def local_infer(self, valid_queue, model, criterion):
objs = utils.AvgrageMeter()
top1 = utils.AvgrageMeter()
top5 = utils.AvgrageMeter()
model.eval()
loss = None
for step, (input, target) in enumerate(valid_queue):
input = input.to(self.device)
target = target.to(self.device)
logits = model(input)
loss = criterion(logits, target)
prec1, prec5 = utils.accuracy(logits, target, topk=(1, 5))
n = input.size(0)
objs.update(loss.item(), n)
top1.update(prec1.item(), n)
top5.update(prec5.item(), n)
if step % self.args.report_freq == 0:
logging.info('client_index = %d, valid %03d %e %f %f',
self.client_index, step, objs.avg, top1.avg,
top5.avg)
return top1.avg / 100.0, objs.avg / 100.0, loss
def local_search(self, train_queue, valid_queue, model, architect,
criterion, optimizer):
objs = utils.AvgrageMeter()
top1 = utils.AvgrageMeter()
top5 = utils.AvgrageMeter()
loss = None
for step, (input, target) in enumerate(train_queue):
# logging.info("epoch %d, step %d START" % (epoch, step))
n = input.size(0)
# model.set_tau(
# self.args.tau_max - self.args.epochs * 1.0 / self.args.epochs * (self.args.tau_max - self.args.tau_min))
input = input.to(self.device)
target = target.to(self.device)
# get a random minibatch from the search queue with replacement
input_search, target_search = next(iter(valid_queue))
input_search = input_search.to(self.device)
target_search = target_search.to(self.device)
architect.step_v2(input, target, input_search, target_search,
self.args.lambda_train_regularizer,
self.args.lambda_valid_regularizer)
optimizer.zero_grad()
logits = model(input)
loss = criterion(logits, target)
loss.backward()
parameters = model.arch_parameters()
nn.utils.clip_grad_norm_(parameters, self.args.grad_clip)
optimizer.step()
# logging.info("step %d. update weight by SGD. FINISH\n" % step)
prec1, prec5 = utils.accuracy(logits, target, topk=(1, 5))
objs.update(loss.item(), n)
top1.update(prec1.item(), n)
top5.update(prec5.item(), n)
# torch.cuda.empty_cache()
if step % self.args.report_freq == 0:
logging.info('client_index = %d, search %03d %e %f %f',
self.client_index, step, objs.avg, top1.avg,
top5.avg)
return top1.avg / 100.0, objs.avg / 100.0, loss
def local_train(self, train_queue, valid_queue, model, criterion,
optimizer):
objs = utils.AvgrageMeter()
top1 = utils.AvgrageMeter()
top5 = utils.AvgrageMeter()
for step, (input, target) in enumerate(train_queue):
# logging.info("epoch %d, step %d START" % (epoch, step))
model.train()
n = input.size(0)
input = input.to(self.device)
target = target.to(self.device)
optimizer.zero_grad()
logits, logits_aux = model(input)
loss = criterion(logits, target)
if self.args.auxiliary:
loss_aux = criterion(logits_aux, target)
loss += self.args.auxiliary_weight * loss_aux
loss.backward()
parameters = model.parameters()
nn.utils.clip_grad_norm_(parameters, self.args.grad_clip)
optimizer.step()
# logging.info("step %d. update weight by SGD. FINISH\n" % step)
prec1, prec5 = utils.accuracy(logits, target, topk=(1, 5))
objs.update(loss.item(), n)
top1.update(prec1.item(), n)
top5.update(prec5.item(), n)
# torch.cuda.empty_cache()
if step % self.args.report_freq == 0:
logging.info('train %03d %e %f %f', step, objs.avg, top1.avg,
top5.avg)
return top1.avg, objs.avg, loss