def _get_default_head(params, weights_name, output_type, name=None):
"""Creates a default head based on a type of a problem."""
if output_type == ModelBuilderOutputType.MODEL_FN_OPS:
if params.regression:
return head_lib.regression_head(
weight_column_name=weights_name,
label_dimension=params.num_outputs,
enable_centered_bias=False,
head_name=name)
else:
return head_lib.multi_class_head(
params.num_classes,
weight_column_name=weights_name,
enable_centered_bias=False,
head_name=name)
else:
if params.regression:
return core_head_lib._regression_head( # pylint:disable=protected-access
weight_column=weights_name,
label_dimension=params.num_outputs,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_NONZERO_WEIGHTS)
else:
return core_head_lib._multi_class_head_with_softmax_cross_entropy_loss( # pylint:disable=protected-access
n_classes=params.num_classes,
weight_column=weights_name,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_NONZERO_WEIGHTS)
def _get_default_head(params, weights_name, output_type, name=None):
"""Creates a default head based on a type of a problem."""
if output_type == ModelBuilderOutputType.MODEL_FN_OPS:
if params.regression:
return head_lib.regression_head(weight_column_name=weights_name,
label_dimension=params.num_outputs,
enable_centered_bias=False,
head_name=name)
else:
return head_lib.multi_class_head(params.num_classes,
weight_column_name=weights_name,
enable_centered_bias=False,
head_name=name)
else:
if params.regression:
return core_head_lib.regression_head(
weight_column=weights_name,
label_dimension=params.num_outputs,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE)
else:
if params.num_classes == 2:
return core_head_lib.binary_classification_head(
weight_column=weights_name,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE)
else:
return core_head_lib.multi_class_head(
n_classes=params.num_classes,
weight_column=weights_name,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE)
def _get_default_head(params, weights_name, output_type, name=None):
"""Creates a default head based on a type of a problem."""
if output_type == ModelBuilderOutputType.MODEL_FN_OPS:
if params.regression:
return head_lib.regression_head(weight_column_name=weights_name,
label_dimension=params.num_outputs,
enable_centered_bias=False,
head_name=name)
else:
return head_lib.multi_class_head(params.num_classes,
weight_column_name=weights_name,
enable_centered_bias=False,
head_name=name)
else:
if params.regression:
return core_head_lib._regression_head( # pylint:disable=protected-access
weight_column=weights_name,
label_dimension=params.num_outputs,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_NONZERO_WEIGHTS)
else:
return core_head_lib._multi_class_head_with_softmax_cross_entropy_loss( # pylint:disable=protected-access
n_classes=params.num_classes,
weight_column=weights_name,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_NONZERO_WEIGHTS)
def _get_default_head(params, weights_name, output_type, name=None):
"""Creates a default head based on a type of a problem."""
if output_type == ModelBuilderOutputType.MODEL_FN_OPS:
if params.regression:
return head_lib.regression_head(
weight_column_name=weights_name,
label_dimension=params.num_outputs,
enable_centered_bias=False,
head_name=name)
else:
return head_lib.multi_class_head(
params.num_classes,
weight_column_name=weights_name,
enable_centered_bias=False,
head_name=name)
else:
if params.regression:
return core_head_lib.regression_head(
weight_column=weights_name,
label_dimension=params.num_outputs,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE)
else:
if params.num_classes == 2:
return core_head_lib.binary_classification_head(
weight_column=weights_name,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE)
else:
return core_head_lib.multi_class_head(
n_classes=params.num_classes,
weight_column=weights_name,
name=name,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE)
def __init__(self,
learner_config,
examples_per_layer,
label_dimension=1,
num_trees=None,
feature_columns=None,
label_name=None,
weight_column_name=None,
model_dir=None,
config=None,
feature_engineering_fn=None,
logits_modifier_function=None,
center_bias=True):
"""Initializes a GradientBoostedDecisionTreeRegressor estimator instance.
Args:
learner_config: A config for the learner.
examples_per_layer: Number of examples to accumulate before growing a
layer. It can also be a function that computes the number of examples
based on the depth of the layer that's being built.
label_dimension: Number of regression labels per example. This is the size
of the last dimension of the labels `Tensor` (typically, this has shape
`[batch_size, label_dimension]`).
num_trees: An int, number of trees to build.
feature_columns: A list of feature columns.
label_name: String, name of the key in label dict. Can be null if label
is a tensor (single headed models).
weight_column_name: Name of the column for weights, or None if not
weighted.
model_dir: Directory for model exports, etc.
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.
logits_modifier_function: A modifier function for the logits.
center_bias: Whether a separate tree should be created for first fitting
the bias.
"""
head = head_lib.regression_head(
label_name=label_name,
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=False)
super(GradientBoostedDecisionTreeRegressor, self).__init__(
model_fn=model.model_builder,
params={
'head': head,
'feature_columns': feature_columns,
'learner_config': learner_config,
'num_trees': num_trees,
'weight_column_name': weight_column_name,
'examples_per_layer': examples_per_layer,
'logits_modifier_function': logits_modifier_function,
'center_bias': center_bias,
},
model_dir=model_dir,
config=config,
feature_engineering_fn=feature_engineering_fn)
def build_estimator(config, hidden_units=None):
# Wide columns and deep columns.
wide_columns = WIDE_COLUMNS
deep_columns = DEEP_COLUMNS
heads = multi_head([
regression_head(head_name="g_wvec", label_name="g_wvec", label_dimension=2),
regression_head(head_name="g_dvec", label_name="g_dvec", label_dimension=2),
])
return DNNLinearCombinedEstimator(
heads,
config=config,
linear_feature_columns=wide_columns,
dnn_feature_columns=deep_columns,
dnn_hidden_units=hidden_units or [12],
fix_global_step_increment_bug=True
)
def __init__(self,
learner_config,
examples_per_layer,
label_dimension=1,
num_trees=None,
feature_columns=None,
label_name=None,
weight_column_name=None,
model_dir=None,
config=None,
feature_engineering_fn=None,
center_bias=True):
"""Initializes a GradientBoostedDecisionTreeRegressor estimator instance.
