def test_image_summary_data():
""" test_image_summary_data """
dataset = get_dataset()
test_data_list = []
i = 1
for next_element in dataset:
tag = "image_" + str(i) + "[:Image]"
dct = {}
dct["name"] = tag
dct["data"] = Tensor(next_element[0])
test_data_list.append(dct)
i += 1
log.debug("begin test_image_summary_sample")
# step 0: create the thread
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_IMAGE") as test_writer:
# step 1: create the test data for summary
# step 2: create the Event
_cache_summary_tensor_data(test_data_list)
test_writer.record(1)
log.debug("finished test_image_summary_sample")
def test_scalar_summary_sample_with_shape_1():
""" test_scalar_summary_sample_with_shape_1 """
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_SCALAR") as test_writer:
for i in range(1, 100):
test_data = get_test_data_shape_1(i)
_cache_summary_tensor_data(test_data)
test_writer.record(i)
def test_image_summary_train():
""" test_image_summary_train """
dataset = get_dataset()
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_IMAGE") as test_writer:
model = get_model()
callback = ImageSummaryCallback(test_writer)
model.train(2, dataset, callbacks=[callback])
def run(self):
"""
Run the explain job and save the result as a summary in summary_dir.
Note:
User should call register_saliency() once before running this function.
Raises:
ValueError: Be raised for any data or settings' value problem.
TypeError: Be raised for any data or settings' type problem.
RuntimeError: Be raised for any runtime problem.
"""
self._verify_data_n_settings(check_all=True)
with SummaryRecord(self._summary_dir) as summary:
print("Start running and writing......")
begin = time()
self._summary_timestamp = self._extract_timestamp(
summary.event_file_name)
if self._summary_timestamp is None:
raise RuntimeError(
"Cannot extract timestamp from summary filename!"
" It should contains a timestamp after 'summary.' .")
self._save_metadata(summary)
imageid_labels = self._run_inference(summary)
if self._is_saliency_registered:
self._run_saliency(summary, imageid_labels)
print("Finish running and writing. Total time elapsed: {:.3f} s".
format(time() - begin))
def test_raise_exception_with_type_error(self, raise_exception):
summary_dir = tempfile.mkdtemp(dir=self.base_summary_dir)
with pytest.raises(TypeError) as exc:
with SummaryRecord(log_dir=summary_dir,
raise_exception=raise_exception):
pass
assert "raise_exception" in str(exc.value)
def run_case(cls, net):
""" run_case """
net.set_train()
steps = 10
with SummaryRecord(cls.summary_dir) as test_writer:
for i in range(1, steps):
net()
test_writer.record(i)
def test_image_summary_sample():
""" test_image_summary_sample """
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_IMAGE") as test_writer:
for i in range(1, 5):
test_data = get_test_data(i)
_cache_summary_tensor_data(test_data)
test_writer.record(i)
test_writer.flush()
def test_summaryrecord_input_invalid_type_dir():
log.debug("begin test_summaryrecord_input_invalid_type_dir")
# step 0: create the thread
try:
SummaryRecord(32)
except:
assert True
else:
assert False
log.debug("finished test_summaryrecord_input_invalid_type_dir")
def test_mulit_layer_directory():
log.debug("begin test_mulit_layer_directory")
# step 0: create the thread
try:
SummaryRecord("./test_temp_summary_event_file/test/t1/")
except:
assert False
else:
assert True
log.debug("finished test_mulit_layer_directory")
def test_summaryrecord_input_null_string():
log.debug("begin test_summaryrecord_input_null_string")
# step 0: create the thread
try:
SummaryRecord("")
except:
assert True
else:
assert False
log.debug("finished test_summaryrecord_input_null_string")
def test_summaryrecord_input_None():
log.debug("begin test_summaryrecord_input_None")
# step 0: create the thread
try:
SummaryRecord(None)
except:
assert True
else:
assert False
log.debug("finished test_summaryrecord_input_None")
def test_summaryrecord_input_relative_dir_1():
log.debug("begin test_summaryrecord_input_relative_dir_1")
# step 0: create the thread
try:
SummaryRecord("./test_temp_summary_event_file/")
except:
assert False
else:
assert True
log.debug("finished test_summaryrecord_input_relative_dir_1")
def test_summaryrecord_input_relative_dir_2():
log.debug("begin test_summaryrecord_input_relative_dir_2")
# step 0: create the thread
try:
SummaryRecord("../summary/")
except:
assert False
else:
assert True
log.debug("finished test_summaryrecord_input_relative_dir_2")
def me_scalar_summary(steps, tag=None, value=None):
with SummaryRecord(SUMMARY_DIR_ME_TEMP) as test_writer:
x = Tensor(np.array([1.1]).astype(np.float32))
y = Tensor(np.array([1.2]).astype(np.float32))
out_me_dict = train_summary_record_scalar_for_1(
test_writer, steps, x, y)
return out_me_dict
def test_graph_summary_callback2():
dataset = get_dataset()
net = Net()
loss = nn.SoftmaxCrossEntropyWithLogits()
optim = Momentum(net.trainable_params(), 0.1, 0.9)
context.set_context(mode=context.GRAPH_MODE)
model = Model(net, loss_fn=loss, optimizer=optim, metrics=None)
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_GRAPH",
network=net) as test_writer:
summary_cb = SummaryStep(test_writer, 1)
model.train(2, dataset, callbacks=summary_cb)
def test_summary():
with tempfile.TemporaryDirectory() as tmp_dir:
steps = 2
with SummaryRecord(tmp_dir) as test_writer:
train_summary_record(test_writer, steps=steps)
file_name = os.path.realpath(test_writer.full_file_name)
with SummaryReader(file_name) as summary_writer:
for _ in range(steps):
event = summary_writer.read_event()
tags = set(value.tag for value in event.summary.value)
assert tags == {'tensor', 'histogram', 'scalar', 'image'}
def test_histogram_summary_empty_tensor():
"""Test histogram summary, input is an empty tensor."""
