def build_subnetwork_spec(self,
name,
subnetwork_builder,
summary,
features,
mode,
labels=None,
previous_ensemble=None,
config=None,
params=None):
del summary
del features
del mode
del labels
del previous_ensemble
del params
del config
return _SubnetworkSpec(
name=name,
subnetwork=None,
builder=subnetwork_builder,
step=tf.Variable(0.),
predictions=None,
loss=None,
train_op=subnetwork_builder.build_subnetwork_train_op(
*[None for _ in range(7)]),
eval_metrics=tu.create_subnetwork_metrics(
metric_fn=lambda: {"a": (tf.constant(1), tf.constant(1))}))
def test_metric_ops_not_duplicated_on_cpu(self):
with context.graph_mode():
self.setup_graph()
metric_fn = lambda: {"metric": (tf.constant(5), tf.constant(5))}
best_candidate_index = 3
mode = tf.estimator.ModeKeys.EVAL
ensemble_metrics = tu.create_ensemble_metrics(metric_fn)
subnetwork_metrics = tu.create_subnetwork_metrics(metric_fn)
iteration_metrics = tu.create_iteration_metrics(
ensemble_metrics=[ensemble_metrics],
subnetwork_metrics=[subnetwork_metrics])
ensemble_ops1 = ensemble_metrics.eval_metrics_ops()
ensemble_ops2 = ensemble_metrics.eval_metrics_ops()
subnetwork_ops1 = subnetwork_metrics.eval_metrics_ops()
subnetwork_ops2 = subnetwork_metrics.eval_metrics_ops()
iteration_ops1 = iteration_metrics.best_eval_metric_ops(
best_candidate_index, mode)
iteration_ops2 = iteration_metrics.best_eval_metric_ops(
best_candidate_index, mode)
self.assertEqual(subnetwork_ops1, subnetwork_ops2)
self.assertEqual(ensemble_ops1, ensemble_ops2)
self.assertEqual(iteration_ops1, iteration_ops2)
for ops in [ensemble_ops1, subnetwork_ops1, iteration_ops1]:
self.assertIsNotNone(ops)
def test_subnetwork_metrics(self, use_tpu):
with context.graph_mode():
self.setup_graph()
spec = self._estimator_spec
if not use_tpu:
spec = spec.as_estimator_spec()
metrics = tu.create_subnetwork_metrics(self._metric_fn,
use_tpu=use_tpu,
features=self._features,
labels=self._labels,
estimator_spec=spec)
actual = self._run_metrics(metrics.eval_metrics_tuple())
expected = {"loss": 2., "metric_1": 1., "metric_2": 2.}
self.assertEqual(actual, expected)
def test_subnetwork_metrics_user_metric_fn_overrides_metrics(self):
with context.graph_mode():
self.setup_graph()
overridden_value = 100.
def _overriding_metric_fn():
value = tf.constant(overridden_value)
return {"metric_1": tf_compat.v1.metrics.mean(value)}
metrics = tu.create_subnetwork_metrics(
_overriding_metric_fn,
features=self._features,
labels=self._labels,
estimator_spec=self._estimator_spec)
actual = self._run_metrics(metrics.eval_metrics_tuple())
expected = {"loss": 2., "metric_1": overridden_value}
self.assertEqual(actual, expected)