Args:
learner_config: A config for the learner.
examples_per_layer: Number of examples to accumulate before growing a
layer. It can also be a function that computes the number of examples
based on the depth of the layer that's being built.
label_dimension: Number of regression labels per example. This is the size
of the last dimension of the labels `Tensor` (typically, this has shape
`[batch_size, label_dimension]`).
num_trees: An int, number of trees to build.
feature_columns: A list of feature columns.
label_name: String, name of the key in label dict. Can be null if label
is a tensor (single headed models).
weight_column_name: Name of the column for weights, or None if not
weighted.
model_dir: Directory for model exports, etc.
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.
center_bias: Whether a separate tree should be created for first fitting
the bias.
"""
head = head_lib.regression_head(
label_name=label_name,
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=False)
super(GradientBoostedDecisionTreeRegressor, self).__init__(
model_fn=model.model_builder,
params={
'head': head,
'feature_columns': feature_columns,
'learner_config': learner_config,
'num_trees': num_trees,
'weight_column_name': weight_column_name,
'examples_per_layer': examples_per_layer,
'center_bias': center_bias
},
model_dir=model_dir,
config=config,
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
example_id_column,
feature_columns,
weight_column_name=None,
model_dir=None,
l1_regularization=0.0,
l2_regularization=1.0,
num_loss_partitions=None,
config=None,
feature_engineering_fn=None,
partitioner=None):
"""Construct a `SDCALinearRegressor` estimator object.
Args:
example_id_column: A string defining the feature column name representing
example ids. Used to initialize the underlying SDCA optimizer.
feature_columns: An iterable containing all the feature columns used by
the model. All items in the iterable should derive from `FeatureColumn`.
Note that the order of the items is ignored at model construction time.
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
model_dir: Directory to save model parameters, graph etc. This can also be
used to load checkpoints from the directory into an estimator to
continue training a previously saved model.
l1_regularization: L1-regularization parameter. Refers to global L1
regularization (across all examples).
l2_regularization: L2-regularization parameter. Refers to global L2
regularization (across all examples).
num_loss_partitions: number of partitions of the (global) loss function
optimized by the underlying optimizer (SDCAOptimizer).
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.
partitioner: Variable partitioner for the primal weights (`div`
partitioning strategy will be used).
Returns:
A `SDCALinearRegressor` estimator.
"""
super(SDCALinearRegressor,
self).__init__(example_id_column=example_id_column,
feature_columns=feature_columns,
weight_column_name=weight_column_name,
model_dir=model_dir,
head=head_lib.regression_head(
weight_column_name=weight_column_name),
l1_regularization=l1_regularization,
l2_regularization=l2_regularization,
num_loss_partitions=num_loss_partitions,
config=config,
feature_engineering_fn=None,
partitioner=partitioner)
def get_default_head(params, weights_name, name=None):
if params.regression:
return head_lib.regression_head(weight_column_name=weights_name,
label_dimension=params.num_outputs,
enable_centered_bias=False,
head_name=name)
else:
return head_lib.multi_class_head(params.num_classes,
weight_column_name=weights_name,
enable_centered_bias=False,
head_name=name)
def __init__(self,
example_id_column,
feature_columns,
weight_column_name=None,
model_dir=None,
l1_regularization=0.0,
l2_regularization=1.0,
num_loss_partitions=None,
config=None,
feature_engineering_fn=None,
partitioner=None):
"""Construct a `SDCALinearRegressor` estimator object.
Args:
example_id_column: A string defining the feature column name representing
example ids. Used to initialize the underlying SDCA optimizer.
feature_columns: An iterable containing all the feature columns used by
the model. All items in the iterable should derive from `FeatureColumn`.
Note that the order of the items is ignored at model construction time.
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
model_dir: Directory to save model parameters, graph etc. This can also be
used to load checkpoints from the directory into an estimator to
continue training a previously saved model.
l1_regularization: L1-regularization parameter. Refers to global L1
regularization (across all examples).
l2_regularization: L2-regularization parameter. Refers to global L2
regularization (across all examples).
num_loss_partitions: number of partitions of the (global) loss function
optimized by the underlying optimizer (SDCAOptimizer).
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.
partitioner: Variable partitioner for the primal weights (`div`
partitioning strategy will be used).
Returns:
A `SDCALinearRegressor` estimator.
"""
super(SDCALinearRegressor, self).__init__(
example_id_column=example_id_column,
feature_columns=feature_columns,
weight_column_name=weight_column_name,
model_dir=model_dir,
head=head_lib.regression_head(weight_column_name=weight_column_name),
l1_regularization=l1_regularization,
l2_regularization=l2_regularization,
num_loss_partitions=num_loss_partitions,
config=config,
feature_engineering_fn=None,
partitioner=partitioner)
def get_default_head(params, weights_name, name=None):
if params.regression:
return head_lib.regression_head(
weight_column_name=weights_name,
label_dimension=params.num_outputs,
enable_centered_bias=False,
head_name=name)
else:
return head_lib.multi_class_head(
params.num_classes,
weight_column_name=weights_name,
enable_centered_bias=False,
head_name=name)
def __init__(self,
model_dir=None,
label_dimension=1,
weight_column_name=None,
config=None,
feature_engineering_fn=None):
"""Initializes a DebugRegressor instance.
