def _test_parsed_sequence_example(
self, col_name, col_fn, col_arg, shape, values):
"""Helper function to check that each FeatureColumn parses correctly.
Args:
col_name: string, name to give to the feature column. Should match
the name that the column will parse out of the features dict.
col_fn: function used to create the feature column. For example,
sequence_numeric_column.
col_arg: second arg that the target feature column is expecting.
shape: the expected dense_shape of the feature after parsing into
a SparseTensor.
values: the expected values at index [0, 2, 6] of the feature
after parsing into a SparseTensor.
"""
example = _make_sequence_example()
columns = [
fc.categorical_column_with_identity('int_ctx', num_buckets=100),
fc.numeric_column('float_ctx'),
col_fn(col_name, col_arg)
]
context, seq_features = parsing_ops.parse_single_sequence_example(
example.SerializeToString(),
context_features=fc.make_parse_example_spec(columns[:2]),
sequence_features=fc.make_parse_example_spec(columns[2:]))
with self.cached_session() as sess:
ctx_result, seq_result = sess.run([context, seq_features])
self.assertEqual(list(seq_result[col_name].dense_shape), shape)
self.assertEqual(
list(seq_result[col_name].values[[0, 2, 6]]), values)
self.assertEqual(list(ctx_result['int_ctx'].dense_shape), [1])
self.assertEqual(ctx_result['int_ctx'].values[0], 5)
self.assertEqual(list(ctx_result['float_ctx'].shape), [1])
self.assertAlmostEqual(ctx_result['float_ctx'][0], 123.6, places=1)
def _parse_example(example):
ctx, seq = parsing_ops.parse_single_sequence_example(
example,
context_features=fc.make_parse_example_spec(ctx_cols),
sequence_features=fc.make_parse_example_spec(seq_cols))
ctx.update(seq)
return ctx
def _test_complete_flow(self, train_input_fn, eval_input_fn,
predict_input_fn, input_dimension, label_dimension,
prediction_length):
feature_columns = [
feature_column_lib.numeric_column('x', shape=(input_dimension, ))
]
est = self._linear_regressor_fn(feature_columns=feature_columns,
label_dimension=label_dimension,
model_dir=self._model_dir)
# TRAIN
# learn y = x
est.train(train_input_fn, steps=200)
# EVALUTE
scores = est.evaluate(eval_input_fn)
self.assertEqual(200, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn(metric_keys.MetricKeys.LOSS, six.iterkeys(scores))
# PREDICT
predictions = np.array(
[x['predictions'] for x in est.predict(predict_input_fn)])
self.assertAllEqual((prediction_length, label_dimension),
predictions.shape)
# EXPORT
feature_spec = feature_column_lib.make_parse_example_spec(
feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = est.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _test_complete_flow(self, train_input_fn, eval_input_fn,
predict_input_fn, input_dimension, label_dimension,
batch_size):
feature_columns = [
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[ops.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 = 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_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _test_complete_flow(
self, train_input_fn, eval_input_fn, predict_input_fn, input_dimension,
label_dimension, batch_size):
feature_columns = [
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)
# EVALUTE
scores = est.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops.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 = 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_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _test_complete_flow(self, train_input_fn, eval_input_fn, predict_input_fn,
input_dimension, label_dimension, batch_size):
feature_columns = [feature_column.numeric_column('x', shape=(input_dimension,))]
est = dnn.DNNRegressorV2(
hidden_units=(2, 2),
feature_columns=feature_columns,
label_dimension=label_dimension,
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[ops.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 = 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(gfile.Exists(export_dir))
def _test_complete_flow(
self, train_input_fn, eval_input_fn, predict_input_fn, input_dimension,
n_classes, batch_size):
feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension,))]
est = dnn.DNNClassifier(
hidden_units=(2, 2),
feature_columns=feature_columns,
n_classes=n_classes,
model_dir=self._model_dir)
# TRAIN
num_steps = 10
est.train(train_input_fn, steps=num_steps)
# EVALUTE
scores = est.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
# PREDICT
predicted_proba = np.array([
x[prediction_keys.PredictionKeys.PROBABILITIES]
for x in est.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, n_classes), predicted_proba.shape)
# EXPORT
feature_spec = 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_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _test_complete_flow(self, train_input_fn, eval_input_fn, predict_input_fn,
input_dimension, n_classes, batch_size):
feature_columns = [
feature_column._numeric_column('x', shape=(input_dimension,))
]
est = dnn_with_layer_annotations.DNNClassifierWithLayerAnnotations(
hidden_units=(2, 2),
feature_columns=feature_columns,
n_classes=n_classes,
model_dir=self._model_dir)
# TRAIN
num_steps = 10
est.train(train_input_fn, steps=num_steps)
# EVALUTE
scores = est.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
# PREDICT
predicted_proba = np.array([
x[prediction_keys.PredictionKeys.PROBABILITIES]
for x in est.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, n_classes), predicted_proba.shape)
# EXPORT
feature_spec = 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_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
placeholders = {}
placeholders[feature_keys.TrainEvalFeatures.TIMES] = (
array_ops.placeholder(
name=feature_keys.TrainEvalFeatures.TIMES,
dtype=dtypes.int64,
shape=[default_batch_size, default_series_length]))
