def _joint_linear_estimator(head, feature_columns):
return estimator.Estimator(model_fn=_base_model_fn,
params={
'model':
composable_model.LinearComposableModel(
num_label_columns=head.logits_dimension,
_joint_weights=True),
'feature_columns':
feature_columns,
'head':
head
})
def __init__(self,
target_column,
model_dir=None,
linear_feature_columns=None,
linear_optimizer=None,
dnn_feature_columns=None,
dnn_optimizer=None,
dnn_hidden_units=None,
dnn_activation_fn=nn.relu,
dnn_dropout=None,
gradient_clip_norm=None,
enable_centered_bias=True,
config=None):
"""Initializes a _DNNLinearCombinedBaseEstimator instance.
Args:
target_column: A _TargetColumn object.
model_dir: Directory to save model parameters, graph and etc.
linear_feature_columns: An iterable containing all the feature columns
used by linear part of the model. All items in the set should be
instances of classes derived from `FeatureColumn`.
linear_optimizer: An instance of `tf.Optimizer` used to apply gradients to
the linear part of the model. If `None`, will use a FTRL optimizer.
dnn_feature_columns: An iterable containing all the feature columns used
by deep part of the model. All items in the set should be instances of
classes derived from `FeatureColumn`.
dnn_optimizer: An instance of `tf.Optimizer` used to apply gradients to
the deep part of the model. If `None`, will use an Adagrad optimizer.
dnn_hidden_units: List of hidden units per layer. All layers are fully
connected.
dnn_activation_fn: Activation function applied to each layer. If `None`,
will use `tf.nn.relu`.
dnn_dropout: When not None, the probability we will drop out
a given coordinate.
gradient_clip_norm: A float > 0. If provided, gradients are clipped
to their global norm with this clipping ratio. See
tf.clip_by_global_norm for more details.
enable_centered_bias: A bool. If True, estimator will learn a centered
bias variable for each class. Rest of the model structure learns the
residual after centered bias.
config: RunConfig object to configure the runtime settings.
Raises:
ValueError: If both linear_feature_columns and dnn_features_columns are
empty at the same time.
"""
super(_DNNLinearCombinedBaseEstimator,
self).__init__(model_dir=model_dir, config=config)
num_ps_replicas = config.num_ps_replicas if config else 0
self._linear_model = composable_model.LinearComposableModel(
num_label_columns=target_column.num_label_columns,
optimizer=linear_optimizer,
gradient_clip_norm=gradient_clip_norm,
num_ps_replicas=num_ps_replicas)
self._dnn_model = composable_model.DNNComposableModel(
num_label_columns=target_column.num_label_columns,
hidden_units=dnn_hidden_units,
optimizer=dnn_optimizer,
activation_fn=dnn_activation_fn,
dropout=dnn_dropout,
gradient_clip_norm=gradient_clip_norm,
num_ps_replicas=num_ps_replicas) if dnn_hidden_units else None
self._linear_feature_columns = linear_feature_columns
self._linear_optimizer = linear_optimizer
self._linear_weight_collection = (
self._linear_model.get_weight_collection_name())
self._dnn_feature_columns = dnn_feature_columns
self._dnn_hidden_units = dnn_hidden_units
self._centered_bias_weight_collection = "centered_bias"
self._enable_centered_bias = enable_centered_bias
self._target_column = target_column
def __init__(
self, # _joint_linear_weights pylint: disable=invalid-name
head,
model_dir=None,
linear_feature_columns=None,
linear_optimizer=None,
_joint_linear_weights=False,
dnn_feature_columns=None,
dnn_optimizer=None,
dnn_hidden_units=None,
dnn_activation_fn=nn.relu,
dnn_dropout=None,
gradient_clip_norm=None,
config=None,
feature_engineering_fn=None,
default_prediction_key=None,
enable_centered_bias=False):
"""Initializes a _DNNLinearCombinedBaseEstimator instance.
Args:
head: A _Head object.
model_dir: Directory to save model parameters, graph and etc. This can
also be used to load checkpoints from the directory into a estimator
to continue training a previously saved model.
linear_feature_columns: An iterable containing all the feature columns
used by linear part of the model. All items in the set should be
instances of classes derived from `FeatureColumn`.
linear_optimizer: An instance of `tf.Optimizer` used to apply gradients to
the linear part of the model. If `None`, will use a FTRL optimizer.
_joint_linear_weights: If True will use a single (possibly partitioned)
variable to store all weights for the linear model. More efficient if
there are many columns, however requires all columns are sparse and
have the 'sum' combiner.
dnn_feature_columns: An iterable containing all the feature columns used
by deep part of the model. All items in the set should be instances of
classes derived from `FeatureColumn`.
dnn_optimizer: An instance of `tf.Optimizer` used to apply gradients to
the deep part of the model. If `None`, will use an Adagrad optimizer.
dnn_hidden_units: List of hidden units per layer. All layers are fully
connected.
dnn_activation_fn: Activation function applied to each layer. If `None`,
will use `tf.nn.relu`.
dnn_dropout: When not None, the probability we will drop out
a given coordinate.
gradient_clip_norm: A float > 0. If provided, gradients are clipped
to their global norm with this clipping ratio. See
tf.clip_by_global_norm for more details.
config: RunConfig object to configure the runtime settings.
feature_engineering_fn: Feature engineering function. Takes features and
labels which are the output of `input_fn` and
returns features and labels which will be fed
into the model.
default_prediction_key: Default prediction key to use with metrics.
enable_centered_bias: A bool. If True, estimator will learn a centered
bias variable for each class. Rest of the model structure learns the
residual after centered bias.
Raises:
ValueError: If both linear_feature_columns and dnn_features_columns are
empty at the same time.
"""
super(_DNNLinearCombinedBaseEstimator,
self).__init__(model_dir=model_dir, config=config)
num_ps_replicas = config.num_ps_replicas if config else 0
self._linear_model = composable_model.LinearComposableModel(
num_label_columns=head.logits_dimension,
optimizer=linear_optimizer,
_joint_weights=_joint_linear_weights,
gradient_clip_norm=gradient_clip_norm,
num_ps_replicas=num_ps_replicas)
self._dnn_model = composable_model.DNNComposableModel(
num_label_columns=head.logits_dimension,
hidden_units=dnn_hidden_units,
optimizer=dnn_optimizer,
activation_fn=dnn_activation_fn,
dropout=dnn_dropout,
gradient_clip_norm=gradient_clip_norm,
num_ps_replicas=num_ps_replicas) if dnn_hidden_units else None
self._linear_feature_columns = linear_feature_columns
self._linear_optimizer = linear_optimizer
self._dnn_feature_columns = dnn_feature_columns
self._dnn_hidden_units = dnn_hidden_units
self._head = head
self._default_prediction_key = default_prediction_key
self._feature_engineering_fn = (feature_engineering_fn
or (lambda features, labels:
(features, labels)))
self._enable_centered_bias = enable_centered_bias
def __init__(self, head, feature_columns):
super(JointLinearEstimator, self).__init__(head, feature_columns)
self._model = composable_model.LinearComposableModel(
num_label_columns=head.logits_dimension, _joint_weights=True)
def __init__(self, target_column, feature_columns):
super(JointLinearEstimator, self).__init__(target_column,
feature_columns)
self._model = composable_model.LinearComposableModel(
num_label_columns=target_column.num_label_columns,
_joint_weights=True)