with tempfile.TemporaryDirectory() as tmp_dir:
with SummaryRecord(tmp_dir, file_suffix="_MS_HISTOGRAM") as test_writer:
test_data = _wrap_test_data(Tensor([]))
_cache_summary_tensor_data(test_data)
test_writer.record(step=1)
file_name = os.path.join(tmp_dir, test_writer.event_file_name)
with SummaryReader(file_name) as reader:
event = reader.read_event()
assert event.summary.value[0].histogram.count == 0
def test_scalar_summary_with_ge_2():
""" test_scalar_summary_with_ge_2 """
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_SCALAR") as test_writer:
net = SummaryDemo()
net.set_train()
steps = 100
for i in range(1, steps):
x = Tensor(np.array([1.1]).astype(np.float32))
y = Tensor(np.array([1.2]).astype(np.float32))
net(x, y)
test_writer.record(i)
def run(self):
"""
Run the explain job and save the result as a summary in summary_dir.
Note:
User should call register_saliency() once before running this function.
Raises:
ValueError: Be raised for any data or settings' value problem.
TypeError: Be raised for any data or settings' type problem.
RuntimeError: Be raised for any runtime problem.
"""
self._verify_data_n_settings(check_all=True)
self._manifest = {
"saliency_map": False,
"benchmark": False,
"uncertainty": False,
"hierarchical_occlusion": False
}
with SummaryRecord(self._summary_dir, raise_exception=True) as summary:
print("Start running and writing......")
begin = time()
self._summary_timestamp = self._extract_timestamp(
summary.event_file_name)
if self._summary_timestamp is None:
raise RuntimeError(
"Cannot extract timestamp from summary filename!"
" It should contains a timestamp after 'summary.' .")
self._save_metadata(summary)
imageid_labels = self._run_inference(summary)
sample_count = self._sample_index
if self._is_saliency_registered:
self._run_saliency(summary, imageid_labels)
if not self._manifest["saliency_map"]:
raise RuntimeError(
f"No saliency map was generated in {sample_count} samples. "
f"Please make sure the dataset, labels, activation function and network are properly trained "
f"and configured.")
if self._is_hoc_registered and not self._manifest[
"hierarchical_occlusion"]:
raise RuntimeError(
f"No Hierarchical Occlusion result was found in {sample_count} samples. "
f"Please make sure the dataset, labels, activation function and network are properly trained "
f"and configured.")
self._save_manifest()
print("Finish running and writing. Total time elapsed: {:.3f} s".
format(time() - begin))
def test_scalar_summary_with_ge_2():
""" test_scalar_summary_with_ge_2 """
log.debug("begin test_scalar_summary_with_ge_2")
# step 0: create the thread
test_writer = SummaryRecord(SUMMARY_DIR, file_suffix="_MS_SCALAR")
# step 1: create the network for summary
x = Tensor(np.array([1.1]).astype(np.float32))
y = Tensor(np.array([1.2]).astype(np.float32))
net = SummaryDemo()
net.set_train()
# step 2: create the Event
steps = 100
for i in range(1, steps):
x = Tensor(np.array([1.1]).astype(np.float32))
y = Tensor(np.array([1.2]).astype(np.float32))
net(x, y)
test_writer.record(i)
# step 3: close the writer
test_writer.close()
log.debug("finished test_scalar_summary_with_ge_2")
def test_image_summary_train():
""" test_image_summary_train """
dataset = get_dataset()
log.debug("begin test_image_summary_sample")
# step 0: create the thread
test_writer = SummaryRecord(SUMMARY_DIR, file_suffix="_MS_IMAGE")
# step 1: create the test data for summary
# step 2: create the Event
model = get_model()
fn = ImageSummaryCallback(test_writer)
summary_recode = SummaryStep(fn, 1)
model.train(2, dataset, callbacks=summary_recode)
# step 3: send the event to mq
# step 4: accept the event and write the file
test_writer.close()
log.debug("finished test_image_summary_sample")
def test_tensor_summary_sample():
""" test_tensor_summary_sample """
log.debug("begin test_tensor_summary_sample")
# step 0: create the thread
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_TENSOR") as test_writer:
# step 1: create the Event
for i in range(1, 100):
test_data = get_test_data(i)
_cache_summary_tensor_data(test_data)
test_writer.record(i)
# step 2: accept the event and write the file
log.debug("finished test_tensor_summary_sample")
def test_scalar_summary_with_ge():
""" test_scalar_summary_with_ge """