# Values are only necessary when filtering. For prediction the default
# value will be ignored.
placeholders[feature_keys.TrainEvalFeatures.VALUES] = (
array_ops.placeholder_with_default(
name=feature_keys.TrainEvalFeatures.VALUES,
input=array_ops.zeros(
shape=[
default_batch_size
if default_batch_size else 0, default_series_length
if default_series_length else 0, self._model.num_features
],
dtype=self._model.dtype),
shape=(default_batch_size, default_series_length,
self._model.num_features)))
if self._model.exogenous_feature_columns:
with ops.Graph().as_default():
# Default placeholders have only an unknown batch dimension. Make them
# in a separate graph, then splice in the series length to the shapes
# and re-create them in the outer graph.
parsed_features = (
feature_column.make_parse_example_spec(
self._model.exogenous_feature_columns))
placeholder_features = parsing_ops.parse_example(
serialized=array_ops.placeholder(
shape=[None], dtype=dtypes.string),
features=parsed_features)
exogenous_feature_shapes = {
key: (value.get_shape(), value.dtype) for key, value
in placeholder_features.items()}
for feature_key, (batch_only_feature_shape, value_dtype) in (
exogenous_feature_shapes.items()):
batch_only_feature_shape = (
batch_only_feature_shape.with_rank_at_least(1).as_list())
feature_shape = ([default_batch_size, default_series_length]
+ batch_only_feature_shape[1:])
placeholders[feature_key] = array_ops.placeholder(
dtype=value_dtype, name=feature_key, shape=feature_shape)
# Models may not know the shape of their state without creating some
# variables/ops. Avoid polluting the default graph by making a new one. We
# use only static metadata from the returned Tensors.
with ops.Graph().as_default():
self._model.initialize_graph()
model_start_state = self._model.get_start_state()
for prefixed_state_name, state_tensor in ts_head_lib.state_to_dictionary(
model_start_state).items():
state_shape_with_batch = tensor_shape.TensorShape(
(default_batch_size,)).concatenate(state_tensor.get_shape())
placeholders[prefixed_state_name] = array_ops.placeholder(
name=prefixed_state_name,
shape=state_shape_with_batch,
dtype=state_tensor.dtype)
return export_lib.ServingInputReceiver(placeholders, placeholders)
def _test_complete_flow(self, train_input_fn, eval_input_fn, predict_input_fn,
input_dimension, label_dimension, prediction_length):
feature_columns = [
feature_column_lib.numeric_column('x', shape=(input_dimension,))
]
est = _baseline_estimator_fn(
label_dimension=label_dimension,
model_dir=self._model_dir)
# TRAIN
# learn y = x
est.train(train_input_fn, steps=200)
# EVALUTE
scores = est.evaluate(eval_input_fn)
self.assertEqual(200, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn(metric_keys.MetricKeys.LOSS, six.iterkeys(scores))
# PREDICT
predictions = np.array(
[x['predictions'] for x in est.predict(predict_input_fn)])
self.assertAllEqual((prediction_length, label_dimension), predictions.shape)
# EXPORT
feature_spec = feature_column_lib.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = est.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def test_complete_flow_with_mode(self, distribution):
label_dimension = 2
input_dimension = label_dimension
batch_size = 10
data = np.linspace(0.,
2.,
batch_size * label_dimension,
dtype=np.float32)
data = data.reshape(batch_size, label_dimension)
train_input_fn = self.dataset_input_fn(
x={'x': data},
y=data,
batch_size=batch_size // len(distribution.worker_devices),
shuffle=True)
eval_input_fn = numpy_io.numpy_input_fn(x={'x': data},
y=data,
batch_size=batch_size,
shuffle=False)
predict_input_fn = numpy_io.numpy_input_fn(x={'x': data},
batch_size=batch_size,
shuffle=False)
linear_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension, ))
]
dnn_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension, ))
]
