Variable(target)
output = model(data)
test_loss += criterion(output, target).data[0]
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
bin_op.restore()
acc = 100. * correct / len(test_loader.dataset)
test_loss /= len(test_loader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
test_loss * 128., correct, len(test_loader.dataset),
100. * correct / len(test_loader.dataset)))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return
def test(save_name, best_acc, sample_weights=torch.Tensor(np.ones((50000,1))/50000.0)):
#global best_acc
model.eval()
test_loss = 0
correct = 0
bin_op.binarization()
sampler = torch.utils.data.sampler.WeightedRandomSampler(sample_weights[:,0].double(), 50000)
testloader_in = torch.utils.data.DataLoader(trainset, batch_size=1000,
shuffle=False, num_workers=4, sampler=sampler)
for data, target in testloader_in:
data, target = Variable(data.cuda()), Variable(target.cuda())
output = model(data)
test_loss += criterion(output, target).data[0]
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
bin_op.restore()
acc = 100. * correct / len(testloader_in.dataset)
if acc > best_acc:
best_acc = acc
save_state(model, best_acc, save_name)
test_loss = test_loss / len(testloader_in.dataset) * 1000
print('\nTrain set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
test_loss , correct, len(testloader_in.dataset),
100. * correct / len(testloader_in.dataset)))
print('Best Train Accuracy: {:.2f}%\n'.format(best_acc))
return best_acc
def test():
net.eval()
# two_nets = TwoNets(model, net)
loss_avg = 0.0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.cuda(), target.cuda()
# adv_data = adversary_test(two_nets, data, target)
#
# # forward
# output = two_nets(adv_data)
generated = model(data, mode='edge_forward').detach()
output = net(generated)
loss = F.cross_entropy(output, target)
# accuracy
pred = output.data.max(1)[1]
correct += pred.eq(target.data).sum().item()
# test loss average
loss_avg += float(loss.data)
state['test_loss'] = loss_avg / len(test_loader)
state['test_accuracy'] = correct / len(test_loader.dataset)
def get_error_output(data, target, batch_sample_weights): data, target = Variable(data.cuda()), Variable(target.cuda()) output = model(data) loss = (criterion_seperated(output, target)*Variable(batch_sample_weights.cuda().float())).mean() return output
def validate(data_loader, net, loss):
start_time = time.time()
net.eval()
metrics = []
for i, (data, target, coord) in enumerate(data_loader):
#with torch.no_grad():
data = Variable(data.cuda(), volatile=True)
target = Variable(target.cuda(), volatile=True)
coord = Variable(coord.cuda(), volatile=True)
output = net(data, coord)
loss_output = loss(output, target, train=False)
loss_output[0] = loss_output[0].item()
metrics.append(loss_output)
end_time = time.time()
metrics = np.asarray(metrics, np.float32)
print(
'Validation: tpr %3.2f, tnr %3.8f, total pos %d, total neg %d, time %3.2f'
% (100.0 * np.sum(metrics[:, 6]) / np.sum(metrics[:, 7]), 100.0 *
np.sum(metrics[:, 8]) / np.sum(metrics[:, 9]), np.sum(
metrics[:, 7]), np.sum(metrics[:, 9]), end_time - start_time))
print(
'loss %2.4f, classify loss %2.4f, regress loss %2.4f, %2.4f, %2.4f, %2.4f'
% (np.mean(metrics[:, 0]), np.mean(
metrics[:, 1]), np.mean(metrics[:, 2]), np.mean(metrics[:, 3]),
np.mean(metrics[:, 4]), np.mean(metrics[:, 5])))
print
print
def train(epoch):
model.train()
for batch_idx, (data, target) in enumerate(trainloader):
# process the weights including binarization
bin_op.binarization()
# forwarding
if torch.cuda.is_available():
data, target = Variable(data.cuda()), Variable(target.cuda())
