def tearDown(self):
from dxl.learn.backend import current_backend, TensorFlow
from dxl.learn.core.config import clear_config
from dxl.learn.core import SubgraphMakerFactory, SubgraphMakerTable
if isinstance(current_backend(), TensorFlow):
current_backend().unbox().reset_default_graph()
clear_config()
SubgraphMakerFactory.reset()
SubgraphMakerTable.reset()
def kernel(self, inputs=None):
KT, KC, KS = self.KEYS.TENSOR, self.KEYS.CONFIG, self.KEYS.GRAPH
self.tensors[KT.LEARNING_RATE] = NotTrainableVariable(
self.info.child_tensor(KT.LEARNING_RATE), [],
current_backend().float32, self.config(KT.LEARNING_RATE))
self.tensors[KT.DECAY_LEARNING_RATE] = self.tensors[
KT.LEARNING_RATE].assign(self.tensors[KT.LEARNING_RATE] *
self.config(KC.DECAY_RATIO))
self.tensors[KS.OPTIMIZER] = self._get_optimizer(
self.config(KC.OPTIMIZER_NAME))
def sandbox():
from dxl.learn.backend import current_backend
return current_backend().sandbox
def sandbox():
with current_backend().sandbox():
yield
class TestCase(current_backend().TestCase()):
def setUp(self):
pass
def tearDown(self):
from dxl.learn.backend import current_backend, TensorFlow
from dxl.learn.core.config import clear_config
from dxl.learn.core import SubgraphMakerFactory, SubgraphMakerTable
if isinstance(current_backend(), TensorFlow):
current_backend().unbox().reset_default_graph()
clear_config()
SubgraphMakerFactory.reset()
SubgraphMakerTable.reset()
def make_dummy_tensor(self, info=None):
from dxl.learn.core import Constant
if info is None:
info = str(uuid.uuid4())
return Constant(0.0, info)
def make_dummy_variable(self, info=None):
from dxl.learn.core import Variable
if info is None:
info == str(uuid.uuid4())
return Variable(info, [])
@property
def resource_path(self):
from .resource import test_resource_path
# return Path(os.getenv('DEV_DXLEARN_TEST_RESOURCE_PATH'))
return test_resource_path
def assertFloatArrayEqual(self, first, second, msg=None):
if msg is None:
msg = ''
return np.testing.assert_array_almost_equal(
np.array(first), np.array(second), err_msg=msg)
def assertNameEqual(self, first, second, with_strip_colon_and_index=True):
names = map(get_object_name_str, [first, second])
if with_strip_colon_and_index:
names = map(name_str_without_colon_and_index, names)
names = list(names)
self.assertEqual(names[0], names[1], 'Name not equal.')
@pytest.mark.skip
# @unittest.skip
@contextmanager
def test_session(self):
from dxl.learn.core.session import TestSession
with super().test_session() as sess:
yield TestSession(sess)
@contextmanager
def variables_initialized_test_session(self):
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
yield sess
@contextmanager
def graph_on_cpu(self):
with tf.device('/cpu:0'):
with tf.Graph().as_default() as g:
yield
def test_default_current_backend(self):
self.assertIsInstance(current_backend(), TensorFlow)