feature_columns = linear_feature_columns + dnn_feature_columns
estimator = dnn_linear_combined.DNNLinearCombinedRegressor(
linear_feature_columns=linear_feature_columns,
dnn_hidden_units=(2, 2),
dnn_feature_columns=dnn_feature_columns,
label_dimension=label_dimension,
model_dir=self._model_dir,
# TODO(isaprykin): Work around the colocate_with error.
dnn_optimizer=adagrad.AdagradOptimizer(0.001),
linear_optimizer=adagrad.AdagradOptimizer(0.001),
config=run_config.RunConfig(train_distribute=distribution))
num_steps = 10
estimator.train(train_input_fn, steps=num_steps)
scores = estimator.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
predictions = np.array([
x[prediction_keys.PredictionKeys.PREDICTIONS]
for x in estimator.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, label_dimension), predictions.shape)
feature_spec = feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = estimator.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def test_complete_flow_with_mode(self, distribution):
label_dimension = 2
input_dimension = label_dimension
batch_size = 10
data = np.linspace(0., 2., batch_size * label_dimension, dtype=np.float32)
data = data.reshape(batch_size, label_dimension)
train_input_fn = self.dataset_input_fn(
x={'x': data},
y=data,
batch_size=batch_size // len(distribution.worker_devices),
shuffle=True)
eval_input_fn = self.dataset_input_fn(
x={'x': data},
y=data,
batch_size=batch_size // len(distribution.worker_devices),
shuffle=False)
predict_input_fn = numpy_io.numpy_input_fn(
x={'x': data}, batch_size=batch_size, shuffle=False)
linear_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension,))
]
dnn_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension,))
]
feature_columns = linear_feature_columns + dnn_feature_columns
estimator = dnn_linear_combined.DNNLinearCombinedRegressor(
linear_feature_columns=linear_feature_columns,
dnn_hidden_units=(2, 2),
dnn_feature_columns=dnn_feature_columns,
label_dimension=label_dimension,
model_dir=self._model_dir,
# TODO(isaprykin): Work around the colocate_with error.
dnn_optimizer=adagrad.AdagradOptimizer(0.001),
linear_optimizer=adagrad.AdagradOptimizer(0.001),
config=run_config.RunConfig(
train_distribute=distribution, eval_distribute=distribution))
num_steps = 10
estimator.train(train_input_fn, steps=num_steps)
scores = estimator.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
predictions = np.array([
x[prediction_keys.PredictionKeys.PREDICTIONS]
for x in estimator.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, label_dimension), predictions.shape)
feature_spec = feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = estimator.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
placeholders = {}
time_placeholder = array_ops.placeholder(
name=feature_keys.TrainEvalFeatures.TIMES,
dtype=dtypes.int64,
shape=[default_batch_size, default_series_length])
placeholders[
feature_keys.TrainEvalFeatures.TIMES] = time_placeholder
# Values are only necessary when filtering. For prediction the default
# value will be ignored.
placeholders[feature_keys.TrainEvalFeatures.VALUES] = (
array_ops.placeholder_with_default(
name=feature_keys.TrainEvalFeatures.VALUES,
input=array_ops.zeros(shape=[
default_batch_size if default_batch_size else 0,
default_series_length if default_series_length else 0,
self._model.num_features
],
dtype=self._model.dtype),
shape=(default_batch_size, default_series_length,
self._model.num_features)))
if self._model.exogenous_feature_columns:
with ops.Graph().as_default():