optimizer.zero_grad()
output = model(data)
# backwarding
loss = criterion(output, target)
loss.backward()
# restore weights
bin_op.restore()
bin_op.updateBinaryGradWeight()
optimizer.step()
if batch_idx % 100 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\tLR: {}'.format(
epoch, batch_idx * len(data), len(trainloader.dataset),
100. * batch_idx / len(trainloader), loss.data[0],
optimizer.param_groups[0]['lr']))
return
def singletest(data,
net,
config,
splitfun,
combinefun,
n_per_run,
margin=64,
isfeat=False):
z, h, w = data.size(2), data.size(3), data.size(4)
print(data.size())
data = splitfun(data, config['max_stride'], margin)
data = Variable(data.cuda())
splitlist = range(0, args.split + 1, n_per_run)
outputlist = []
featurelist = []
for i in range(len(splitlist) - 1):
if isfeat:
output, feature = net(data[splitlist[i]:splitlist[i + 1]])
featurelist.append(feature)
else:
output = net(data[splitlist[i]:splitlist[i + 1]])
output = output.data.cpu().numpy()
outputlist.append(output)
output = np.concatenate(outputlist, 0)
output = combinefun(output, z / config['stride'], h / config['stride'],
w / config['stride'])
if isfeat:
feature = np.concatenate(featurelist, 0).transpose([0, 2, 3, 4, 1])
feature = combinefun(feature, z / config['stride'],
h / config['stride'], w / config['stride'])
return output, feature
else:
return output
def validate(data_loader, net, loss):
start_time = time.time()
net.eval()
metrics = []
for i, (data, target, coord) in enumerate(data_loader):
data = Variable(data.cuda())
target = Variable(target.cuda())
coord = Variable(coord.cuda())
output = net(data, coord)
loss_output = loss(output, target, train=False)
# print('output',len(output))
# for numi in range(len(output)):
# print('-------',output[numi].shape,target[numi].shape)
# print('target',len(target))
# loss_output[0] = loss_output[0].data[0]
loss_output[0] = loss_output[0].item()
metrics.append(loss_output)
end_time = time.time()
metrics = np.asarray(metrics, np.float32)
print(
'Validation: tpr %3.2f, tnr %3.8f, total pos %d, total neg %d, time %3.2f'
% (100.0 * np.sum(metrics[:, 6]) / np.sum(metrics[:, 7]), 100.0 *
np.sum(metrics[:, 8]) / np.sum(metrics[:, 9]), np.sum(
metrics[:, 7]), np.sum(metrics[:, 9]), end_time - start_time))
print(
'loss %2.4f, classify loss %2.4f, regress loss %2.4f, %2.4f, %2.4f, %2.4f'
% (np.mean(metrics[:, 0]), np.mean(
metrics[:, 1]), np.mean(metrics[:, 2]), np.mean(metrics[:, 3]),
np.mean(metrics[:, 4]), np.mean(metrics[:, 5])))
print
print
def test():
global best_acc
model.eval()
test_loss = 0
correct = 0
bin_op.binarization()
for data, target in testloader:
data, target = Variable(data.cuda()), Variable(target.cuda())
output = model(data)
test_loss += criterion(output, target).data.item()
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
bin_op.restore()
acc = 100. * float(correct) / len(testloader.dataset)
if acc > best_acc:
best_acc = acc
save_state(model, best_acc)
test_loss /= len(testloader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
test_loss * 128., correct, len(testloader.dataset),
100. * float(correct) / len(testloader.dataset)))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return
def train(epoch):
model.train()
for batch_idx, (data, target) in enumerate(trainloader):
# forwarding
if torch.cuda.is_available():
data = data.cuda()
target = target.cuda()
data, target = Variable(data), Variable(target)
optimizer.zero_grad()
output = model(data)
# backwarding
loss = criterion(output, target)
loss.backward()
optimizer.step()
if batch_idx % 100 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\tLR: {}'.
