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 build_subnetwork_spec(self,
name,
subnetwork_builder,
summary,
features,
mode,
labels=None,
previous_ensemble=None,
params=None):
del summary
del features
del mode
del labels
del previous_ensemble
del params
is_training = False
if subnetwork_builder:
is_training = "training" in subnetwork_builder.name
return _SubnetworkSpec(
name=name,
subnetwork=None,
builder=subnetwork_builder,
step=tf.Variable(0.),
is_training=is_training,
predictions=None,
loss=None,
train_op=subnetwork_builder.build_subnetwork_train_op(
*[None for _ in range(7)]),
eval_metrics=None)
def build_subnetwork_spec(self,
name,
subnetwork_builder,
iteration_step,
summary,
features,
mode,
labels=None,
previous_ensemble=None,
params=None):
del iteration_step
del summary
del features
del mode
del labels
del previous_ensemble
del params
return _SubnetworkSpec(
name=name,
subnetwork=None,
builder=subnetwork_builder,
predictions=None,
loss=None,
train_op=subnetwork_builder.build_subnetwork_train_op(
*[None for _ in range(7)]),
eval_metrics=None)
def create_iteration_metrics(subnetwork_metrics=None,
ensemble_metrics=None,
use_tpu=False,
iteration_number=1):
"""Creates an instance of the _IterationMetrics class.
Args:
subnetwork_metrics: List of _SubnetworkMetrics objects.
ensemble_metrics: List of _EnsembleMetrics objects.
use_tpu: Whether to use TPU-specific variable sharing logic.
iteration_number: What number iteration these metrics are for.
Returns:
An instance of _IterationMetrics that has been populated with the
input metrics.
"""
subnetwork_metrics = subnetwork_metrics or []
ensemble_metrics = ensemble_metrics or []
candidates = []
for i, metric in enumerate(ensemble_metrics):
spec = _EnsembleSpec(name="ensemble_{}".format(i),
ensemble=None,
architecture=None,
subnetwork_builders=None,
predictions=None,
step=None,
variables=None,
eval_metrics=metric)
candidate = _Candidate(ensemble_spec=spec,
adanet_loss=tf.constant(i),
variables=None)
candidates.append(candidate)
subnetwork_specs = []
for i, metric in enumerate(subnetwork_metrics):
spec = _SubnetworkSpec(name="subnetwork_{}".format(i),
subnetwork=None,
builder=None,
predictions=None,
step=None,
loss=None,
train_op=None,
asset_dir=None,
eval_metrics=metric,
variables=None)
subnetwork_specs.append(spec)
return _IterationMetrics(iteration_number,
candidates,
subnetwork_specs=subnetwork_specs,
use_tpu=use_tpu)
