def test_new_errors(self,
number=0,
candidates=lambda: [_dummy_candidate()],
estimator_spec=tu.dummy_estimator_spec(),
best_candidate_index=0,
subnetwork_reports=lambda: []):
with self.assertRaises(ValueError):
_Iteration(number=number,
candidates=candidates(),
subnetwork_specs=None,
estimator_spec=estimator_spec,
best_candidate_index=best_candidate_index,
summaries=[],
subnetwork_reports=subnetwork_reports(),
train_manager=_TrainManager([], [],
self.test_subdirectory,
is_chief=True))
def test_new_errors(self,
number=0,
candidates=lambda: [_dummy_candidate()],
estimator_spec=tu.dummy_estimator_spec(),
best_candidate_index=0,
is_over_fn=lambda: True,
subnetwork_reports=lambda: []):
with self.test_session():
with self.assertRaises(ValueError):
_Iteration(number=number,
candidates=candidates(),
subnetwork_specs=None,
estimator_spec=estimator_spec,
best_candidate_index=best_candidate_index,
summaries=[],
is_over_fn=is_over_fn,
subnetwork_reports=subnetwork_reports())
class IterationTest(parameterized.TestCase, tf.test.TestCase):
# pylint: disable=g-long-lambda
@parameterized.named_parameters(
{
"testcase_name": "single_candidate",
"number": 0,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"is_over_fn": lambda: True,
}, {
"testcase_name": "two_candidates",
"number": 0,
"candidates": [_dummy_candidate(),
_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"is_over_fn": lambda: True,
}, {
"testcase_name": "positive_number",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"is_over_fn": lambda: True,
}, {
"testcase_name": "false_is_over",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"is_over_fn": lambda: False,
}, {
"testcase_name": "zero_best_predictions",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"is_over_fn": lambda: True,
}, {
"testcase_name": "zero_best_loss",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"is_over_fn": lambda: True,
}, {
"testcase_name": "pass_subnetwork_report",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"is_over_fn": lambda: True,
"subnetwork_reports_fn": lambda: {
"foo":
SubnetworkReport(hparams={"dropout": 1.0},
attributes={"aoo": tf.constant("aoo")},
metrics={
"moo":
(tf.constant("moo1"), tf.constant("moo2"))
})
},
})
def test_new(self,
number,
candidates,
estimator_spec,
best_candidate_index,
is_over_fn,
subnetwork_reports_fn=None,
step=0):
if subnetwork_reports_fn is None:
subnetwork_reports = {}
else:
subnetwork_reports = subnetwork_reports_fn()
with self.test_session():
iteration = _Iteration(number=number,
candidates=candidates,
estimator_spec=estimator_spec,
best_candidate_index=best_candidate_index,
summaries=[],
is_over_fn=is_over_fn,
subnetwork_reports=subnetwork_reports,
step=step)
self.assertEqual(iteration.number, number)
self.assertEqual(iteration.candidates, candidates)
self.assertEqual(iteration.estimator_spec, estimator_spec)
self.assertEqual(iteration.best_candidate_index,
best_candidate_index)
self.assertEqual(iteration.is_over_fn(), is_over_fn())
self.assertEqual(iteration.subnetwork_reports, subnetwork_reports)
self.assertEqual(iteration.step, step)
@parameterized.named_parameters(
{
"testcase_name": "negative_number",
"number": -1,
}, {
"testcase_name": "float_number",
"number": 1.213,
}, {
"testcase_name": "none_number",
"number": None,
}, {
"testcase_name": "empty_candidates",
"candidates": lambda: [],
}, {
"testcase_name": "none_candidates",
"candidates": lambda: None,
}, {
"testcase_name": "non_list_candidates",
"candidates": lambda: {
"foo": _dummy_candidate()
},
}, {
"testcase_name": "none_estimator_spec",
"estimator_spec": None,
}, {
"testcase_name": "none_best_candidate_index",
"best_candidate_index": None,
