def test_auto_ensemble_estimator_lifecycle(self,
candidate_pool,
want_loss,
max_train_steps=30):
features = {"input_1": [[1., 0.]]}
labels = [[1.]]
run_config = tf.estimator.RunConfig(tf_random_seed=42)
head = regression_head.RegressionHead()
# Always create optimizers in a lambda to prevent error like:
# `RuntimeError: Cannot set `iterations` to a new Variable after the
# Optimizer weights have been created`
optimizer = lambda: tf.keras.optimizers.SGD(lr=.01)
feature_columns = [tf.feature_column.numeric_column("input_1", shape=[2])]
def train_input_fn():
input_features = {}
for key, feature in features.items():
input_features[key] = tf.constant(feature, name=key)
input_labels = tf.constant(labels, name="labels")
return input_features, input_labels
def test_input_fn():
dataset = tf.data.Dataset.from_tensors([tf.constant(features["input_1"])])
input_features = tf.compat.v1.data.make_one_shot_iterator(
dataset).get_next()
return {"input_1": input_features}, None
estimator = AutoEnsembleEstimator(
head=head,
candidate_pool=candidate_pool(head, feature_columns, optimizer),
max_iteration_steps=10,
force_grow=True,
model_dir=self.test_subdirectory,
config=run_config)
# Train for three iterations.
estimator.train(input_fn=train_input_fn, max_steps=max_train_steps)
# Evaluate.
eval_results = estimator.evaluate(input_fn=train_input_fn, steps=1)
self.assertAllClose(max_train_steps, eval_results["global_step"])
self.assertAllClose(want_loss, eval_results["loss"], atol=.3)
# Predict.
predictions = estimator.predict(input_fn=test_input_fn)
for prediction in predictions:
self.assertIsNotNone(prediction["predictions"])
# Export SavedModel.
def serving_input_fn():
"""Input fn for serving export, starting from serialized example."""
serialized_example = tf.compat.v1.placeholder(
dtype=tf.string, shape=(None), name="serialized_example")
for key, value in features.items():
features[key] = tf.constant(value)
return export.SupervisedInputReceiver(
features=features,
labels=tf.constant(labels),
receiver_tensors=serialized_example)
export_dir_base = os.path.join(self.test_subdirectory, "export")
estimator.export_saved_model(
export_dir_base=export_dir_base,
serving_input_receiver_fn=serving_input_fn)
def test_auto_ensemble_estimator_lifecycle(self):
features = {"input_1": [[1., 0.]]}
labels = [[1.]]
run_config = tf.estimator.RunConfig(tf_random_seed=42)
head = tf.contrib.estimator.regression_head(
loss_reduction=tf.losses.Reduction.SUM_OVER_BATCH_SIZE)
optimizer = tf.train.GradientDescentOptimizer(learning_rate=.01)
feature_columns = [
tf.feature_column.numeric_column("input_1", shape=[2])
]
def train_input_fn():
input_features = {}
for key, feature in features.items():
input_features[key] = tf.constant(feature, name=key)
input_labels = tf.constant(labels, name="labels")
return input_features, input_labels
def test_input_fn():
input_features = tf.data.Dataset.from_tensors([
tf.constant(features["input_1"])
]).make_one_shot_iterator().get_next()
return {"input_1": input_features}, None
if hasattr(tf.estimator, "LinearEstimator"):
linear_estimator_fn = tf.estimator.LinearEstimator
else:
linear_estimator_fn = tf.contrib.estimator.LinearEstimator
if hasattr(tf.estimator, "DNNEstimator"):
dnn_estimator_fn = tf.estimator.DNNEstimator
else:
dnn_estimator_fn = tf.contrib.estimator.DNNEstimator
estimator = AutoEnsembleEstimator(
head=head,
candidate_pool=[
linear_estimator_fn(head=head,
feature_columns=feature_columns,
optimizer=optimizer),
dnn_estimator_fn(head=head,
feature_columns=feature_columns,
optimizer=optimizer,
hidden_units=[3]),
],
max_iteration_steps=4,
force_grow=True,
model_dir=self.test_subdirectory,
config=run_config)
# Train.
estimator.train(input_fn=train_input_fn, max_steps=12)
# Evaluate.
eval_results = estimator.evaluate(input_fn=train_input_fn, steps=3)
self.assertIsNotNone(eval_results["loss"])
# Predict.
predictions = estimator.predict(input_fn=test_input_fn)
for prediction in predictions:
self.assertIsNotNone(prediction["predictions"])
# Export SavedModel.
def serving_input_fn():
"""Input fn for serving export, starting from serialized example."""
serialized_example = tf.placeholder(dtype=tf.string,
shape=(None),
name="serialized_example")
for key, value in features.items():
features[key] = tf.constant(value)
return export.SupervisedInputReceiver(
features=features,
labels=tf.constant(labels),
receiver_tensors=serialized_example)
export_dir_base = os.path.join(self.test_subdirectory, "export")
tf.contrib.estimator.export_saved_model_for_mode(
estimator,
export_dir_base=export_dir_base,
input_receiver_fn=serving_input_fn,
mode=tf.estimator.ModeKeys.PREDICT)
def test_auto_ensemble_estimator_lifecycle(self,
candidate_pool,
want_loss,
max_train_steps=30):
features = {"input_1": [[1., 0.]]}
labels = [[1.]]
