dali_iter = DALIClassificationIterator(train_pipes, size)
while it < iters + 1:
for batches in tqdm(dali_iter):
if it == lr_steps[lr_counter]:
trainer.set_learning_rate(trainer.learning_rate * 0.1)
lr_counter += 1
datas, labels = split_and_load(batches, num_gpu)
with ag.record():
ots = [net(X) for X in datas]
embedds = [ot[0] for ot in ots]
outputs = [ot[1] for ot in ots]
losses = [
loss(yhat, y, emb)
for yhat, y, emb in zip(outputs, labels, embedds)
]
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(
it = epoch * batches_per_epoch + i
lr_scheduler.update(i, epoch)
datas = gluon.utils.split_and_load(batch[0],
ctx_list=ctx,
batch_axis=0,
even_split=False)
labels = gluon.utils.split_and_load(batch[1],
ctx_list=ctx,
batch_axis=0,
even_split=False)
with ag.record():
ots = [net(X) for X in datas]
outputs = [ot[1] for ot in ots]
losses = [loss(yhat, y) for yhat, y in zip(outputs, labels)]
for l in losses:
ag.backward(l)
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'
# metrics
loss_mtc, acc_mtc = mx.metric.Loss(), mx.metric.Accuracy()
tic = time.time()
btic = time.time()
# train loop
for epoch in range(epochs):
for i, batch in enumerate(train_iter):
it = epoch * num_batches + i
data = batch[0].data[0]
label = batch[0].label[0]
with ag.record():
embedding, output = net(data)
batch_loss = loss(output, label, embedding)
ag.backward(batch_loss)
trainer.step(batch_size)
acc_mtc.update([label], [output])
loss_mtc.update(0, [batch_loss])
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 | '
'learning rate: %.8f speed: %.2f samples/s, time: %.6f' %
(epoch, it, train_loss, train_acc, trainer.learning_rate,