# Default placeholders have only an unknown batch dimension. Make them
# in a separate graph, then splice in the series length to the shapes
# and re-create them in the outer graph.
parsed_features = (feature_column.make_parse_example_spec(
self._model.exogenous_feature_columns))
placeholder_features = parsing_ops.parse_example(
serialized=array_ops.placeholder(shape=[None],
dtype=dtypes.string),
features=parsed_features)
exogenous_feature_shapes = {
key: (value.get_shape(), value.dtype)
for key, value in placeholder_features.items()
}
for feature_key, (batch_only_feature_shape, value_dtype) in (
exogenous_feature_shapes.items()):
batch_only_feature_shape = (
batch_only_feature_shape.with_rank_at_least(
1).as_list())
feature_shape = (
[default_batch_size, default_series_length] +
batch_only_feature_shape[1:])
placeholders[feature_key] = array_ops.placeholder(
dtype=value_dtype,
name=feature_key,
shape=feature_shape)
batch_size_tensor = array_ops.shape(time_placeholder)[0]
placeholders.update(
self._model_start_state_placeholders(
batch_size_tensor, static_batch_size=default_batch_size))
return export_lib.ServingInputReceiver(placeholders, placeholders)
def test_complete_flow(self):
label_dimension = 2
batch_size = 10
feature_columns = [feature_column.numeric_column('x', shape=(2,))]
est = dnn.DNNRegressor(
hidden_units=(2, 2),
feature_columns=feature_columns,
label_dimension=label_dimension,
model_dir=self._model_dir)
data = np.linspace(0., 2., batch_size * label_dimension, dtype=np.float32)
data = data.reshape(batch_size, label_dimension)
# TRAIN
# learn y = x
train_input_fn = numpy_io.numpy_input_fn(
x={'x': data},
y=data,
batch_size=batch_size,
num_epochs=None,
shuffle=True)
num_steps = 200
est.train(train_input_fn, steps=num_steps)
# EVALUTE
eval_input_fn = numpy_io.numpy_input_fn(
x={'x': data},
y=data,
batch_size=batch_size,
shuffle=False)
scores = est.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
# PREDICT
predict_input_fn = numpy_io.numpy_input_fn(
x={'x': data},
batch_size=batch_size,
shuffle=False)
predictions = np.array([
x[prediction_keys.PredictionKeys.PREDICTIONS]
for x in est.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, label_dimension), predictions.shape)
# TODO(ptucker): Deterministic test for predicted values?
# EXPORT
feature_spec = 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_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
placeholders = {}
time_placeholder = array_ops.placeholder(
name=feature_keys.TrainEvalFeatures.TIMES,
dtype=dtypes.int64,
shape=[default_batch_size, default_series_length])
placeholders[feature_keys.TrainEvalFeatures.TIMES] = time_placeholder
# Values are only necessary when filtering. For prediction the default
# value will be ignored.
placeholders[feature_keys.TrainEvalFeatures.VALUES] = (
array_ops.placeholder_with_default(
name=feature_keys.TrainEvalFeatures.VALUES,
input=array_ops.zeros(
shape=[
default_batch_size
if default_batch_size else 0, default_series_length
if default_series_length else 0, self._model.num_features
],
dtype=self._model.dtype),
shape=(default_batch_size, default_series_length,
self._model.num_features)))
if self._model.exogenous_feature_columns:
with ops.Graph().as_default():
# Default placeholders have only an unknown batch dimension. Make them
# in a separate graph, then splice in the series length to the shapes
# and re-create them in the outer graph.
parsed_features = (
feature_column.make_parse_example_spec(
self._model.exogenous_feature_columns))
placeholder_features = parsing_ops.parse_example(
serialized=array_ops.placeholder(
shape=[None], dtype=dtypes.string),
features=parsed_features)
exogenous_feature_shapes = {
key: (value.get_shape(), value.dtype) for key, value
in placeholder_features.items()}
for feature_key, (batch_only_feature_shape, value_dtype) in (
exogenous_feature_shapes.items()):
batch_only_feature_shape = (
batch_only_feature_shape.with_rank_at_least(1).as_list())
feature_shape = ([default_batch_size, default_series_length]
+ batch_only_feature_shape[1:])
placeholders[feature_key] = array_ops.placeholder(
dtype=value_dtype, name=feature_key, shape=feature_shape)
batch_size_tensor = array_ops.shape(time_placeholder)[0]