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_SCALAR") as test_writer:
net = SummaryDemo()
net.set_train()
# step 2: create the Event
steps = 100
for i in range(1, steps):
x = Tensor(
np.array([1.1 + random.uniform(1, 10)]).astype(np.float32))
y = Tensor(
np.array([1.2 + random.uniform(1, 10)]).astype(np.float32))
net(x, y)
test_writer.record(i)
def test_validate():
with SummaryRecord(SUMMARY_DIR) as sr:
sr.record(1)
with pytest.raises(ValueError):
sr.record(False)
with pytest.raises(ValueError):
sr.record(2.0)
with pytest.raises(ValueError):
sr.record((1, 3))
with pytest.raises(ValueError):
sr.record([2, 3])
with pytest.raises(ValueError):
sr.record("str")
with pytest.raises(ValueError):
sr.record(sr)
def me_train_tensor(net, input_np, label_np, epoch_size=2):
context.set_context(mode=context.GRAPH_MODE)
loss = SoftmaxCrossEntropyWithLogits(is_grad=False, sparse=True)
opt = ApplyMomentum(Tensor(np.array([0.1])), Tensor(np.array([0.9])),
filter(lambda x: x.requires_grad, net.get_parameters()))
Model(net, loss, opt)
_network = wrap.WithLossCell(net, loss)
_train_net = MsWrapper(wrap.TrainOneStepCell(_network, opt))
_train_net.set_train()
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_GRAPH", network=_train_net) as summary_writer:
for epoch in range(0, epoch_size):
print(f"epoch %d" % (epoch))
output = _train_net(Tensor(input_np), Tensor(label_np))
summary_writer.record(i)
print("********output***********")
print(output.asnumpy())
def test_summary_step2_summary_record1():
"""Test record 10 step summary."""
if platform.system() == "Windows":
# Summary does not support windows currently.
return
with tempfile.TemporaryDirectory() as tmp_dir:
steps = 2
with SummaryRecord(tmp_dir) as test_writer:
train_summary_record(test_writer, steps=steps)
file_name = os.path.realpath(test_writer.full_file_name)
with SummaryReader(file_name) as summary_writer:
for _ in range(steps):
event = summary_writer.read_event()
tags = set(value.tag for value in event.summary.value)
assert tags == {'tensor', 'histogram', 'scalar', 'image'}
def test_histogram_summary_same_value():
"""Test histogram summary, input is an ones tensor."""
with tempfile.TemporaryDirectory() as tmp_dir:
with SummaryRecord(tmp_dir, file_suffix="_MS_HISTOGRAM") as test_writer:
dim1 = 100
dim2 = 100
test_data = _wrap_test_data(Tensor(np.ones([dim1, dim2])))
_cache_summary_tensor_data(test_data)
test_writer.record(step=1)
file_name = os.path.join(tmp_dir, test_writer.event_file_name)
with SummaryReader(file_name) as reader:
event = reader.read_event()
LOG.debug(event)
assert len(event.summary.value[0].histogram.buckets) == _calc_histogram_bins(dim1 * dim2)
def test_histogram_summary_all_nan_inf():
"""Test histogram summary, input tensor has no valid number."""
with tempfile.TemporaryDirectory() as tmp_dir:
with SummaryRecord(tmp_dir, file_suffix="_MS_HISTOGRAM") as test_writer:
test_data = _wrap_test_data(Tensor(np.array([np.nan, np.nan, np.nan, np.inf, -np.inf])))
_cache_summary_tensor_data(test_data)
test_writer.record(step=1)
file_name = os.path.join(tmp_dir, test_writer.event_file_name)
with SummaryReader(file_name) as reader:
event = reader.read_event()
LOG.debug(event)
histogram = event.summary.value[0].histogram
assert histogram.nan_count == 3
assert histogram.pos_inf_count == 1
assert histogram.neg_inf_count == 1
def test_image_summary_data():
""" test_image_summary_data """
dataset = get_dataset()
test_data_list = []
i = 1
for next_element in dataset:
tag = "image_" + str(i) + "[:Image]"
dct = {}
dct["name"] = tag
dct["data"] = Tensor(next_element[0])
test_data_list.append(dct)
i += 1
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_IMAGE") as test_writer:
_cache_summary_tensor_data(test_data_list)
test_writer.record(1)
def test_scalar_summary_sample_with_shape_1():
""" test_scalar_summary_sample_with_shape_1 """
log.debug("begin test_scalar_summary_sample_with_shape_1")
# step 0: create the thread
with SummaryRecord(SUMMARY_DIR, file_suffix="_MS_SCALAR") as test_writer:
# step 1: create the test data for summary
# step 2: create the Event
for i in range(1, 100):
test_data = get_test_data_shape_1(i)
_cache_summary_tensor_data(test_data)
test_writer.record(i)
# step 3: send the event to mq
# step 4: accept the event and write the file
log.debug("finished test_scalar_summary_sample")