format(epoch, batch_idx * len(data),
len(trainloader.dataset),
100. * batch_idx / len(trainloader), loss.data.item(),
optimizer.param_groups[0]['lr']))
return
def train(data_loader, net, loss, epoch, optimizer, get_lr, save_freq,
save_dir):
start_time = time.time()
net.train()
lr = get_lr(epoch)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
metrics = []
for i, (data, target, coord) in enumerate(data_loader):
data = Variable(data.cuda(), requires_grad=True)
target = Variable(target.cuda())
coord = Variable(coord.cuda(), requires_grad=True)
output = net(data, coord)
# print('---------', data.size(),coord.size(),target.size(),output.size())#torch.Size([1, 1, 96, 96, 96]), torch.Size([1, 3, 24, 24, 24]), torch.Size([1, 24, 24, 24, 3, 5]), torch.Size([1, 24, 24, 24, 3, 5]))
loss_output = loss(output, target)
optimizer.zero_grad()
# print('loss_output[0]',loss_output.shape, loss_output[0].shape)
loss_output[0].backward()
optimizer.step()
# loss_output[0] = loss_output[0].data[0]
loss_output[0] = loss_output[0].item()
# print('loss_output[1]',loss_output[0])
metrics.append(loss_output)
# metrics = np.asarray(metrics)
# print('metrics',type(metrics))
if epoch % args.save_freq == 0:
state_dict = net.module.state_dict()
for key in state_dict.keys():
state_dict[key] = state_dict[key].cpu()
torch.save(
{
'epoch': epoch,
'save_dir': save_dir,
'state_dict': state_dict,
'args': args
}, os.path.join(save_dir, '%03d.ckpt' % epoch))
end_time = time.time()
metrics = np.asarray(metrics, dtype=np.float32)
print('Epoch %03d (lr %.5f)' % (epoch, lr))
print(
'Train: tpr %3.2f, tnr %3.2f, total pos %d, total neg %d, time %3.2f'
% (100.0 * np.sum(metrics[:, 6]) / np.sum(metrics[:, 7]), 100.0 *
np.sum(metrics[:, 8]) / np.sum(metrics[:, 9]), np.sum(
metrics[:, 7]), np.sum(metrics[:, 9]), end_time - start_time))
print(
'loss %2.4f, classify loss %2.4f, regress loss %2.4f, %2.4f, %2.4f, %2.4f'
% (np.mean(metrics[:, 0]), np.mean(
metrics[:, 1]), np.mean(metrics[:, 2]), np.mean(metrics[:, 3]),
np.mean(metrics[:, 4]), np.mean(metrics[:, 5])))
print
def test():
global best_acc
model.eval() # 모듈을 평가모드로 설정
test_loss = 0
correct = 0
bin_op.binarization()
for data, target in testloader:
data, target = Variable(data.cuda()), Variable(target.cuda())
output = model(data)
test_loss += criterion(output, target).data.item()
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
bin_op.restore()
acc = 100. * float(correct) / len(testloader.dataset)
if acc > best_acc:
best_acc = acc
save_state(model, best_acc)
test_loss /= len(testloader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
test_loss * 128., correct, len(testloader.dataset),
100. * float(correct) / len(testloader.dataset)))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return
def test(model, best_acc, studflag=True):
test_loss = 0.0
correct = 0
model.eval()
bin_op.binarization()
for data, target in testloader:
data, target = Variable(data.cuda()), Variable(target.cuda())
output, h1_student, h2_student = model(data)
test_loss += criterion(output, target).data[0]
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
bin_op.restore()
acc = 100. * correct / len(testloader.dataset)
if studflag == False:
print("Teacher showing student")
else:
print(acc, best_acc)
if acc > best_acc:
best_acc = acc
save_state(model, best_acc)
test_loss /= len(testloader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
test_loss * 128., correct, len(testloader.dataset),
100. * correct / len(testloader.dataset)))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return best_acc
def train(epoch, sample_weights=torch.Tensor(np.ones((50000, 1)) / 50000.0)):
adjust_learning_rate(optimizer, epoch)
model.train()
sampler = torch.utils.data.sampler.WeightedRandomSampler(
sample_weights[:, 0].double(), 50000)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=128, \
shuffle=False, num_workers=4, \
sampler=sampler)
for batch_idx, (data, target) in enumerate(trainloader):
bin_op.binarization()
data, target = Variable(data.cuda()), Variable(target.cuda())
optimizer.zero_grad()
output = model(data)
# backwarding
loss = criterion(output, target)
loss.backward()
# restore weights
bin_op.restore()
bin_op.updateBinaryGradWeight()
optimizer.step()
if batch_idx % 100 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\tLR: {}'.
format(epoch, batch_idx * len(data),
len(trainloader.dataset),
100. * batch_idx / len(trainloader), loss.data[0],
optimizer.param_groups[0]['lr']))
return trainloader
def get_batch(source, i, evaluation=False):
seq_len = min(args.bptt, len(source) - 1 - i)
data = Variable(source[i:i + seq_len], volatile=evaluation)
target = Variable(source[i + 1:i + 1 + seq_len].view(-1))
if args.cuda:
data = data.cuda()
return data, target
def train(epoch):
model.train()
for batch_idx, (data, target) in enumerate(trainloader):
# process the weights including binarization
bin_op.binarization()
# forwarding
data, target = Variable(data.cuda()), Variable(target.cuda())
optimizer.zero_grad()
output = model(data)
# backwarding
loss = criterion(output, target)
loss.backward()
# restore weights
bin_op.restore()
bin_op.updateBinaryGradWeight()
optimizer.step()
if batch_idx % 100 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\tLR: {}'.format(
epoch, batch_idx * len(data), len(trainloader.dataset),
100. * batch_idx / len(trainloader), loss.data.item(),
optimizer.param_groups[0]['lr']))
return
def train(epoch, trainloader, model, criterion, optimizer):
model.train()
for batch_idx, (data, target) in enumerate(trainloader):
# process the weights including binarization
# quantize the conv weights, and store the full-precision weights
if args.quantization != 'none':
bin_op.binarization()
# forwarding
data, target = Variable(data.cuda()), Variable(target.cuda())
optimizer.zero_grad()
output = model(data)
# backwarding
loss = criterion(output, target)
loss.backward()
# restore the full precision weights
if args.quantization != 'none':
bin_op.restore()
bin_op.updateBinaryGradWeight()
optimizer.step()
if batch_idx % 100 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\tLR: {}'.
