def rev_valid(valid_loader, network, criterion, extra_info, print_freq):
losses, top1, top5 = AverageMeter(), AverageMeter(), AverageMeter()
network.eval()
network.apply(change_key('search_mode', 'search'))
with torch.no_grad():
for i, (inputs, targets) in enumerate(valid_loader):
targets = targets.cuda(non_blocking=True)
logits, expected_flop = network(inputs)
loss = criterion(logits, targets)
prec1, prec5 = obtain_accuracy(logits.data,
targets.data,
topk=(1, 5))
losses.update(loss.item(), inputs.size()[0])
top1.update(prec1.item(), inputs.size()[0])
top5.update(prec5.item(), inputs.size()[0])
if i % print_freq == 0 or (i + 1) == len(valid_loader):
print('**Valid** [{:}][{:03d}/{:03d}]'.format(
extra_info, i, len(valid_loader)))
print(
'**VALID** Prec@1 {top1.avg:.2f} Prec@5 {top5.avg:.2f} Error@1 {error1:.2f} Error@5 {error5:.2f}'
.format(top1=top1,
top5=top5,
error1=100 - top1.avg,
error5=100 - top5.avg))
return losses.avg, top1.avg, top5.avg
def finetune_procedure(xloader, network, criterion, scheduler, optimizer, mode,
config, extra_info, print_freq):
losses, top1, top5 = AverageMeter(), AverageMeter(), AverageMeter()
if mode == 'train':
network.train()
elif mode == 'valid':
network.eval()
else:
raise ValueError
for i, inputs_tuple in enumerate(xloader):
if mode == 'train':
inputs, targets, _, __ = inputs_tuple
scheduler.update(None, 1.0 * i / len(xloader))
elif mode == 'valid':
inputs, targets = inputs_tuple
inputs = inputs.cuda(non_blocking=True)
targets = targets.cuda(non_blocking=True)
if mode == 'train':
optimizer.zero_grad()
features, logits = network(inputs)
if isinstance(logits, list):
logits, logits_aux = logits
else:
logits, logits_aux = logits, None
loss = criterion(logits, targets)
if config is not None and hasattr(
config, 'auxiliary') and config.auxiliary > 0:
loss_aux = criterion(logits_aux, targets)
loss += config.auxiliary * loss_aux
if mode == 'train':
loss.backward()
optimizer.step()
prec1, prec5 = obtain_accuracy(logits.data, targets.data, topk=(1, 5))
losses.update(loss.item(), inputs.size(0))
top1.update(prec1.item(), inputs.size(0))
top5.update(prec5.item(), inputs.size(0))
if mode == 'valid' and i % print_freq == 0:
print('**Valid** [{:}][{:03d}/{:03d}]'.format(
extra_info, i, len(xloader)))
return losses.avg, top1.avg, top5.avg
def search_train(search_loader, network, criterion, scheduler, base_optimizer,
arch_optimizer, optim_config, extra_info, print_freq):
arch_losses, top1, top5 = AverageMeter(), AverageMeter(), AverageMeter()
arch_cls_losses, arch_flop_losses = AverageMeter(), AverageMeter()
epoch_str, flop_need, flop_weight, flop_tolerant = extra_info[
'epoch-str'], extra_info['FLOP-exp'], extra_info[
'FLOP-weight'], extra_info['FLOP-tolerant']
# train
network.train()
print('[Search] : {:}, FLOP-Require={:.2f}, FLOP-Weight={:.2f}'.format(
epoch_str, flop_need, flop_weight))
network.apply(change_key('search_mode', 'search'))
for step, (base_inputs, base_targets, arch_inputs,
arch_targets) in enumerate(search_loader):
#scheduler.update(None, 1*step / len(search_loader))
#base_targets = base_targets.cuda(non_blocking = True)
arch_inputs = arch_inputs.cuda(non_blocking=True)
arch_targets = arch_targets.cuda(non_blocking=True)
# update arch parameters
arch_optimizer.zero_grad()
logits, expected_flop = network(arch_inputs)
flop_cur = network.get_flop('genotype', None, None)
#flop_cur = network.module.get_flop('genotype', None, None)
flop_loss, flop_loss_scale = get_flop_loss(expected_flop, flop_cur,
flop_need, flop_tolerant)
loss = criterion(logits, arch_targets)
arch_loss = loss + flop_loss * flop_weight
arch_loss.backward()
arch_optimizer.step()
# record
prec1, prec5 = obtain_accuracy(logits.data,
arch_targets.data,
topk=(1, 5))
top1.update(prec1.item(), arch_inputs.size()[0])
top5.update(prec5.item(), arch_inputs.size()[0])
arch_losses.update(arch_loss.item(), arch_inputs.size()[0])
arch_flop_losses.update(flop_loss_scale, arch_inputs.size()[0])
arch_cls_losses.update(loss.item(), arch_inputs.size()[0])
print('**TRAIN** Prec@1 {top1.avg:.2f} Prec@5 {top5.avg:.2f} Error@1 {error1:.2f} Error@5 {error5:.2f}\
Arch-loss:{archloss:.3f}, arch_flop_loss:{archflop:.3f} arch_cls_loss:{clsloss:.3f}' .format(top1=top1, top5=top5, error1 = 100-top1.avg, error5 = 100-top5.avg,\
archloss=arch_losses.avg, archflop=arch_flop_losses.avg, clsloss=arch_cls_losses.avg))
print('Current FLOP at {:} is {:}'.format(epoch_str, flop_cur))
return arch_losses.avg, top1.avg, top5.avg