def _test_complete_flow(self, train_input_fn, eval_input_fn,
predict_input_fn, input_dimension, label_dimension,
batch_size):
feature_columns = [
tf.feature_column.numeric_column('x', shape=(input_dimension, ))
]
est = linear.LinearEstimator(
head=head_lib._regression_head(label_dimension=label_dimension),
feature_columns=feature_columns,
model_dir=self._model_dir)
# Train
num_steps = 10
est.train(train_input_fn, steps=num_steps)
# Evaluate
scores = est.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[tf.compat.v1.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
# Predict
predictions = np.array([
x[prediction_keys.PredictionKeys.PREDICTIONS]
for x in est.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, label_dimension), predictions.shape)
# Export
feature_spec = tf.compat.v1.feature_column.make_parse_example_spec(
feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = est.export_saved_model(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(tf.compat.v1.gfile.Exists(export_dir))
def _baseline_estimator_fn(weight_column=None,
label_dimension=1,
**kwargs):
return baseline.BaselineEstimator(
head=head_lib._regression_head(
weight_column=weight_column,
label_dimension=label_dimension,
loss_reduction=tf.compat.v1.losses.Reduction.SUM),
**kwargs)
def _linear_estimator_fn(weight_column=None, label_dimension=1, **kwargs):
"""Returns a LinearEstimator that uses regression_head."""
return linear.LinearEstimator(
head=head_lib._regression_head(
weight_column=weight_column,
label_dimension=label_dimension,
# Tests in core (from which this test inherits) test the sum loss.
loss_reduction=tf.compat.v1.losses.Reduction.SUM),
**kwargs)
def _test_ckpt_converter(self, train_input_fn, eval_input_fn,
predict_input_fn, input_dimension, label_dimension,
batch_size, dnn_optimizer, linear_optimizer):
# Create checkpoint in CannedEstimator v1.
linear_feature_columns_v1 = [
feature_column._numeric_column('x', shape=(input_dimension,))
]
dnn_feature_columns_v1 = [
feature_column._numeric_column('x', shape=(input_dimension,))
]
est_v1 = dnn_linear_combined.DNNLinearCombinedEstimator(
head=head_lib._regression_head(label_dimension=label_dimension),
linear_feature_columns=linear_feature_columns_v1,
dnn_feature_columns=dnn_feature_columns_v1,
dnn_hidden_units=(2, 2),
model_dir=self._old_ckpt_dir,
dnn_optimizer=dnn_optimizer,
linear_optimizer=linear_optimizer)
# Train
num_steps = 10
est_v1.train(train_input_fn, steps=num_steps)
self.assertIsNotNone(est_v1.latest_checkpoint())
self.assertTrue(est_v1.latest_checkpoint().startswith(self._old_ckpt_dir))
# Convert checkpoint from v1 to v2.
source_checkpoint = os.path.join(self._old_ckpt_dir, 'model.ckpt-10')
source_graph = os.path.join(self._old_ckpt_dir, 'graph.pbtxt')
target_checkpoint = os.path.join(self._new_ckpt_dir, 'model.ckpt-10')
checkpoint_converter.convert_checkpoint('combined', source_checkpoint,
source_graph, target_checkpoint)
# Create CannedEstimator V2 and restore from the converted checkpoint.
linear_feature_columns_v2 = [
tf.feature_column.numeric_column('x', shape=(input_dimension,))
]
dnn_feature_columns_v2 = [
tf.feature_column.numeric_column('x', shape=(input_dimension,))
]
est_v2 = dnn_linear_combined.DNNLinearCombinedEstimatorV2(
head=regression_head.RegressionHead(label_dimension=label_dimension),
linear_feature_columns=linear_feature_columns_v2,
dnn_feature_columns=dnn_feature_columns_v2,
dnn_hidden_units=(2, 2),
model_dir=self._new_ckpt_dir,
dnn_optimizer=dnn_optimizer,
linear_optimizer=linear_optimizer)
# Train
extra_steps = 10
est_v2.train(train_input_fn, steps=extra_steps)
self.assertIsNotNone(est_v2.latest_checkpoint())
self.assertTrue(est_v2.latest_checkpoint().startswith(self._new_ckpt_dir))