Args:
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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
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.
Returns:
A `DebugRegressor` estimator.
"""
params = {
"head":
head_lib.regression_head(
weight_column_name=weight_column_name,
label_dimension=label_dimension,
enable_centered_bias=True)
}
super(DebugRegressor, self).__init__(
model_fn=debug_model_fn,
model_dir=model_dir,
config=config,
params=params,
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
model_dir=None,
label_dimension=1,
weight_column_name=None,
config=None,
feature_engineering_fn=None):
"""Initializes a DebugRegressor instance.
Args:
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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
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.
Returns:
A `DebugRegressor` estimator.
"""
params = {
"head":
head_lib.regression_head(
weight_column_name=weight_column_name,
label_dimension=label_dimension,
enable_centered_bias=True)
}
super(DebugRegressor, self).__init__(
model_fn=debug_model_fn,
model_dir=model_dir,
config=config,
params=params,
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
example_id_column,
feature_columns,
weight_column_name=None,
model_dir=None,
l1_regularization=0.0,
l2_regularization=1.0,
num_loss_partitions=None,
config=None,
feature_engineering_fn=None):
"""Construct a `SDCARegressor` estimator object. See _SDCAEstimator."""
super(SDCARegressor, self).__init__(
example_id_column=example_id_column,
feature_columns=feature_columns,
weight_column_name=weight_column_name,
model_dir=model_dir,
head=head_lib.regression_head(weight_column_name=weight_column_name),
l1_regularization=l1_regularization,
l2_regularization=l2_regularization,
num_loss_partitions=num_loss_partitions,
config=config,
feature_engineering_fn=None)
def __init__(self,
example_id_column,
feature_columns,
weight_column_name=None,
model_dir=None,
l1_regularization=0.0,
l2_regularization=1.0,
num_loss_partitions=None,
config=None,
feature_engineering_fn=None):
"""Construct a `SDCARegressor` estimator object. See _SDCAEstimator."""
super(SDCARegressor,
self).__init__(example_id_column=example_id_column,
feature_columns=feature_columns,
weight_column_name=weight_column_name,
model_dir=model_dir,
head=head_lib.regression_head(
weight_column_name=weight_column_name),
l1_regularization=l1_regularization,
l2_regularization=l2_regularization,
num_loss_partitions=num_loss_partitions,
config=config,
feature_engineering_fn=None)
def __init__(self,
hidden_units,
feature_columns,
model_dir=None,
weight_column_name=None,
optimizer=None,
activation_fn=nn.relu,
dropout=None,
gradient_clip_norm=None,
enable_centered_bias=False,
config=None,
feature_engineering_fn=None,
label_dimension=1,
embedding_lr_multipliers=None,
input_layer_min_slice_size=None):
"""Initializes a `DNNRegressor` instance.
Args:
hidden_units: List of hidden units per layer. All layers are fully
connected. Ex. `[64, 32]` means first layer has 64 nodes and second one
has 32.
feature_columns: An iterable containing all the feature columns used by
the model. All items in the set should be instances of classes derived
from `FeatureColumn`.
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.
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
optimizer: An instance of `tf.Optimizer` used to train the model. If
`None`, will use an Adagrad optimizer.
activation_fn: Activation function applied to each layer. If `None`, will
use `tf.nn.relu`.
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.
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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
embedding_lr_multipliers: Optional. A dictionary from `EbeddingColumn` to
a `float` multiplier. Multiplier will be used to multiply with
learning rate for the embedding variables.
input_layer_min_slice_size: Optional. The min slice size of input layer
partitions. If not provided, will use the default of 64M.
Returns:
A `DNNRegressor` estimator.
"""
self._feature_columns = tuple(feature_columns or [])
super(DNNRegressor, self).__init__(
model_fn=_dnn_model_fn,
model_dir=model_dir,
config=config,
params={
"head":
head_lib.regression_head(
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=enable_centered_bias),
"hidden_units":
hidden_units,
"feature_columns":
self._feature_columns,
"optimizer":
optimizer,
"activation_fn":
activation_fn,
"dropout":
dropout,
"gradient_clip_norm":
gradient_clip_norm,
"embedding_lr_multipliers":
embedding_lr_multipliers,
"input_layer_min_slice_size":
input_layer_min_slice_size,
},
feature_engineering_fn=feature_engineering_fn)
def __init__(self, # _joint_weights: pylint: disable=invalid-name
feature_columns,
model_dir=None,
weight_column_name=None,
optimizer=None,
gradient_clip_norm=None,
enable_centered_bias=False,
label_dimension=1,
_joint_weights=False,
config=None,
feature_engineering_fn=None):
"""Construct a `LinearRegressor` estimator object.
Args:
feature_columns: An iterable containing all the feature columns used by
the model. All items in the set should be instances of classes derived
from `FeatureColumn`.
model_dir: Directory to save model parameters, graph, etc. This can
also be used to load checkpoints from the directory into a estimator
to continue training a previously saved model.
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
optimizer: An instance of `tf.Optimizer` used to train the model. If
`None`, will use an Ftrl optimizer.
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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
_joint_weights: If True use a single (possibly partitioned) variable to
store the weights. It's faster, but requires all feature columns are
sparse and have the 'sum' combiner. Incompatible with SDCAOptimizer.
