def executor_test_settings(func):
if hu.is_sandcastle() or hu.is_travis():
return settings(
max_examples=CI_MAX_EXAMPLES,
timeout=CI_TIMEOUT
)(func)
else:
return func
def executor_test_settings(func):
if hu.is_sandcastle() or hu.is_travis():
return hu.settings(
max_examples=CI_MAX_EXAMPLES,
deadline=CI_TIMEOUT * 1000 # deadline is in ms
)(func)
else:
return func
def executor_test_model_names():
if not hu.is_travis():
return conv_model_generators().keys()
else:
return ["MLP"]
def executor_test_model_names():
if hu.is_sandcastle() or hu.is_travis():
return ["MLP"]
else:
return conv_model_generators().keys()
import unittest
EXECUTORS = ["dag", "async_dag"]
ITERATIONS = 2
CI_MAX_EXAMPLES = 2
CI_TIMEOUT = 600
def test_settings(func):
if hu.is_sandcastle() or hu.is_travis():
return settings(max_examples=CI_MAX_EXAMPLES, timeout=CI_TIMEOUT)(func)
else:
return func
@unittest.skipIf(hu.is_travis(), "Disabled in Travis")
class ExecutorCPUConvNetTest(ExecutorTestBase):
@given(executor=st.sampled_from(EXECUTORS),
model_name=st.sampled_from(conv_model_generators().keys()),
batch_size=st.sampled_from([8]),
num_workers=st.sampled_from([8]))
@test_settings
def test_executor(self, executor, model_name, batch_size, num_workers):
model = build_conv_model(model_name, batch_size)
model.Proto().num_workers = num_workers
def run_model():
iterations = ITERATIONS
if model_name == "MLP":
iterations = 1 # avoid numeric instability with MLP gradients
workspace.RunNet(model.net, iterations)
def executor_test_model_names():
if hu.is_sandcastle() or hu.is_travis():
return ["MLP"]
else:
return conv_model_generators().keys()