placeholders.update(
self._model_start_state_placeholders(
batch_size_tensor, static_batch_size=default_batch_size))
return export_lib.ServingInputReceiver(placeholders, placeholders)
def test_complete_flow(self):
label_dimension = 2
batch_size = 10
feature_columns = [feature_column_lib.numeric_column('x', shape=(2, ))]
est = linear.LinearRegressor(feature_columns=feature_columns,
label_dimension=label_dimension,
model_dir=self._model_dir)
data = np.linspace(0.,
2.,
batch_size * label_dimension,
dtype=np.float32)
data = data.reshape(batch_size, label_dimension)
# TRAIN
# learn y = x
train_input_fn = numpy_io.numpy_input_fn(x={'x': data},
y=data,
batch_size=batch_size,
num_epochs=None,
shuffle=True)
est.train(train_input_fn, steps=200)
# EVALUTE
eval_input_fn = numpy_io.numpy_input_fn(x={'x': data},
y=data,
batch_size=batch_size,
num_epochs=1,
shuffle=False)
scores = est.evaluate(eval_input_fn)
self.assertEqual(200, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn(metric_keys.MetricKeys.LOSS, six.iterkeys(scores))
# PREDICT
predict_input_fn = numpy_io.numpy_input_fn(x={'x': data},
y=None,
batch_size=batch_size,
num_epochs=1,
shuffle=False)
predictions = list(
[x['predictions'] for x in est.predict(predict_input_fn)])
self.assertAllClose(data, predictions, atol=0.01)
# EXPORT
feature_spec = feature_column_lib.make_parse_example_spec(
feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = est.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _test_complete_flow(self, train_input_fn, eval_input_fn, predict_input_fn,
input_dimension, label_dimension, batch_size,
fc_impl):
linear_feature_columns = [
fc_impl.numeric_column('x', shape=(input_dimension,))
]
dnn_feature_columns = [
fc_impl.numeric_column('x', shape=(input_dimension,))
]
feature_columns = linear_feature_columns + dnn_feature_columns
feature_spec = feature_column.make_parse_example_spec(feature_columns)
self._test_complete_flow_helper(linear_feature_columns, dnn_feature_columns,
feature_spec, train_input_fn, eval_input_fn,
predict_input_fn, input_dimension,
label_dimension, batch_size)
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
times_column = feature_column.numeric_column(
key=feature_keys.TrainEvalFeatures.TIMES, dtype=dtypes.int64)
values_column = feature_column.numeric_column(
key=feature_keys.TrainEvalFeatures.VALUES,
dtype=values_input_dtype,
shape=(self._model.num_features, ))
parsed_features_no_sequence = (
feature_column.make_parse_example_spec(
list(self._model.exogenous_feature_columns) +
[times_column, values_column]))
parsed_features = {}
for key, feature_spec in parsed_features_no_sequence.items():
if isinstance(feature_spec, parsing_ops.FixedLenFeature):
if key == feature_keys.TrainEvalFeatures.VALUES:
parsed_features[key] = feature_spec._replace(
shape=((values_proto_length, ) +
feature_spec.shape))
else:
parsed_features[key] = feature_spec._replace(
shape=((filtering_length + prediction_length, ) +
feature_spec.shape))
elif feature_spec.dtype == dtypes.string:
parsed_features[key] = parsing_ops.FixedLenFeature(
shape=(filtering_length + prediction_length, ),
dtype=dtypes.string)
else: # VarLenFeature
raise ValueError(
"VarLenFeatures not supported, got %s for key %s" %
(feature_spec, key))
tfexamples = array_ops.placeholder(shape=[default_batch_size],
dtype=dtypes.string,
name="input")
features = parsing_ops.parse_example(serialized=tfexamples,
features=parsed_features)
features[feature_keys.TrainEvalFeatures.TIMES] = array_ops.squeeze(
features[feature_keys.TrainEvalFeatures.TIMES], axis=-1)
features[feature_keys.TrainEvalFeatures.VALUES] = math_ops.cast(
features[feature_keys.TrainEvalFeatures.VALUES],
dtype=self._model.dtype)[:, :filtering_length]
features.update(
self._model_start_state_placeholders(
batch_size_tensor=array_ops.shape(
features[feature_keys.TrainEvalFeatures.TIMES])[0],
static_batch_size=default_batch_size))
return export_lib.ServingInputReceiver(features,
{"examples": tfexamples})
def _test_complete_flow_mix2(self, train_input_fn, eval_input_fn,
predict_input_fn, input_dimension,
label_dimension, batch_size, fc_impl):
del fc_impl
linear_feature_columns = [
feature_column_v2.numeric_column('x', shape=(input_dimension,))
]