format(epoch, batch_idx * len(data),
len(trainloader.dataset),
100. * batch_idx / len(trainloader), loss.data.item(),
optimizer.param_groups[0]['lr']))
return
def test(evaluate=False):
global best_acc
model.eval()
test_loss = 0
correct = 0
for data, target in test_loader:
if args.cuda:
data, target = data.cuda(), target.cuda()
data, target = Variable(data, volatile=True), Variable(target)
output = model(data)
test_loss += criterion(output, target).data[0]
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(target.data.view_as(pred)).cpu().sum()
acc = 100. * correct / len(test_loader.dataset)
if ((args.prune == 'node') and (not args.retrain)) or (acc > best_acc):
best_acc = acc
if not evaluate:
save_state(model, best_acc)
test_loss /= len(test_loader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
test_loss * args.batch_size, correct, len(test_loader.dataset),
100. * correct / len(test_loader.dataset)))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return
def test():
net.eval()
# two_nets = TwoNets(model, net)
loss_avg = 0.0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.cuda(), target.cuda()
interp_z = torch.zeros_like(data)[:,0:1,:,:].uniform_(0, 1).cuda()
generated1 = model(data, mode='just_cannyedge1').detach().cuda()
generated2 = model(data, mode='just_catedge2').detach().cuda()
generated = interp_z * generated1 + (1 - interp_z) * generated2
output = net(generated)
loss = F.cross_entropy(output, target)
# accuracy
pred = output.data.max(1)[1]
correct += pred.eq(target.data).sum().item()
# test loss average
loss_avg += float(loss.data)
state['test_loss'] = loss_avg / len(test_loader)
state['test_accuracy'] = correct / len(test_loader.dataset)
def train(model, train_loader, epoch, optimizer, criterion):
print('Training %s...' % model_str)
train_total = 0
train_correct = 0
for i, (data, labels) in enumerate(train_loader):
data = data.cuda()
labels = labels.cuda()
# Forward + Backward + Optimize
optimizer.zero_grad()
_, logits = model(data, revision=False)
prec1, = accuracy(logits, labels, topk=(1, ))
train_total += 1
train_correct += prec1
loss = criterion(logits, labels)
loss.backward()
optimizer.step()
if (i + 1) % args.print_freq == 0:
print(
'Epoch [%d/%d], Iter [%d/%d] Training Accuracy: %.4F, Loss: %.4f'
% (epoch + 1, args.n_epoch_1, i + 1,
len(train_dataset) // batch_size, prec1, loss.item()))
train_acc = float(train_correct) / float(train_total)
return train_acc
def test():
global best_acc
model.eval()
flag=True
training=False
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
bin_op.binarization()
for batch_idx,(data, target) in enumerate(testloader):
target = target.cuda(async=True)
data_var = torch.autograd.Variable(data.cuda(), volatile=True)
target_var = torch.autograd.Variable(target, volatile=True)
output,activations = model(data_var)
#Layers to run PCA on
key_idx = range(0,19)
for i in key_idx: #Run PCA layer-wise
size_keyidx = activations[i].size()
activation_i = activations[i]
run_PCA(activations,i,size_keyidx[1], threshold=0.99)
loss= criterion(output, target_var)
prec1, prec5 = accuracy(output.data, target, training, topk=(1, 5))
losses.update(loss.data[0], data.size(0))
top1.update(prec1[0], data.size(0))
top5.update(prec5[0], data.size(0))
if flag == True:
if batch_idx % 10 == 0:
print('[{0}/{1}({2:.0f}%)]\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
batch_idx, len(testloader), 100. *float(batch_idx)/len(testloader),
loss=losses, top1=top1, top5=top5))
else:
if batch_idx % 10 == 0:
print('Epoch: [{0}][{1}/{2}({3:.0f}%)]\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
epoch, batch_idx, len(testloader), 100. *float(batch_idx)/len(testloader),
loss=losses, top1=top1, top5=top5))
bin_op.restore()
acc = top1.avg
if acc > best_acc:
best_acc = acc
if flag == False:
save_state(model, best_acc)