# Make sure estimator v2 restores from the converted checkpoint, and
# continues training extra steps.
self.assertEqual(num_steps + extra_steps,
est_v2.get_variable_value(tf.compat.v1.GraphKeys.GLOBAL_STEP))
def __init__(self,
model_dir=None,
linear_feature_columns=None,
linear_optimizer='Ftrl',
dnn_feature_columns=None,
dnn_optimizer='Adagrad',
dnn_hidden_units=None,
dnn_activation_fn=tf.nn.relu,
dnn_dropout=None,
label_dimension=1,
weight_column=None,
input_layer_partitioner=None,
config=None,
warm_start_from=None,
loss_reduction=tf.compat.v1.losses.Reduction.SUM,
batch_norm=False,
linear_sparse_combiner='sum'):
self._feature_columns = _validate_feature_columns(
linear_feature_columns=linear_feature_columns,
dnn_feature_columns=dnn_feature_columns)
estimator._canned_estimator_api_gauge.get_cell('Regressor').set(
'DNNLinearCombined') # pylint: disable=protected-access
head = head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction)
def _model_fn(features, labels, mode, config):
"""Call the _dnn_linear_combined_model_fn."""
return _dnn_linear_combined_model_fn(
features=features,
labels=labels,
mode=mode,
head=head,
linear_feature_columns=linear_feature_columns,
linear_optimizer=linear_optimizer,
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,
input_layer_partitioner=input_layer_partitioner,
config=config,
batch_norm=batch_norm,
linear_sparse_combiner=linear_sparse_combiner)
super(DNNLinearCombinedRegressor, self).__init__(
model_fn=_model_fn,
model_dir=model_dir,
config=config,
warm_start_from=warm_start_from)
def __init__(self,
model_dir=None,
label_dimension=1,
weight_column=None,
optimizer='Ftrl',
config=None,
loss_reduction=losses.Reduction.SUM):
"""Initializes a BaselineRegressor 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: A string or a `_NumericColumn` created by
`tf.feature_column.numeric_column` defining feature column representing
weights. It will be multiplied by the loss of the example.
optimizer: String, `tf.Optimizer` object, or callable that creates the
optimizer to use for training. If not specified, will use
`FtrlOptimizer` with a default learning rate of 0.3.
config: `RunConfig` object to configure the runtime settings.
loss_reduction: One of `tf.losses.Reduction` except `NONE`. Describes how
to reduce training loss over batch. Defaults to `SUM`.
Returns:
A `BaselineRegressor` estimator.
"""
# TODO(b/117517419): Update this reference once head moves to core.
head = head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction)
def _model_fn(features, labels, mode, config):
return _baseline_model_fn(features=features,
labels=labels,
mode=mode,
head=head,
optimizer=optimizer,
config=config)
super(BaselineRegressor, self).__init__(model_fn=_model_fn,
model_dir=model_dir,
config=config)
def _model_fn(features, labels, mode, config):
"""Call the defined shared _dnn_model_fn."""
return _dnn_model_fn(
features=features,
labels=labels,
mode=mode,
head=head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction),
hidden_units=hidden_units,
feature_columns=tuple(feature_columns or []),
optimizer=optimizer,
activation_fn=activation_fn,
dropout=dropout,
input_layer_partitioner=input_layer_partitioner,
config=config,
batch_norm=batch_norm)
def __init__(
self,
hidden_units,
feature_columns,
model_dir=None,
label_dimension=1,
weight_column=None,
optimizer='Adagrad',
activation_fn=tf.nn.relu,
dropout=None,
input_layer_partitioner=None,
config=None,
warm_start_from=None,
loss_reduction=tf.compat.v1.losses.Reduction.SUM,
batch_norm=False,
):
head = head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction)
estimator._canned_estimator_api_gauge.get_cell('Regressor').set('DNN') # pylint: disable=protected-access
def _model_fn(features, labels, mode, config):
"""Call the defined shared _dnn_model_fn."""