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.
Returns:
A `LinearRegressor` estimator.
"""
self._feature_columns = tuple(feature_columns or [])
assert self._feature_columns
chief_hook = None
if (isinstance(optimizer, sdca_optimizer.SDCAOptimizer) and
enable_centered_bias):
enable_centered_bias = False
logging.warning("centered_bias is not supported with SDCA, "
"please disable it explicitly.")
head = head_lib.regression_head(
weight_column_name=weight_column_name,
label_dimension=label_dimension,
enable_centered_bias=enable_centered_bias)
params = {
"head": head,
"feature_columns": feature_columns,
"optimizer": optimizer,
}
if isinstance(optimizer, sdca_optimizer.SDCAOptimizer):
assert label_dimension == 1, "SDCA only applies for label_dimension=1."
assert not _joint_weights, ("_joint_weights is incompatible with"
" SDCAOptimizer.")
model_fn = sdca_model_fn
# The model_fn passes the model parameters to the chief_hook. We then use
# the hook to update weights and shrink step only on the chief.
chief_hook = _SdcaUpdateWeightsHook()
params.update({
"weight_column_name": weight_column_name,
"update_weights_hook": chief_hook,
})
else:
model_fn = _linear_model_fn
params.update({
"gradient_clip_norm": gradient_clip_norm,
"joint_weights": _joint_weights,
})
super(LinearRegressor, self).__init__(
model_fn=model_fn,
model_dir=model_dir,
config=config,
params=params,
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
learner_config,
examples_per_layer,
label_dimension=1,
num_trees=None,
feature_columns=None,
label_name=None,
weight_column_name=None,
model_dir=None,
config=None,
feature_engineering_fn=None,
logits_modifier_function=None,
center_bias=True,
use_core_libs=False,
output_leaf_index=False,
override_global_step_value=None,
num_quantiles=100):
"""Initializes a GradientBoostedDecisionTreeRegressor estimator instance.
Args:
learner_config: A config for the learner.
examples_per_layer: Number of examples to accumulate before growing a
layer. It can also be a function that computes the number of examples
based on the depth of the layer that's being built.
label_dimension: Number of regression labels per example. This is the size
of the last dimension of the labels `Tensor` (typically, this has shape
`[batch_size, label_dimension]`).
num_trees: An int, number of trees to build.
feature_columns: A list of feature columns.
label_name: String, name of the key in label dict. Can be null if label is
a tensor (single headed models).
weight_column_name: Name of the column for weights, or None if not
weighted.
model_dir: Directory for model exports, etc.
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.
logits_modifier_function: A modifier function for the logits.
center_bias: Whether a separate tree should be created for first fitting
the bias.
use_core_libs: Whether feature columns and loss are from the core (as
opposed to contrib) version of tensorflow.
output_leaf_index: whether to output leaf indices along with predictions
during inference. The leaf node indexes are available in predictions
dict by the key 'leaf_index'. For example, result_dict =
classifier.predict(...)
for example_prediction_result in result_dict: # access leaf index list
by example_prediction_result["leaf_index"] # which contains one leaf
index per tree
override_global_step_value: If after the training is done, global step
value must be reset to this value. This should be used to reset global
step to a number > number of steps used to train the current ensemble.
For example, the usual way is to train a number of trees and set a very
large number of training steps. When the training is done (number of
trees were trained), this parameter can be used to set the global step
to a large value, making it look like that number of training steps ran.
If None, no override of global step will happen.
num_quantiles: Number of quantiles to build for numeric feature values.
"""
head = head_lib.regression_head(
label_name=label_name,
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=False)
if label_dimension == 1:
learner_config.num_classes = 2
else:
learner_config.num_classes = label_dimension
super(GradientBoostedDecisionTreeRegressor, self).__init__(
model_fn=model.model_builder,
params={
'head': head,
'feature_columns': feature_columns,
'learner_config': learner_config,
'num_trees': num_trees,
'weight_column_name': weight_column_name,
'examples_per_layer': examples_per_layer,
'logits_modifier_function': logits_modifier_function,
'center_bias': center_bias,
'use_core_libs': use_core_libs,
'output_leaf_index': False,
'override_global_step_value': override_global_step_value,
'num_quantiles': num_quantiles,
},
model_dir=model_dir,
config=config,
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
hidden_units,
feature_columns,
model_dir=None,
weight_column_name=None,
optimizer=None,
activation_fn=nn.relu,
dropout=None,
gradient_clip_norm=None,
enable_centered_bias=False,
config=None,
feature_engineering_fn=None,
label_dimension=1,
embedding_lr_multipliers=None,
input_layer_min_slice_size=None):
"""Initializes a `DNNRegressor` instance.
Args:
hidden_units: List of hidden units per layer. All layers are fully
connected. Ex. `[64, 32]` means first layer has 64 nodes and second one
has 32.
feature_columns: An iterable containing all the feature columns used by
the model. All items in the set should be instances of classes derived
from `FeatureColumn`.
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.
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
optimizer: An instance of `tf.Optimizer` used to train the model. If
`None`, will use an Adagrad optimizer.
activation_fn: Activation function applied to each layer. If `None`, will
use `tf.nn.relu`.
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.
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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
embedding_lr_multipliers: Optional. A dictionary from `EbeddingColumn` to
a `float` multiplier. Multiplier will be used to multiply with
learning rate for the embedding variables.
input_layer_min_slice_size: Optional. The min slice size of input layer
partitions. If not provided, will use the default of 64M.