}, {
"testcase_name": "none_subnetwork_reports",
"subnetwork_reports": lambda: None,
}, {
"testcase_name": "none_step",
"step": None,
})
def test_new_errors(self,
number=0,
candidates=lambda: [_dummy_candidate()],
estimator_spec=tu.dummy_estimator_spec(),
best_candidate_index=0,
is_over_fn=lambda: True,
subnetwork_reports=lambda: [],
step=0):
with self.test_session():
with self.assertRaises(ValueError):
_Iteration(number=number,
candidates=candidates(),
estimator_spec=estimator_spec,
best_candidate_index=best_candidate_index,
summaries=[],
is_over_fn=is_over_fn,
subnetwork_reports=subnetwork_reports(),
step=step)
class IterationTest(tu.AdanetTestCase):
# pylint: disable=g-long-lambda
@parameterized.named_parameters(
{
"testcase_name": "single_candidate",
"number": 0,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
}, {
"testcase_name": "two_candidates",
"number": 0,
"candidates": [_dummy_candidate(),
_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
}, {
"testcase_name": "positive_number",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
}, {
"testcase_name": "zero_best_predictions",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
}, {
"testcase_name": "zero_best_loss",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
}, {
"testcase_name": "pass_subnetwork_report",
"number": 1,
"candidates": [_dummy_candidate()],
"estimator_spec": tu.dummy_estimator_spec(),
"best_candidate_index": 0,
"subnetwork_reports_fn": lambda: {
"foo":
SubnetworkReport(hparams={"dropout": 1.0},
attributes={"aoo": tf.constant("aoo")},
metrics={
"moo":
(tf.constant("moo1"), tf.constant("moo2"))
})
},
})
@test_util.run_in_graph_and_eager_modes
def test_new(self,
number,
candidates,
estimator_spec,
best_candidate_index,
subnetwork_reports_fn=None):
if subnetwork_reports_fn is None:
subnetwork_reports = {}
else:
subnetwork_reports = subnetwork_reports_fn()
iteration = _Iteration(number=number,
candidates=candidates,
subnetwork_specs=None,
estimator_spec=estimator_spec,
best_candidate_index=best_candidate_index,
summaries=[],
subnetwork_reports=subnetwork_reports,
train_manager=_TrainManager(
[], [],
self.test_subdirectory,
is_chief=True))
self.assertEqual(iteration.number, number)
self.assertEqual(iteration.candidates, candidates)
self.assertEqual(iteration.estimator_spec, estimator_spec)
self.assertEqual(iteration.best_candidate_index, best_candidate_index)
self.assertEqual(iteration.subnetwork_reports, subnetwork_reports)
@parameterized.named_parameters(
{
"testcase_name": "negative_number",
"number": -1,
}, {
"testcase_name": "float_number",
"number": 1.213,
}, {
"testcase_name": "none_number",
"number": None,
}, {
"testcase_name": "empty_candidates",
"candidates": lambda: [],
}, {
"testcase_name": "none_candidates",
"candidates": lambda: None,
}, {
"testcase_name": "non_list_candidates",
"candidates": lambda: {
"foo": _dummy_candidate()
},
}, {
"testcase_name": "none_estimator_spec",
"estimator_spec": None,
}, {
"testcase_name": "none_best_candidate_index",
"best_candidate_index": None,
}, {
"testcase_name": "none_subnetwork_reports",
"subnetwork_reports": lambda: None,
})
@test_util.run_in_graph_and_eager_modes
def test_new_errors(self,
number=0,
candidates=lambda: [_dummy_candidate()],
estimator_spec=tu.dummy_estimator_spec(),
best_candidate_index=0,
subnetwork_reports=lambda: []):
with self.assertRaises(ValueError):
_Iteration(number=number,
candidates=candidates(),
subnetwork_specs=None,
estimator_spec=estimator_spec,
best_candidate_index=best_candidate_index,
summaries=[],
subnetwork_reports=subnetwork_reports(),
train_manager=_TrainManager([], [],
self.test_subdirectory,
is_chief=True))