return _dnn_model_fn(
features=features,
labels=labels,
mode=mode,
head=head,
hidden_units=hidden_units,
feature_columns=tuple(feature_columns or []),
optimizer=optimizer,
activation_fn=activation_fn,
dropout=dropout,
input_layer_partitioner=input_layer_partitioner,
config=config,
batch_norm=batch_norm)
super(DNNRegressor, self).__init__(
model_fn=_model_fn,
model_dir=model_dir,
config=config,
warm_start_from=warm_start_from)
def _linear_only_estimator_fn(feature_columns,
model_dir=None,
label_dimension=1,
weight_column=None,
optimizer='Ftrl',
config=None,
partitioner=None,
sparse_combiner='sum'):
return dnn_linear_combined.DNNLinearCombinedEstimator(
head=head_lib._regression_head(
weight_column=weight_column,
label_dimension=label_dimension,
# Tests in core (from which this test inherits) test the sum loss.
loss_reduction=losses.Reduction.SUM),
model_dir=model_dir,
linear_feature_columns=feature_columns,
linear_optimizer=optimizer,
input_layer_partitioner=partitioner,
config=config,
linear_sparse_combiner=sparse_combiner)
def _model_fn(features, labels, mode, config):
"""Call the _dnn_linear_combined_model_fn."""
return _dnn_linear_combined_model_fn(
features=features,
labels=labels,
mode=mode,
head=head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction),
linear_feature_columns=linear_feature_columns,
linear_optimizer=linear_optimizer,
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,
input_layer_partitioner=input_layer_partitioner,
config=config,
batch_norm=batch_norm,
linear_sparse_combiner=linear_sparse_combiner)
def __init__(self,
feature_columns,
model_dir=None,
label_dimension=1,
weight_column=None,
optimizer='Ftrl',
config=None,
partitioner=None,
warm_start_from=None,
loss_reduction=losses.Reduction.SUM,
sparse_combiner='sum'):
_validate_linear_sdca_optimizer_for_linear_regressor(
feature_columns=feature_columns,
label_dimension=label_dimension,
optimizer=optimizer,
sparse_combiner=sparse_combiner)
head = head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction)
def _model_fn(features, labels, mode, config):
"""Call the defined shared _linear_model_fn."""
return _linear_model_fn(
features=features,
labels=labels,
mode=mode,
head=head,
feature_columns=tuple(feature_columns or []),
optimizer=optimizer,
partitioner=partitioner,
config=config,
sparse_combiner=sparse_combiner)
super(LinearRegressor, self).__init__(
model_fn=_model_fn,
model_dir=model_dir,
config=config,
warm_start_from=warm_start_from)
def __init__(self,
model_dir=None,
label_dimension=1,
weight_column=None,
optimizer='Ftrl',
config=None,
loss_reduction=losses.Reduction.SUM):
head = head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction)
def _model_fn(features, labels, mode, config):
return _baseline_model_fn(features=features,
labels=labels,
mode=mode,
head=head,
optimizer=optimizer,
config=config)
super(BaselineRegressor, self).__init__(model_fn=_model_fn,
model_dir=model_dir,
config=config)
def _dnn_only_estimator_fn(hidden_units,
feature_columns,
model_dir=None,
label_dimension=1,
weight_column=None,
optimizer='Adagrad',
activation_fn=nn.relu,
dropout=None,
input_layer_partitioner=None,
config=None):
return dnn_linear_combined.DNNLinearCombinedEstimator(
head=head_lib._regression_head(
weight_column=weight_column,
label_dimension=label_dimension,
# Tests in core (from which this test inherits) test the sum loss.
loss_reduction=losses.Reduction.SUM),
model_dir=model_dir,
dnn_feature_columns=feature_columns,
dnn_optimizer=optimizer,
dnn_hidden_units=hidden_units,
dnn_activation_fn=activation_fn,
dnn_dropout=dropout,
input_layer_partitioner=input_layer_partitioner,
config=config)
def regression_head(weight_column=None,
label_dimension=1,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE,
loss_fn=None,
inverse_link_fn=None,
name=None):
"""Creates a `_Head` for regression using the `mean_squared_error` loss.