Returns:
A `DNNRegressor` estimator.
"""
self._feature_columns = tuple(feature_columns or [])
super(DNNRegressor, self).__init__(
model_fn=_dnn_model_fn,
model_dir=model_dir,
config=config,
params={
"head":
head_lib.regression_head(
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=enable_centered_bias),
"hidden_units": hidden_units,
"feature_columns": self._feature_columns,
"optimizer": optimizer,
"activation_fn": activation_fn,
"dropout": dropout,
"gradient_clip_norm": gradient_clip_norm,
"embedding_lr_multipliers": embedding_lr_multipliers,
"input_layer_min_slice_size": input_layer_min_slice_size,
},
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
dnn_hidden_units,
dnn_feature_columns,
tree_learner_config,
num_trees,
tree_examples_per_layer,
weight_column_name=None,
model_dir=None,
config=None,
label_name=None,
label_dimension=1,
feature_engineering_fn=None,
dnn_optimizer="Adagrad",
dnn_activation_fn=nn.relu,
dnn_dropout=None,
dnn_input_layer_partitioner=None,
dnn_input_layer_to_tree=True,
dnn_steps_to_train=10000,
predict_with_tree_only=False,
tree_feature_columns=None,
tree_center_bias=False,
dnn_to_tree_distillation_param=None,
use_core_versions=False):
"""Initializes a DNNBoostedTreeCombinedRegressor instance.
Args:
dnn_hidden_units: List of hidden units per layer for DNN.
dnn_feature_columns: An iterable containing all the feature columns
used by the model's DNN.
tree_learner_config: A config for the tree learner.
num_trees: Number of trees to grow model to after training DNN.
tree_examples_per_layer: Number of examples to accumulate before
growing the tree a layer. This value has a big impact on model
quality and should be set equal to the number of examples in
training dataset if possible. It can also be a function that computes
the number of examples based on the depth of the layer that's
being built.
weight_column_name: The name of weight column.
model_dir: Directory for model exports.
config: `RunConfig` of the estimator.
label_name: String, name of the key in label dict. Can be null if label
is a tensor (single headed models).
label_dimension: Number of regression labels per example. This is the size
of the last dimension of the labels `Tensor` (typically, this has shape
`[batch_size, label_dimension]`).
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.
dnn_optimizer: string, `Optimizer` object, or callable that defines the
optimizer to use for training the DNN. If `None`, will use the Adagrad
optimizer with default learning rate.
dnn_activation_fn: Activation function applied to each layer of the DNN.
If `None`, will use `tf.nn.relu`.
dnn_dropout: When not `None`, the probability to drop out a given
unit in the DNN.
dnn_input_layer_partitioner: Partitioner for input layer of the DNN.
Defaults to `min_max_variable_partitioner` with `min_slice_size`
64 << 20.
dnn_input_layer_to_tree: Whether to provide the DNN's input layer
as a feature to the tree.
dnn_steps_to_train: Number of steps to train dnn for before switching
to gbdt.
predict_with_tree_only: Whether to use only the tree model output as the
final prediction.
tree_feature_columns: An iterable containing all the feature columns
used by the model's boosted trees. If dnn_input_layer_to_tree is
set to True, these features are in addition to dnn_feature_columns.
tree_center_bias: Whether a separate tree should be created for
first fitting the bias.
dnn_to_tree_distillation_param: A Tuple of (float, loss_fn), where the
float defines the weight of the distillation loss, and the loss_fn, for
computing distillation loss, takes dnn_logits, tree_logits and weight
tensor. If the entire tuple is None, no distillation will be applied. If
only the loss_fn is None, we will take the sigmoid/softmax cross entropy
loss be default. When distillation is applied, `predict_with_tree_only`
will be set to True.
use_core_versions: Whether feature columns and loss are from the core (as
opposed to contrib) version of tensorflow.
"""
head = head_lib.regression_head(
label_name=label_name,
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=False)
# num_classes needed for GradientBoostedDecisionTreeModel
if label_dimension == 1:
tree_learner_config.num_classes = 2
else:
tree_learner_config.num_classes = label_dimension
def _model_fn(features, labels, mode, config):
return _dnn_tree_combined_model_fn(
features=features,
labels=labels,
mode=mode,
head=head,
dnn_hidden_units=dnn_hidden_units,
dnn_feature_columns=dnn_feature_columns,
tree_learner_config=tree_learner_config,
num_trees=num_trees,
tree_examples_per_layer=tree_examples_per_layer,
config=config,
dnn_optimizer=dnn_optimizer,
dnn_activation_fn=dnn_activation_fn,
dnn_dropout=dnn_dropout,
dnn_input_layer_partitioner=dnn_input_layer_partitioner,
dnn_input_layer_to_tree=dnn_input_layer_to_tree,
dnn_steps_to_train=dnn_steps_to_train,
predict_with_tree_only=predict_with_tree_only,
tree_feature_columns=tree_feature_columns,
tree_center_bias=tree_center_bias,
dnn_to_tree_distillation_param=dnn_to_tree_distillation_param,
use_core_versions=use_core_versions)
super(DNNBoostedTreeCombinedRegressor, self).__init__(
model_fn=_model_fn,
model_dir=model_dir,
config=config,
feature_engineering_fn=feature_engineering_fn)
def __init__(self, # _joint_linear_weights pylint: disable=invalid-name
model_dir=None,
weight_column_name=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,
enable_centered_bias=False,
label_dimension=1,
config=None,
feature_engineering_fn=None,
embedding_lr_multipliers=None,
input_layer_min_slice_size=None):
"""Initializes a DNNLinearCombinedRegressor instance.