#test_loss /= len(testloader.dataset)
print(' * Prec@1 {top1.avg:.3f} Prec@5 {top5.avg:.3f} Loss {loss.avg:.4f}'
.format(top1=top1, top5=top5,loss = losses))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return acc, losses.avg
def test(): #Testing function
global best_acc
flag = False
model.eval()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
bin_op.binarization()
for batch_idx, (data, target) in enumerate(testloader):
target = target.cuda(async=True)
data_var = torch.autograd.Variable(data.cuda(), volatile=True)
target_var = torch.autograd.Variable(target, volatile=True)
#data_var, target_var = Variable(data.cuda()), Variable(target.cuda())
output = model(data_var)
loss = criterion(output, target_var)
prec1, prec5 = accuracy(output.data, target, False, topk=(1, 5))
losses.update(loss.data[0], data.size(0))
top1.update(prec1[0], data.size(0))
top5.update(prec5[0], data.size(0))
if flag == True:
if batch_idx % 10 == 0:
print('[{0}/{1}({2:.0f}%)]\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
batch_idx,
len(testloader),
100. * float(batch_idx) / len(testloader),
loss=losses,
top1=top1,
top5=top5))
else:
if batch_idx % 10 == 0:
print('Epoch: [{0}][{1}/{2}({3:.0f}%)]\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
epoch,
batch_idx,
len(testloader),
100. * float(batch_idx) / len(testloader),
loss=losses,
top1=top1,
top5=top5))
bin_op.restore()
acc = top1.avg
if acc > best_acc:
best_acc = acc
save_state(model, best_acc)
#test_loss /= len(testloader.dataset)
print(' * Prec@1 {top1.avg:.3f} Prec@5 {top5.avg:.3f}'.format(top1=top1,
top5=top5))
#print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(
# test_loss * 128., correct, len(testloader.dataset),
# 100. * correct / len(testloader.dataset)))
print('Best Accuracy: {:.2f}%\n'.format(best_acc))
return acc, losses.avg
def batchify(data, bsz):
print("batchify")
nbatch = data.size(0) // bsz
data = data.narrow(0, 0, nbatch * bsz)
data = data.view(bsz, -1).t().contiguous()
if USE_CUDA:
data = data.cuda()
return data
def batchify(data, bsz):
# Work out how cleanly we can divide the dataset into bsz parts.
nbatch = data.size(0) // bsz
# Trim off any extra elements that wouldn't cleanly fit (remainders).
data = data.narrow(0, 0, nbatch * bsz)
# Evenly divide the data across the bsz batches.
data = data.view(bsz, -1).t().contiguous()
return data.cuda() if args.cuda else data
def batchify(data, bsz): # 这里的data类型是Tensor
nbatch = data.size(0) // bsz # 总词数除以batch_size得到需要的batch轮数
data = data.narrow(0, 0, nbatch *
bsz) # 第一个参数用来确定行(0)、列(1)，第二个参数确定开始的行/列，第三个参数确定行数/列数
data = data.view(bsz, -1).t().contiguous(
) # the size -1 is inferred from other dimensions, view成bsz行nbatch列，经过t()行列颠倒成nbatch行bsz列
if args.cuda:
data = data.cuda()
return data # 返回的是nbatch行bsz列的矩阵
def test_get_batch(source, evaluation=False):
seq_len = len(source) - 1
data = Variable(source[:seq_len], volatile=evaluation)
target = Variable(source[1:1 + seq_len].view(-1))
# This is where data should be CUDA-fied to lessen OOM errors
if args.cuda:
return data.cuda(), target.cuda()
else:
return data, target
def get_batch(source, i, evaluation=False):
seq_len = min(args.bptt, len(source) - 1 - i)
data = Variable(source[i:i + seq_len], volatile=evaluation)
target = Variable(source[i + 1:i + 1 + seq_len].view(-1))
#This is where data should be CUDA-fied to lessen OOM errors
if args.cuda:
return data.cuda(), target.cuda()
else:
return data, target
def batchify(data, bsz):
# Work out how cleanly we can divide the dataset into bsz parts.
nbatch = data.size(0) // bsz
# Trim off any extra elements that wouldn't cleanly fit (remainders).
data = data.narrow(0, 0, nbatch * bsz)
# Evenly divide the data across the bsz batches.
data = data.view(bsz, -1).t().contiguous()
if args.cuda:
data = data.cuda()
return data