The loss is the weighted sum over all input dimensions. Namely, if the input
labels have shape `[batch_size, label_dimension]`, the loss is the weighted
sum over both `batch_size` and `label_dimension`.
The head expects `logits` with shape `[D0, D1, ... DN, label_dimension]`.
In many applications, the shape is `[batch_size, label_dimension]`.
The `labels` shape must match `logits`, namely
`[D0, D1, ... DN, label_dimension]`. If `label_dimension=1`, shape
`[D0, D1, ... DN]` is also supported.
If `weight_column` is specified, weights must be of shape
`[D0, D1, ... DN]`, `[D0, D1, ... DN, 1]` or
`[D0, D1, ... DN, label_dimension]`.
Supports custom `loss_fn`. `loss_fn` takes `(labels, logits)` or
`(labels, logits, features)` as arguments and returns unreduced loss with
shape `[D0, D1, ... DN, label_dimension]`.
Also supports custom `inverse_link_fn`, also known as 'mean function'.
`inverse_link_fn` is only used in `PREDICT` mode. It takes `logits` as
argument and returns predicted values. This function is the inverse of the
link function defined in
https://en.wikipedia.org/wiki/Generalized_linear_model#Link_function
Namely, for poisson regression, set `inverse_link_fn=tf.exp`.
The head can be used with a canned estimator. Example:
```python
my_head = tf.contrib.estimator.regression_head()
my_estimator = tf.contrib.estimator.DNNEstimator(
head=my_head,
hidden_units=...,
feature_columns=...)
```
It can also be used with a custom `model_fn`. Example:
```python
def _my_model_fn(features, labels, mode):
my_head = tf.contrib.estimator.regression_head()
logits = tf.keras.Model(...)(features)
return my_head.create_estimator_spec(
features=features,
mode=mode,
labels=labels,
optimizer=tf.AdagradOptimizer(learning_rate=0.1),
logits=logits)
my_estimator = tf.estimator.Estimator(model_fn=_my_model_fn)
```
Args:
weight_column: A string or a `_NumericColumn` created by
`tf.feature_column.numeric_column` defining feature column representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
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]`).
loss_reduction: One of `tf.losses.Reduction` except `NONE`. Describes how to
reduce training loss over batch and label dimension. Defaults to
`SUM_OVER_BATCH_SIZE`, namely weighted sum of losses divided by
`batch size * label_dimension`. See `tf.losses.Reduction`.
loss_fn: Optional loss function. Defaults to `mean_squared_error`.
inverse_link_fn: Optional inverse link function, also known as 'mean
function'. Defaults to identity.
name: name of the head. If provided, summary and metrics keys will be
suffixed by `"/" + name`. Also used as `name_scope` when creating ops.
Returns:
An instance of `_Head` for linear regression.
Raises:
ValueError: If `label_dimension` or `loss_reduction` is invalid.
"""
return head_lib._regression_head( # pylint:disable=protected-access
weight_column=weight_column,
label_dimension=label_dimension,
loss_reduction=loss_reduction,
loss_fn=loss_fn,
inverse_link_fn=inverse_link_fn,
name=name)
def train_and_evaluate_estimator():
"""Runs Estimator distributed training."""