Args:
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.
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
linear_feature_columns: An iterable containing all the feature columns
used by linear part of the model. All items in the set must 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 a single (possibly partitioned) variable
will be used to store the linear model weights. It's faster, but
requires that 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 must 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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
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.
embedding_lr_multipliers: Optional. A dictionary from `EmbeddingColumn` to
a `float` multiplier. Multiplier will be used to multiply with
learning rate for the embedding variables.
input_layer_min_slice_size: Optional. The min slice size of input layer
partitions. If not provided, will use the default of 64M.
Raises:
ValueError: If both linear_feature_columns and dnn_features_columns are
empty at the same time.
"""
linear_feature_columns = linear_feature_columns or []
dnn_feature_columns = dnn_feature_columns or []
self._feature_columns = linear_feature_columns + dnn_feature_columns
if not self._feature_columns:
raise ValueError("Either linear_feature_columns or dnn_feature_columns "
"must be defined.")
head = head_lib.regression_head(
weight_column_name=weight_column_name,
label_dimension=label_dimension,
enable_centered_bias=enable_centered_bias)
super(DNNLinearCombinedRegressor, self).__init__(
model_fn=_dnn_linear_combined_model_fn,
model_dir=model_dir,
config=config,
params={
"head": head,
"linear_feature_columns": linear_feature_columns,
"linear_optimizer": linear_optimizer,
"joint_linear_weights": _joint_linear_weights,
"dnn_feature_columns": dnn_feature_columns,
"dnn_optimizer": dnn_optimizer,
"dnn_hidden_units": dnn_hidden_units,
"dnn_activation_fn": dnn_activation_fn,
"dnn_dropout": dnn_dropout,
"gradient_clip_norm": gradient_clip_norm,
"embedding_lr_multipliers": embedding_lr_multipliers,
"input_layer_min_slice_size": input_layer_min_slice_size,
},
feature_engineering_fn=feature_engineering_fn)
def linear_regressor_model_fn(features, labels, mode, params, config=None):
"""A model_fn for linear models that use a gradient-based optimizer.
Args:
features: `Tensor` or dict of `Tensor` (depends on data passed to `fit`).
labels: `Tensor` of shape [batch_size, 1] or [batch_size] labels of
dtype `int32` or `int64` in the range `[0, n_classes)`.
mode: Defines whether this is training, evaluation or prediction.
See `ModeKeys`.
params: A dict of hyperparameters.
The following hyperparameters are expected:
* head: A `Head` instance.
* feature_columns: An iterable containing all the feature columns
used by
the model.
* optimizer: string, `Optimizer` object, or callable that defines the
optimizer to use for training. If `None`, will use a FTRL
optimizer.
* gradient_clip_norm: A float > 0. If provided, gradients are
clipped to their global norm with this clipping ratio.
* joint_weights: If True, the weights for all columns will be stored
in a
single (possibly partitioned) variable. It's more efficient, but it's
incompatible with SDCAOptimizer, and requires all feature columns are
sparse and use the 'sum' combiner.
config: `RunConfig` object to configure the runtime settings.
Returns:
A `ModelFnOps` instance.
Raises:
ValueError: If mode is not any of the `ModeKeys`.
"""
# head = head_lib.multi_class_head(
# params["n_classes"],
# weight_column_name=params["weight_column_name"],
# enable_centered_bias=params["enable_centered_bias"],
# label_keys=params["label_keys"])
head = head_lib.regression_head(
weight_column_name=params["weight_column_name"],
label_dimension=params["label_dimension"],
enable_centered_bias=params["enable_centered_bias"])
feature_columns = [layers.real_valued_column(i) for i in features.keys()]
optimizer = _get_default_optimizer(feature_columns)
gradient_clip_norm = params.get("gradient_clip_norm", None)
num_ps_replicas = config.num_ps_replicas if config else 0
joint_weights = params.get("joint_weights", False)
if not isinstance(features, dict):
features = {"": features}
parent_scope = "linear"
partitioner = partitioned_variables.min_max_variable_partitioner(
max_partitions=num_ps_replicas, min_slice_size=64 << 20)
with variable_scope.variable_scope(parent_scope,
values=tuple(six.itervalues(features)),
partitioner=partitioner) as scope:
if all([
isinstance(fc, feature_column._FeatureColumn)
# pylint: disable=protected-access
for fc in feature_columns
]):
if joint_weights:
layer_fn = layers.joint_weighted_sum_from_feature_columns
else:
layer_fn = layers.weighted_sum_from_feature_columns
logits, _, _ = layer_fn(columns_to_tensors=features,
feature_columns=feature_columns,
num_outputs=head.logits_dimension,
weight_collections=[parent_scope],
scope=scope)
else:
logits = fc_core.linear_model(features=features,
feature_columns=feature_columns,
units=head.logits_dimension,
weight_collections=[parent_scope])
def _train_op_fn(loss):
global_step = contrib_variables.get_global_step()
my_vars = ops.get_collection(parent_scope)
grads = gradients.gradients(loss, my_vars)
if gradient_clip_norm:
grads, _ = clip_ops.clip_by_global_norm(
grads, gradient_clip_norm)
return (_get_optimizer(optimizer).apply_gradients(
zip(grads, my_vars), global_step=global_step))
return head.create_model_fn_ops(features=features,
mode=mode,
labels=labels,
train_op_fn=_train_op_fn,
logits=logits)
def __init__(self,
hidden_units,
feature_columns,
model_dir=None,
optimizer=None,
activation_fn=nn.relu,
dropout=None,
gradient_clip_norm=None,
enable_centered_bias=False,
config=None,
feature_engineering_fn=None,
label_dimension=1):
"""Initializes a `DNNRegressor` instance.
