def _run_network(self, dataset_sink_mode=False, num_samples=2, **kwargs):
lenet = LeNet5()
loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
optim = Momentum(lenet.trainable_params(),
learning_rate=0.1,
momentum=0.9)
model = Model(lenet,
loss_fn=loss,
optimizer=optim,
metrics={'loss': Loss()})
summary_dir = tempfile.mkdtemp(dir=self.base_summary_dir)
summary_collector = SummaryCollector(summary_dir=summary_dir,
collect_freq=2,
**kwargs)
ds_train = create_dataset(os.path.join(self.mnist_path, "train"),
num_samples=num_samples)
model.train(1,
ds_train,
callbacks=[summary_collector],
dataset_sink_mode=dataset_sink_mode)
ds_eval = create_dataset(os.path.join(self.mnist_path, "test"))
model.eval(ds_eval,
dataset_sink_mode=dataset_sink_mode,
callbacks=[summary_collector])
return summary_dir
def test_summary_ops(self):
"""Test summary operators."""
ds_train = create_mnist_dataset('train', num_samples=1, batch_size=1)
ds_train_iter = ds_train.create_dict_iterator()
expected_data = next(ds_train_iter)['image'].asnumpy()
net = LeNet5()
loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
optim = Momentum(net.trainable_params(),
learning_rate=0.1,
momentum=0.9)
model = Model(net,
loss_fn=loss,
optimizer=optim,
metrics={'loss': Loss()})
model.train(1, ds_train, dataset_sink_mode=False)
summary_data = _get_summary_tensor_data()
image_data = summary_data['x[:Image]'].asnumpy()
tensor_data = summary_data['x[:Tensor]'].asnumpy()
x_fc3 = summary_data['x_fc3[:Scalar]'].asnumpy()
assert np.allclose(expected_data, image_data)
assert np.allclose(expected_data, tensor_data)
assert not np.allclose(0, x_fc3)
def test_loss():
"""test_loss"""
num = 5
inputs = np.random.rand(num)
loss = Loss()
for k in range(num):
loss.update(Tensor(np.array([inputs[k]])))
assert inputs.mean() == loss.eval()
loss.clear()
with pytest.raises(RuntimeError):
loss.eval()
config.label_smooth_factor = 0.0
loss = CrossEntropySmooth(sparse=True,
reduction="mean",
smooth_factor=config.label_smooth_factor,
num_classes=config.class_num)
else:
loss = SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
loss_scale = FixedLossScaleManager(config.loss_scale,
drop_overflow_update=False)
model = Model(net,
loss_fn=loss,
optimizer=opt,
loss_scale_manager=loss_scale,
metrics={
"top1": Top1CategoricalAccuracy(),
"loss": Loss()
},
amp_level="O2",
keep_batchnorm_fp32=False)
else:
# GPU target
if args.dataset == "imagenet2012":
if not config.use_label_smooth:
config.label_smooth_factor = 0.0
loss = CrossEntropySmooth(sparse=True,
reduction="mean",
smooth_factor=config.label_smooth_factor,
num_classes=config.class_num)
else:
loss = SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
def test_loss_shape_error():
loss = Loss()
inp = np.ones(shape=[2, 2])
with pytest.raises(ValueError):
loss.update(inp)
def test_loss_inputs_error():
loss = Loss()
with pytest.raises(ValueError):
loss(np.array(1), np.array(2))