dnn_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension,))
]
feature_columns = linear_feature_columns + dnn_feature_columns
feature_spec = feature_column.make_parse_example_spec(feature_columns)
self._test_complete_flow_helper(linear_feature_columns, dnn_feature_columns,
feature_spec, train_input_fn, eval_input_fn,
predict_input_fn, input_dimension,
label_dimension, batch_size)
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
times_column = feature_column.numeric_column(
key=feature_keys.TrainEvalFeatures.TIMES, dtype=dtypes.int64)
values_column = feature_column.numeric_column(
key=feature_keys.TrainEvalFeatures.VALUES, dtype=values_input_dtype,
shape=(self._model.num_features,))
parsed_features_no_sequence = (
feature_column.make_parse_example_spec(
list(self._model.exogenous_feature_columns)
+ [times_column, values_column]))
parsed_features = {}
for key, feature_spec in parsed_features_no_sequence.items():
if isinstance(feature_spec, parsing_ops.FixedLenFeature):
if key == feature_keys.TrainEvalFeatures.VALUES:
parsed_features[key] = feature_spec._replace(
shape=((values_proto_length,)
+ feature_spec.shape))
else:
parsed_features[key] = feature_spec._replace(
shape=((filtering_length + prediction_length,)
+ feature_spec.shape))
elif feature_spec.dtype == dtypes.string:
parsed_features[key] = parsing_ops.FixedLenFeature(
shape=(filtering_length + prediction_length,),
dtype=dtypes.string)
else: # VarLenFeature
raise ValueError("VarLenFeatures not supported, got %s for key %s"
% (feature_spec, key))
tfexamples = array_ops.placeholder(
shape=[default_batch_size], dtype=dtypes.string, name="input")
features = parsing_ops.parse_example(
serialized=tfexamples,
features=parsed_features)
features[feature_keys.TrainEvalFeatures.TIMES] = array_ops.squeeze(
features[feature_keys.TrainEvalFeatures.TIMES], axis=-1)
features[feature_keys.TrainEvalFeatures.VALUES] = math_ops.cast(
features[feature_keys.TrainEvalFeatures.VALUES],
dtype=self._model.dtype)[:, :filtering_length]
features.update(
self._model_start_state_placeholders(
batch_size_tensor=array_ops.shape(
features[feature_keys.TrainEvalFeatures.TIMES])[0],
static_batch_size=default_batch_size))
return export_lib.ServingInputReceiver(
features, {"examples": tfexamples})
def _get_exogenous_embedding_shape(self):
"""Computes the shape of the vector returned by _process_exogenous_features.
Returns:
The shape as a list. Does not include a batch dimension.
"""
if not self._exogenous_feature_columns:
return (0,)
with ops.Graph().as_default():
parsed_features = (
feature_column.make_parse_example_spec(
self._exogenous_feature_columns))
placeholder_features = parsing_ops.parse_example(
serialized=array_ops.placeholder(shape=[None], dtype=dtypes.string),
features=parsed_features)
embedded = feature_column.input_layer(
features=placeholder_features,
feature_columns=self._exogenous_feature_columns)
return embedded.get_shape().as_list()[1:]
def test_complete_flow(self):
label_dimension = 2
batch_size = 10
feature_columns = [
feature_column_lib.numeric_column('x', shape=(2,))
]
est = linear.LinearRegressor(
feature_columns=feature_columns, label_dimension=label_dimension,
model_dir=self._model_dir)
data = np.linspace(0., 2., batch_size * label_dimension, dtype=np.float32)
data = data.reshape(batch_size, label_dimension)
# TRAIN
# learn y = x
train_input_fn = numpy_io.numpy_input_fn(
x={'x': data}, y=data, batch_size=batch_size, num_epochs=None,
shuffle=True)
est.train(train_input_fn, steps=200)
# EVALUTE
eval_input_fn = numpy_io.numpy_input_fn(
x={'x': data}, y=data, batch_size=batch_size, num_epochs=1,
shuffle=False)
scores = est.evaluate(eval_input_fn)
self.assertEqual(200, scores[ops.GraphKeys.GLOBAL_STEP])
self.assertIn(metric_keys.MetricKeys.LOSS, six.iterkeys(scores))
# PREDICT
predict_input_fn = numpy_io.numpy_input_fn(
x={'x': data}, y=None, batch_size=batch_size, num_epochs=1,
shuffle=False)
predictions = list(
[x['predictions'] for x in est.predict(predict_input_fn)])
self.assertAllClose(data, predictions, atol=0.01)
# EXPORT
feature_spec = feature_column_lib.make_parse_example_spec(
feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = est.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def test_complete_flow(self):