Args:
hidden_units: List of hidden units per layer. All layers are fully
connected. Ex. `[64, 32]` means first layer has 64 nodes and second one
has 32.
feature_columns: An iterable containing all the feature columns used by
the model. All items in the set should be instances of classes derived
from `FeatureColumn`.
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.
optimizer: An instance of `tf.Optimizer` used to train the model. If
`None`, will use an Adagrad optimizer.
activation_fn: Activation function applied to each layer. If `None`, will
use `tf.nn.relu`. Note that a string containing the unqualified name of
the op may also be provided, e.g., "relu", "tanh", or "sigmoid".
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.
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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
Returns:
A `DNNRegressor` estimator.
"""
self._feature_columns = tuple(feature_columns or [])
super(MyRegressor, self).__init__(
model_fn=_dnn_model_fn,
model_dir=model_dir,
config=config,
params={
"head":
head_lib.regression_head(
label_dimension=label_dimension,
enable_centered_bias=enable_centered_bias),
"hidden_units":
hidden_units,
"feature_columns":
self._feature_columns,
"optimizer":
optimizer,
"activation_fn":
activation_fn,
"dropout":
dropout,
"gradient_clip_norm":
gradient_clip_norm,
},
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
learner_config,
examples_per_layer,
label_dimension=1,
num_trees=None,
feature_columns=None,
label_name=None,
weight_column_name=None,
model_dir=None,
config=None,
feature_engineering_fn=None,
logits_modifier_function=None,
center_bias=True,
use_core_libs=False,
output_leaf_index=False,
override_global_step_value=None):
"""Initializes a GradientBoostedDecisionTreeRegressor estimator instance.
Args:
learner_config: A config for the learner.
examples_per_layer: Number of examples to accumulate before growing a
layer. It can also be a function that computes the number of examples
based on the depth of the layer that's being built.
label_dimension: Number of regression labels per example. This is the size
of the last dimension of the labels `Tensor` (typically, this has shape
`[batch_size, label_dimension]`).
num_trees: An int, number of trees to build.
feature_columns: A list of feature columns.
label_name: String, name of the key in label dict. Can be null if label
is a tensor (single headed models).
weight_column_name: Name of the column for weights, or None if not
weighted.
model_dir: Directory for model exports, etc.
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.
logits_modifier_function: A modifier function for the logits.
center_bias: Whether a separate tree should be created for first fitting
the bias.
use_core_libs: Whether feature columns and loss are from the core (as
opposed to contrib) version of tensorflow.
output_leaf_index: whether to output leaf indices along with predictions
during inference. The leaf node indexes are available in predictions
dict by the key 'leaf_index'. For example,
result_dict = classifier.predict(...)
for example_prediction_result in result_dict:
# access leaf index list by example_prediction_result["leaf_index"]
# which contains one leaf index per tree
override_global_step_value: If after the training is done, global step
value must be reset to this value. This should be used to reset global
step to a number > number of steps used to train the current ensemble.
For example, the usual way is to train a number of trees and set a very
large number of training steps. When the training is done (number of
trees were trained), this parameter can be used to set the global step
to a large value, making it look like that number of training steps ran.
If None, no override of global step will happen.
"""
head = head_lib.regression_head(
label_name=label_name,
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=False)
if label_dimension == 1:
learner_config.num_classes = 2
else:
learner_config.num_classes = label_dimension
super(GradientBoostedDecisionTreeRegressor, self).__init__(
model_fn=model.model_builder,
params={
'head': head,
'feature_columns': feature_columns,
'learner_config': learner_config,
'num_trees': num_trees,
'weight_column_name': weight_column_name,
'examples_per_layer': examples_per_layer,
'logits_modifier_function': logits_modifier_function,
'center_bias': center_bias,
'use_core_libs': use_core_libs,
'output_leaf_index': False,
'override_global_step_value': override_global_step_value
},
model_dir=model_dir,
config=config,
feature_engineering_fn=feature_engineering_fn)
def __init__(self, # _joint_linear_weights pylint: disable=invalid-name
model_dir=None,
weight_column_name=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,
enable_centered_bias=False,
label_dimension=1,
config=None,
feature_engineering_fn=None,
embedding_lr_multipliers=None,
input_layer_min_slice_size=None,
fix_global_step_increment_bug=False):
"""Initializes a DNNLinearCombinedRegressor instance.
Note: New users must set `fix_global_step_increment_bug=True` when creating
an estimator.
Args:
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.
weight_column_name: A string defining feature column name representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
linear_feature_columns: An iterable containing all the feature columns
used by linear part of the model. All items in the set must 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 a single (possibly partitioned) variable
will be used to store the linear model weights. It's faster, but
requires that 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 must 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.
label_dimension: Number of regression targets per example. This is the
size of the last dimension of the labels and logits `Tensor` objects
(typically, these have shape `[batch_size, label_dimension]`).
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.
embedding_lr_multipliers: Optional. A dictionary from `EmbeddingColumn` to
a `float` multiplier. Multiplier will be used to multiply with
learning rate for the embedding variables.
input_layer_min_slice_size: Optional. The min slice size of input layer
partitions. If not provided, will use the default of 64M.
fix_global_step_increment_bug: If `False`, the estimator needs two fit
steps to optimize both linear and dnn parts. If `True`, this bug is
fixed. New users must set this to `True`, but it the default value is
`False` for backwards compatibility.
