def export(self,
export_dir,
input_fn,
signature_fn=None,
input_feature_key=None,
default_batch_size=1):
"""See BaseEstimator.export."""
# Reset model function with basic device assigner.
# Servo doesn't support distributed inference
# but it will try to respect device assignments if they're there.
# pylint: disable=protected-access
orig_model_fn = self._estimator._model_fn
self._estimator._model_fn = get_model_fn(
self.params, self.graph_builder_class,
tensor_forest.RandomForestDeviceAssigner(),
weights_name=self.weights_name)
result = self._estimator.export(
export_dir=export_dir,
input_fn=input_fn,
input_feature_key=input_feature_key,
use_deprecated_input_fn=False,
signature_fn=(signature_fn or
(export.regression_signature_fn
if self.params.regression else
export.classification_signature_fn_with_prob)),
default_batch_size=default_batch_size,
prediction_key=eval_metrics.INFERENCE_PROB_NAME)
self._estimator._model_fn = orig_model_fn
# pylint: enable=protected-access
return result
def __init__(self,
params,
device_assigner=None,
model_dir=None,
graph_builder_class=tensor_forest.RandomForestGraphs,
master='',
accuracy_metric=None,
tf_random_seed=None,
config=None,
feature_engineering_fn=None):
self.params = params.fill()
self.accuracy_metric = (
accuracy_metric
or ('r2' if self.params.regression else 'accuracy'))
self.data_feeder = None
self.device_assigner = (device_assigner
or tensor_forest.RandomForestDeviceAssigner())
self.graph_builder_class = graph_builder_class
self.training_args = {}
self.construction_args = {}
self._feature_engineering_fn = (feature_engineering_fn
or (lambda features, targets:
(features, targets)))
super(TensorForestEstimator, self).__init__(model_dir=model_dir,
config=config)
def __init__(self,
params,
device_assigner=None,
model_dir=None,
graph_builder_class=tensor_forest.RandomForestGraphs,
master='',
accuracy_metric=None,
tf_random_seed=None,
continue_training=False,
verbose=1,
max_to_keep=5,
save_checkpoint_secs=300):
self.params = params.fill()
self.accuracy_metric = (
accuracy_metric
or ('r2' if self.params.regression else 'accuracy'))
self.data_feeder = None
self.device_assigner = (device_assigner
or tensor_forest.RandomForestDeviceAssigner())
self.graph_builder_class = graph_builder_class
self.training_args = {}
self.construction_args = {}
config = run_config.RunConfig(
master=master,
tf_random_seed=(tf_random_seed or int(
(time.time() * 1000) % 1000)),
save_checkpoints_secs=save_checkpoint_secs,
keep_checkpoint_max=max_to_keep)
super(TensorForestEstimator, self).__init__(model_dir=model_dir,
config=config)
def __init__(self,
params,
device_assigner=None,
optimizer_class=adagrad.AdagradOptimizer,
**kwargs):
self.device_assigner = (
device_assigner or tensor_forest.RandomForestDeviceAssigner())
self.params = params
self.optimizer = optimizer_class(self.params.learning_rate)
self.is_regression = params.regression
self.regularizer = None
if params.regularization == "l1":
self.regularizer = layers.l1_regularizer(
self.params.regularization_strength)
elif params.regularization == "l2":
self.regularizer = layers.l2_regularizer(
self.params.regularization_strength)
def __init__(self, params, layer_num, device_assigner, *args, **kwargs):
self.layer_num = layer_num
self.device_assigner = (device_assigner
or tensor_forest.RandomForestDeviceAssigner())
self.params = params
self._define_vars(params, **kwargs)