# The tf.estimator.RunConfig automatically parses the TF_CONFIG environment
# variables during construction.
# For more information on how tf.estimator.RunConfig uses TF_CONFIG, see
# https://www.tensorflow.org/api_docs/python/tf/estimator/RunConfig.
config = tf.estimator.RunConfig(
tf_random_seed=42,
save_checkpoints_steps=10,
save_checkpoints_secs=None,
# Keep all checkpoints to avoid checkpoint GC causing failures during
# evaluation.
# TODO: Prevent checkpoints that are currently being
# evaluated by another process from being garbage collected.
keep_checkpoint_max=None,
model_dir=FLAGS.model_dir,
session_config=tf_compat.v1.ConfigProto(
log_device_placement=False,
# Ignore other workers; only talk to parameter servers.
# Otherwise, when a chief/worker terminates, the others will hang.
device_filters=["/job:ps"]))
def input_fn():
input_features = {"x": tf.constant(features, name="x")}
input_labels = tf.constant(labels, name="y")
return tf.data.Dataset.from_tensors((input_features, input_labels)).repeat()
kwargs = {
"max_iteration_steps": 100,
"force_grow": True,
"delay_secs_per_worker": .2,
"max_worker_delay_secs": 1,
"worker_wait_secs": 1,
# Set low timeout to reduce wait time for failures.
"worker_wait_timeout_secs": 180,
"evaluator": Evaluator(input_fn, steps=10),
"config": config
}
head = head_lib._regression_head( # pylint: disable=protected-access
loss_reduction=tf.losses.Reduction.SUM_OVER_BATCH_SIZE)
features = [[1., 0.], [0., 0], [0., 1.], [1., 1.]]
labels = [[1.], [0.], [1.], [0.]]
estimator_type = FLAGS.estimator_type
if FLAGS.placement_strategy == "round_robin":
kwargs["experimental_placement_strategy"] = RoundRobinStrategy()
if estimator_type == "autoensemble":
feature_columns = [tf.feature_column.numeric_column("x", shape=[2])]
# pylint: disable=g-long-lambda
# TODO: Switch optimizers to tf.keras.optimizers.Adam once the
# distribution bug is fixed.
candidate_pool = {
"linear":
tf.estimator.LinearEstimator(
head=head,
feature_columns=feature_columns,
optimizer=lambda: tf_compat.v1.train.AdamOptimizer(
learning_rate=.001)),
"dnn":
tf.estimator.DNNEstimator(
head=head,
feature_columns=feature_columns,
optimizer=lambda: tf_compat.v1.train.AdamOptimizer(
learning_rate=.001),
hidden_units=[3]),
"dnn2":
tf.estimator.DNNEstimator(
head=head,
feature_columns=feature_columns,
optimizer=lambda: tf_compat.v1.train.AdamOptimizer(
learning_rate=.001),
hidden_units=[10, 10]),
}
# pylint: enable=g-long-lambda
estimator = AutoEnsembleEstimator(
head=head, candidate_pool=candidate_pool, **kwargs)
elif estimator_type == "estimator":
subnetwork_generator = SimpleGenerator([
_DNNBuilder("dnn1", config, layer_size=3),
_DNNBuilder("dnn2", config, layer_size=4),
_DNNBuilder("dnn3", config, layer_size=5),
])
estimator = Estimator(
head=head, subnetwork_generator=subnetwork_generator, **kwargs)
elif FLAGS.estimator_type == "autoensemble_trees_multiclass":
if not bt_losses:
logging.warning(
"Skipped autoensemble_trees_multiclass test since contrib is missing."
)
return
n_classes = 3
head = head_lib._multi_class_head_with_softmax_cross_entropy_loss( # pylint: disable=protected-access
n_classes=n_classes,
loss_reduction=tf.losses.Reduction.SUM_OVER_BATCH_SIZE)
def tree_loss_fn(labels, logits):
result = bt_losses.per_example_maxent_loss(
labels=labels, logits=logits, num_classes=n_classes, weights=None)
return result[0]
tree_head = head_lib._multi_class_head_with_softmax_cross_entropy_loss( # pylint: disable=protected-access
loss_fn=tree_loss_fn,
n_classes=n_classes,
loss_reduction=tf.losses.Reduction.SUM_OVER_BATCH_SIZE)
labels = [[1], [0], [1], [2]]
feature_columns = [tf.feature_column.numeric_column("x", shape=[2])]