Raises:
ValueError: If both linear_feature_columns and dnn_features_columns are
empty at the same time.
"""
linear_feature_columns = linear_feature_columns or []
dnn_feature_columns = dnn_feature_columns or []
self._feature_columns = linear_feature_columns + dnn_feature_columns
if not self._feature_columns:
raise ValueError("Either linear_feature_columns or dnn_feature_columns "
"must be defined.")
# TODO(b/35922130): Replace with `input_layer_partitioner` arg.
input_layer_partitioner = None
if input_layer_min_slice_size is not None:
input_layer_partitioner = (
partitioned_variables.min_max_variable_partitioner(
max_partitions=config.num_ps_replicas if config else 0,
min_slice_size=input_layer_min_slice_size))
head = head_lib.regression_head(
weight_column_name=weight_column_name,
label_dimension=label_dimension,
enable_centered_bias=enable_centered_bias)
super(DNNLinearCombinedRegressor, self).__init__(
model_fn=_dnn_linear_combined_model_fn,
model_dir=model_dir,
config=config,
params={
"head": head,
"linear_feature_columns": linear_feature_columns,
"linear_optimizer": linear_optimizer,
"joint_linear_weights": _joint_linear_weights,
"dnn_feature_columns": dnn_feature_columns,
"dnn_optimizer": dnn_optimizer,
"dnn_hidden_units": dnn_hidden_units,
"dnn_activation_fn": dnn_activation_fn,
"dnn_dropout": dnn_dropout,
"gradient_clip_norm": gradient_clip_norm,
"embedding_lr_multipliers": embedding_lr_multipliers,
"input_layer_partitioner": input_layer_partitioner,
"fix_global_step_increment_bug": fix_global_step_increment_bug,
},
feature_engineering_fn=feature_engineering_fn)
def __init__(self,
dnn_hidden_units,
dnn_feature_columns,
tree_learner_config,
num_trees,
tree_examples_per_layer,
weight_column_name=None,
model_dir=None,
config=None,
label_name=None,
label_dimension=1,
feature_engineering_fn=None,
dnn_optimizer="Adagrad",
dnn_activation_fn=nn.relu,
dnn_dropout=None,
dnn_input_layer_partitioner=None,
dnn_input_layer_to_tree=True,
dnn_steps_to_train=10000,
tree_feature_columns=None,
tree_center_bias=False,
use_core_versions=False):
"""Initializes a DNNBoostedTreeCombinedRegressor instance.
Args:
dnn_hidden_units: List of hidden units per layer for DNN.
dnn_feature_columns: An iterable containing all the feature columns
used by the model's DNN.
tree_learner_config: A config for the tree learner.
num_trees: Number of trees to grow model to after training DNN.
tree_examples_per_layer: Number of examples to accumulate before
growing the tree a layer. This value has a big impact on model
quality and should be set equal to the number of examples in
training dataset if possible. It can also be a function that computes
the number of examples based on the depth of the layer that's
being built.
weight_column_name: The name of weight column.
model_dir: Directory for model exports.
config: `RunConfig` of the estimator.
label_name: String, name of the key in label dict. Can be null if label
is a tensor (single headed models).
label_dimension: Number of regression labels per example. This is the size
of the last dimension of the labels `Tensor` (typically, this has shape
`[batch_size, label_dimension]`).
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.
dnn_optimizer: string, `Optimizer` object, or callable that defines the
optimizer to use for training the DNN. If `None`, will use the Adagrad
optimizer with default learning rate.
dnn_activation_fn: Activation function applied to each layer of the DNN.
If `None`, will use `tf.nn.relu`.
dnn_dropout: When not `None`, the probability to drop out a given
unit in the DNN.
dnn_input_layer_partitioner: Partitioner for input layer of the DNN.
Defaults to `min_max_variable_partitioner` with `min_slice_size`
64 << 20.
dnn_input_layer_to_tree: Whether to provide the DNN's input layer
as a feature to the tree.
dnn_steps_to_train: Number of steps to train dnn for before switching
to gbdt.
tree_feature_columns: An iterable containing all the feature columns
used by the model's boosted trees. If dnn_input_layer_to_tree is
set to True, these features are in addition to dnn_feature_columns.
tree_center_bias: Whether a separate tree should be created for
first fitting the bias.
use_core_versions: Whether feature columns and loss are from the core (as
opposed to contrib) version of tensorflow.
"""
head = head_lib.regression_head(label_name=label_name,
label_dimension=label_dimension,
weight_column_name=weight_column_name,
enable_centered_bias=False)
# num_classes needed for GradientBoostedDecisionTreeModel
if label_dimension == 1:
tree_learner_config.num_classes = 2
else:
tree_learner_config.num_classes = label_dimension
def _model_fn(features, labels, mode, config):
return _dnn_tree_combined_model_fn(
features, labels, mode, head, dnn_hidden_units,
dnn_feature_columns, tree_learner_config, num_trees,
tree_examples_per_layer, config, dnn_optimizer,
dnn_activation_fn, dnn_dropout, dnn_input_layer_partitioner,
dnn_input_layer_to_tree, dnn_steps_to_train,
tree_feature_columns, tree_center_bias, use_core_versions)
super(DNNBoostedTreeCombinedRegressor,
self).__init__(model_fn=_model_fn,
model_dir=model_dir,
config=config,
feature_engineering_fn=feature_engineering_fn)