run_config = tf.estimator.RunConfig(tf_random_seed=42)
head = tf.contrib.estimator.regression_head(
loss_reduction=tf.losses.Reduction.SUM_OVER_BATCH_SIZE)
optimizer = tf.train.GradientDescentOptimizer(learning_rate=.01)
feature_columns = [
tf.feature_column.numeric_column("input_1", shape=[2])
]
def train_input_fn():
input_features = {}
for key, feature in features.items():
input_features[key] = tf.constant(feature, name=key)
input_labels = tf.constant(labels, name="labels")
return input_features, input_labels
def test_input_fn():
input_features = tf.data.Dataset.from_tensors([
tf.constant(features["input_1"])
]).make_one_shot_iterator().get_next()
return {"input_1": input_features}, None
estimator = AutoEnsembleEstimator(head=head,
candidate_pool=candidate_pool(
head, feature_columns,
optimizer),
max_iteration_steps=10,
force_grow=True,
model_dir=self.test_subdirectory,
config=run_config)
# Train for three iterations.
estimator.train(input_fn=train_input_fn, max_steps=max_train_steps)
# Evaluate.
eval_results = estimator.evaluate(input_fn=train_input_fn, steps=1)
self.assertAllClose(max_train_steps, eval_results["global_step"])
self.assertAllClose(want_loss, eval_results["loss"], atol=.3)
# Predict.
predictions = estimator.predict(input_fn=test_input_fn)
for prediction in predictions:
self.assertIsNotNone(prediction["predictions"])
# Export SavedModel.
def serving_input_fn():
"""Input fn for serving export, starting from serialized example."""
serialized_example = tf.placeholder(dtype=tf.string,
shape=(None),
name="serialized_example")
for key, value in features.items():
features[key] = tf.constant(value)
return export.SupervisedInputReceiver(
features=features,
labels=tf.constant(labels),
receiver_tensors=serialized_example)
export_dir_base = os.path.join(self.test_subdirectory, "export")
export_saved_model_fn = getattr(estimator, "export_saved_model", None)
if not callable(export_saved_model_fn):
export_saved_model_fn = estimator.export_savedmodel
export_saved_model_fn(export_dir_base=export_dir_base,
serving_input_receiver_fn=serving_input_fn)
def test_auto_ensemble_estimator_lifecycle(self, list_candidate_pool):
features = {"input_1": [[1., 0.]]}
labels = [[1.]]
run_config = tf.estimator.RunConfig(tf_random_seed=42)
head = tf.contrib.estimator.regression_head(
loss_reduction=tf.losses.Reduction.SUM_OVER_BATCH_SIZE)
optimizer = tf.train.GradientDescentOptimizer(learning_rate=.01)
feature_columns = [tf.feature_column.numeric_column("input_1", shape=[2])]
def train_input_fn():
input_features = {}
for key, feature in features.items():
input_features[key] = tf.constant(feature, name=key)
input_labels = tf.constant(labels, name="labels")
return input_features, input_labels
def test_input_fn():
input_features = tf.data.Dataset.from_tensors([
tf.constant(features["input_1"])
]).make_one_shot_iterator().get_next()
return {"input_1": input_features}, None
if hasattr(tf.estimator, "LinearEstimator"):
linear_estimator_fn = tf.estimator.LinearEstimator
else:
linear_estimator_fn = tf.contrib.estimator.LinearEstimator
if hasattr(tf.estimator, "DNNEstimator"):
dnn_estimator_fn = tf.estimator.DNNEstimator
else:
dnn_estimator_fn = tf.contrib.estimator.DNNEstimator
candidate_pool = {
"linear":
linear_estimator_fn(
head=head, feature_columns=feature_columns,
optimizer=optimizer),
"dnn":
dnn_estimator_fn(
head=head,
feature_columns=feature_columns,
optimizer=optimizer,
hidden_units=[3])
}
if list_candidate_pool:
candidate_pool = [candidate_pool[k] for k in sorted(candidate_pool)]
estimator = AutoEnsembleEstimator(
head=head,
candidate_pool=candidate_pool,
max_iteration_steps=10,
force_grow=True,
model_dir=self.test_subdirectory,
config=run_config)
# Train for three iterations.
estimator.train(input_fn=train_input_fn, max_steps=30)
# Evaluate.
eval_results = estimator.evaluate(input_fn=train_input_fn, steps=3)
want_loss = .209
if tf_compat.version_greater_or_equal("1.10.0") and (
not tf_compat.version_greater_or_equal("1.12.0")):
# Only TF 1.10 and 1.11.
want_loss = .079514
self.assertAllClose(want_loss, eval_results["loss"], atol=.05)
# Predict.
predictions = estimator.predict(input_fn=test_input_fn)
for prediction in predictions:
self.assertIsNotNone(prediction["predictions"])
# Export SavedModel.
def serving_input_fn():
"""Input fn for serving export, starting from serialized example."""
serialized_example = tf.placeholder(
dtype=tf.string, shape=(None), name="serialized_example")
for key, value in features.items():
features[key] = tf.constant(value)
return export.SupervisedInputReceiver(
features=features,
labels=tf.constant(labels),
receiver_tensors=serialized_example)
export_dir_base = os.path.join(self.test_subdirectory, "export")
export_saved_model_fn = getattr(estimator, "export_saved_model", None)
if not callable(export_saved_model_fn):
export_saved_model_fn = estimator.export_savedmodel
export_saved_model_fn(
export_dir_base=export_dir_base,
serving_input_receiver_fn=serving_input_fn)