# TODO: Switch optimizers to tf.keras.optimizers.Adam once the
# distribution bug is fixed.
candidate_pool = lambda config: { # pylint: disable=g-long-lambda
"linear":
tf.estimator.LinearEstimator(
head=head,
feature_columns=feature_columns,
optimizer=tf_compat.v1.train.AdamOptimizer(
learning_rate=.001),
config=config),
"gbdt":
tf.estimator.BoostedTreesEstimator(
head=tree_head,
feature_columns=feature_columns,
n_trees=10,
n_batches_per_layer=1,
center_bias=False,
config=config),
}
estimator = AutoEnsembleEstimator(
head=head, candidate_pool=candidate_pool, **kwargs)
elif estimator_type == "estimator_with_experimental_multiworker_strategy":
def _model_fn(features, labels, mode):
"""Test model_fn."""
layer = tf.keras.layers.Dense(1)
logits = layer(features["x"])
if mode == tf.estimator.ModeKeys.PREDICT:
predictions = {"logits": logits}
return tf.estimator.EstimatorSpec(mode, predictions=predictions)
loss = tf.losses.mean_squared_error(
labels=labels,
predictions=logits,
reduction=tf.losses.Reduction.SUM_OVER_BATCH_SIZE)
if mode == tf.estimator.ModeKeys.EVAL:
return tf.estimator.EstimatorSpec(mode, loss=loss)
if mode == tf.estimator.ModeKeys.TRAIN:
optimizer = tf.train.GradientDescentOptimizer(0.2)
train_op = optimizer.minimize(
loss, global_step=tf.train.get_global_step())
return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op)
if json.loads(os.environ["TF_CONFIG"])["task"]["type"] == "evaluator":
# The evaluator job would crash if MultiWorkerMirroredStrategy is called.
distribution = None
else:
distribution = tf.distribute.experimental.MultiWorkerMirroredStrategy()
multiworker_config = tf.estimator.RunConfig(
tf_random_seed=42,
model_dir=FLAGS.model_dir,
train_distribute=distribution,
session_config=tf_compat.v1.ConfigProto(log_device_placement=False))
# TODO: Replace with adanet.Estimator. Currently this just verifies
# that the distributed testing framework supports distribute strategies.
estimator = tf.estimator.Estimator(
model_fn=_model_fn, config=multiworker_config)
train_hooks = [
tf.estimator.ProfilerHook(save_steps=50, output_dir=FLAGS.model_dir)
]
# Train for three iterations.
train_spec = tf.estimator.TrainSpec(
input_fn=input_fn, max_steps=300, hooks=train_hooks)
eval_spec = tf.estimator.EvalSpec(
input_fn=input_fn, steps=1, start_delay_secs=.5, throttle_secs=.05)
# Calling train_and_evaluate is the official way to perform distributed
# training with an Estimator. Calling Estimator#train directly results
# in an error when the TF_CONFIG is setup for a cluster.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
def poisson_regression_head(
weight_column=None,
label_dimension=1,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE,
compute_full_loss=True,
name=None):
"""Creates a `_Head` for poisson regression using `tf.nn.log_poisson_loss`.
The loss is the weighted sum over all input dimensions. Namely, if the input
labels have shape `[batch_size, label_dimension]`, the loss is the weighted
sum over both `batch_size` and `label_dimension`.
The head expects `logits` with shape `[D0, D1, ... DN, label_dimension]`.
In many applications, the shape is `[batch_size, label_dimension]`.
The `labels` shape must match `logits`, namely
`[D0, D1, ... DN, label_dimension]`. If `label_dimension=1`, shape
`[D0, D1, ... DN]` is also supported.
If `weight_column` is specified, weights must be of shape
`[D0, D1, ... DN]`, `[D0, D1, ... DN, 1]` or
`[D0, D1, ... DN, label_dimension]`.
This is implemented as a generalized linear model, see
https://en.wikipedia.org/wiki/Generalized_linear_model.
The head can be used with a canned estimator. Example:
```python
my_head = tf.contrib.estimator.poisson_regression_head()
my_estimator = tf.contrib.estimator.DNNEstimator(
head=my_head,
hidden_units=...,
feature_columns=...)
