trainer.step(batch_size)
acc_mtc.update(labels, outputs)
loss_mtc.update(0, losses)
if (it % opt.save_frequency) == 0:
_, train_loss = loss_mtc.get()
train_metric_name, train_acc = acc_mtc.get()
toc = time.time()
logger.info(
'\nEpoch[%d] Batch[%d]\tSpeed: %f samples/sec\t%s=%f\tlr=%f train loss: %.6f'
% (epoch, it, batch_size * opt.save_frequency /
(toc - tic), train_metric_name, train_acc,
trainer.learning_rate, train_loss))
acc_list = validate(logger, net, ctx, val_datas, targets, epoch,
it)
# save flags
do_save = False
is_highest = False
# check if save or not
if len(acc_list) > 0:
score = sum(acc_list)
if acc_list[-1] >= highest_acc[-1]:
if acc_list[-1] > highest_acc[-1]:
is_highest = True
else:
if score >= highest_acc[0]:
is_highest = True
highest_acc[0] = score
highest_acc[-1] = acc_list[-1]
for l in losses:
ag.backward(l)
trainer.step(batch_size)
acc_mtc.update(labels, outputs)
loss_mtc.update(0, losses)
if (it % save_period) == 0 and it != 0:
_, train_loss = loss_mtc.get()
_, train_acc = acc_mtc.get()
toc = time.time()
logger.info(
'\n[epoch % 2d] [it % 3d] train loss: %.6f, train_acc: %.6f | '
'speed: %.2f samples/s, time: %.6f' %
(epoch, it, train_loss, train_acc, batch_size /
(toc - btic), toc - tic))
logger.info("Radius {}".format(
loss.R.data(ctx=mx.gpu(0)).asscalar()))
results = validate(net, ctx, val_datas, targets)
for result in results:
logger.info('{}'.format(result))
loss_mtc.reset()
acc_mtc.reset()
tic = time.time()
net.save_parameters(
"../../models/mobilefacenet-ring-it-%d.params" % it)
btic = time.time()
it += 1
epoch += 1
dali_iter.reset()