```
It can also be used with a custom `model_fn`. Example:
```python
def _my_model_fn(features, labels, mode):
my_head = tf.contrib.estimator.poisson_regression_head()
logits = tf.keras.Model(...)(features)
return my_head.create_estimator_spec(
features=features,
mode=mode,
labels=labels,
optimizer=tf.AdagradOptimizer(learning_rate=0.1),
logits=logits)
my_estimator = tf.estimator.Estimator(model_fn=_my_model_fn)
```
Args:
weight_column: A string or a `_NumericColumn` created by
`tf.feature_column.numeric_column` defining feature column representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
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]`).
loss_reduction: One of `tf.losses.Reduction` except `NONE`. Describes how to
reduce training loss over batch and label dimension. Defaults to
`SUM_OVER_BATCH_SIZE`, namely weighted sum of losses divided by
`batch size * label_dimension`. See `tf.losses.Reduction`.
compute_full_loss: Whether to include the constant `log(z!)` term in
computing the poisson loss. See `tf.nn.log_poisson_loss` for the full
documentation.
name: name of the head. If provided, summary and metrics keys will be
suffixed by `"/" + name`. Also used as `name_scope` when creating ops.
Returns:
An instance of `_Head` for poisson regression.
Raises:
ValueError: If `label_dimension` or `loss_reduction` is invalid.
"""
def _poisson_loss(labels, logits):
return nn.log_poisson_loss(targets=labels,
log_input=logits,
compute_full_loss=compute_full_loss)
return head_lib._regression_head( # pylint:disable=protected-access
weight_column=weight_column,
label_dimension=label_dimension,
loss_reduction=loss_reduction,
loss_fn=_poisson_loss,
inverse_link_fn=math_ops.exp,
name=name)
def logistic_regression_head(
weight_column=None,
loss_reduction=losses.Reduction.SUM_OVER_BATCH_SIZE,
name=None):
"""Creates a `_Head` for logistic regression.
Uses `sigmoid_cross_entropy_with_logits` loss, which is the same as
`binary_classification_head`. The differences compared to
`binary_classification_head` are:
* Does not support `label_vocabulary`. Instead, labels must be float in the
range [0, 1].
* Does not calculate some metrics that do not make sense, such as AUC.
* In `PREDICT` mode, only returns logits and predictions
(`=tf.sigmoid(logits)`), whereas `binary_classification_head` also returns
probabilities, classes, and class_ids.
* Export output defaults to `RegressionOutput`, whereas
`binary_classification_head` defaults to `PredictOutput`.
The head expects `logits` with shape `[D0, D1, ... DN, 1]`.
In many applications, the shape is `[batch_size, 1]`.
The `labels` shape must match `logits`, namely
`[D0, D1, ... DN]` or `[D0, D1, ... DN, 1]`.
If `weight_column` is specified, weights must be of shape
`[D0, D1, ... DN]` or `[D0, D1, ... DN, 1]`.
This is implemented as a generalized linear model, see
https://en.wikipedia.org/wiki/Generalized_linear_model.
The head can be used with a canned estimator. Example:
```python
my_head = tf.contrib.estimator.logistic_regression_head()
my_estimator = tf.contrib.estimator.DNNEstimator(
head=my_head,
hidden_units=...,
feature_columns=...)
```
It can also be used with a custom `model_fn`. Example:
```python
def _my_model_fn(features, labels, mode):
my_head = tf.contrib.estimator.logistic_regression_head()
logits = tf.keras.Model(...)(features)
return my_head.create_estimator_spec(
features=features,
mode=mode,
labels=labels,
optimizer=tf.AdagradOptimizer(learning_rate=0.1),
logits=logits)
my_estimator = tf.estimator.Estimator(model_fn=_my_model_fn)
```
Args:
weight_column: A string or a `_NumericColumn` created by
`tf.feature_column.numeric_column` defining feature column representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example.
loss_reduction: One of `tf.losses.Reduction` except `NONE`. Describes how to
reduce training loss over batch and label dimension. Defaults to
`SUM_OVER_BATCH_SIZE`, namely weighted sum of losses divided by
`batch size * label_dimension`. See `tf.losses.Reduction`.
name: name of the head. If provided, summary and metrics keys will be
suffixed by `"/" + name`. Also used as `name_scope` when creating ops.
Returns:
An instance of `_Head` for logistic regression.
Raises:
ValueError: If `loss_reduction` is invalid.
"""
def _logistic_loss(labels, logits):
labels = head_lib._assert_range( # pylint:disable=protected-access
labels,
n_classes=2,
message='Labels must be in range [0, 1]')
return nn.sigmoid_cross_entropy_with_logits(labels=labels,
logits=logits)
return head_lib._regression_head( # pylint:disable=protected-access
weight_column=weight_column,
label_dimension=1,
loss_reduction=loss_reduction,
loss_fn=_logistic_loss,
inverse_link_fn=math_ops.sigmoid,
name=name)
def __init__(self,
feature_columns,
model_dir=None,
label_dimension=1,
weight_column=None,
optimizer='Ftrl',
config=None,
partitioner=None,
warm_start_from=None,
loss_reduction=losses.Reduction.SUM,
sparse_combiner='sum'):
"""Initializes a `LinearRegressor` instance.
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 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: A string or a `_NumericColumn` created by
`tf.feature_column.numeric_column` defining feature column representing
weights. It is used to down weight or boost examples during training. It
will be multiplied by the loss of the example. If it is a string, it is
used as a key to fetch weight tensor from the `features`. If it is a
`_NumericColumn`, raw tensor is fetched by key `weight_column.key`,
then weight_column.normalizer_fn is applied on it to get weight tensor.
optimizer: An instance of `tf.Optimizer` or
`tf.estimator.experimental.LinearSDCA` used to train the model. Can
also be a string (one of 'Adagrad', 'Adam', 'Ftrl', 'RMSProp', 'SGD'),
or callable. Defaults to FTRL optimizer.
config: `RunConfig` object to configure the runtime settings.
partitioner: Optional. Partitioner for input layer.
warm_start_from: A string filepath to a checkpoint to warm-start from, or
a `WarmStartSettings` object to fully configure warm-starting. If the
string filepath is provided instead of a `WarmStartSettings`, then all
weights and biases are warm-started, and it is assumed that vocabularies
and Tensor names are unchanged.
loss_reduction: One of `tf.losses.Reduction` except `NONE`. Describes how
to reduce training loss over batch. Defaults to `SUM`.
sparse_combiner: A string specifying how to reduce if a categorical column
is multivalent. One of "mean", "sqrtn", and "sum" -- these are
effectively different ways to do example-level normalization, which can
be useful for bag-of-words features. for more details, see
`tf.feature_column.linear_model`.
"""
if isinstance(optimizer, LinearSDCA):
if sparse_combiner != 'sum':
raise ValueError(
'sparse_combiner must be "sum" when optimizer '
'is a LinearSDCA object.')
if not feature_column_v2.is_feature_column_v2(feature_columns):
raise ValueError('V2 feature columns required when optimizer '
'is a LinearSDCA object.')
if label_dimension > 1:
raise ValueError(
'LinearSDCA can only be used with one-dimensional '
'label.')
head = head_lib._regression_head( # pylint: disable=protected-access
label_dimension=label_dimension,
weight_column=weight_column,
loss_reduction=loss_reduction)
def _model_fn(features, labels, mode, config):
"""Call the defined shared _linear_model_fn."""
return _linear_model_fn(features=features,
labels=labels,
mode=mode,
head=head,
feature_columns=tuple(feature_columns
or []),
optimizer=optimizer,
partitioner=partitioner,
config=config,
sparse_combiner=sparse_combiner)
super(LinearRegressor, self).__init__(model_fn=_model_fn,
model_dir=model_dir,
config=config,
warm_start_from=warm_start_from)