n_classes = 3
input_dimension = 2
batch_size = 12
data = np.linspace(0.,
n_classes - 1.,
batch_size * input_dimension,
dtype=np.float32)
x_data = data.reshape(batch_size, input_dimension)
categorical_data = np.random.random_integers(0,
len(x_data),
size=len(x_data))
y_data = np.reshape(self._as_label(data[:batch_size]), (batch_size, 1))
train_input_fn = numpy_io.numpy_input_fn(x={
'x': x_data,
'categories': categorical_data
},
y=y_data,
batch_size=batch_size,
num_epochs=None,
shuffle=True)
eval_input_fn = numpy_io.numpy_input_fn(x={
'x': x_data,
'categories': categorical_data
},
y=y_data,
batch_size=batch_size,
shuffle=False)
predict_input_fn = numpy_io.numpy_input_fn(x={
'x':
x_data,
'categories':
categorical_data
},
batch_size=batch_size,
shuffle=False)
feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension, )),
feature_column.embedding_column(
feature_column.categorical_column_with_vocabulary_list(
'categories',
vocabulary_list=np.linspace(0.,
len(x_data),
len(x_data),
dtype=np.int64)), 1)
]
estimator = dnn.DNNClassifier(hidden_units=(2, 2),
feature_columns=feature_columns,
n_classes=n_classes,
model_dir=self._model_dir)
def optimizer_fn():
return optimizers.get_optimizer_instance('Adagrad',
learning_rate=0.05)
estimator = estimator_lib.Estimator(
model_fn=replicate_model_fn.replicate_model_fn(
estimator.model_fn,
optimizer_fn,
devices=['/gpu:0', '/gpu:1', '/gpu:2']),
model_dir=estimator.model_dir,
config=estimator.config,
params=estimator.params)
num_steps = 10
estimator.train(train_input_fn, steps=num_steps)
scores = estimator.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops_lib.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
predicted_proba = np.array([
x[prediction_keys.PredictionKeys.PROBABILITIES]
for x in estimator.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, n_classes), predicted_proba.shape)
feature_spec = feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = estimator.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def input_layer_with_layer_annotations(features,
feature_columns,
weight_collections=None,
trainable=True,
cols_to_vars=None,
cols_to_output_tensors=None):
"""Returns a dense `Tensor` as input layer based on given `feature_columns`.
Generally a single example in training data is described with
FeatureColumns.
At the first layer of the model, this column oriented data should be
converted
to a single `Tensor`.
This is like tf.feature_column.input_layer, except with added
Integrated-Gradient annotations.
Args:
features: A mapping from key to tensors. `_FeatureColumn`s look up via
these keys. For example `numeric_column('price')` will look at 'price'
key in this dict. Values can be a `SparseTensor` or a `Tensor` depends
on corresponding `_FeatureColumn`.
feature_columns: An iterable containing the FeatureColumns to use as
inputs to your model. All items should be instances of classes derived
from `_DenseColumn` such as `numeric_column`, `embedding_column`,
`bucketized_column`, `indicator_column`. If you have categorical
features, you can wrap them with an `embedding_column` or
`indicator_column`.
weight_collections: A list of collection names to which the Variable will
be added. Note that variables will also be added to collections
`tf.GraphKeys.GLOBAL_VARIABLES` and `ops.GraphKeys.MODEL_VARIABLES`.
trainable: If `True` also add the variable to the graph collection
`GraphKeys.TRAINABLE_VARIABLES` (see `tf.Variable`).
cols_to_vars: If not `None`, must be a dictionary that will be filled with
a mapping from `_FeatureColumn` to list of `Variable`s. For example,
after the call, we might have cols_to_vars = {_EmbeddingColumn(
categorical_column=_HashedCategoricalColumn( key='sparse_feature',
hash_bucket_size=5, dtype=tf.string), dimension=10): [<tf.Variable
'some_variable:0' shape=(5, 10), <tf.Variable 'some_variable:1'
shape=(5, 10)]} If a column creates no variables, its value will be an
empty list.
cols_to_output_tensors: If not `None`, must be a dictionary that will be
filled with a mapping from '_FeatureColumn' to the associated output
`Tensor`s.
Returns:
A `Tensor` which represents input layer of a model. Its shape
is (batch_size, first_layer_dimension) and its dtype is `float32`.
first_layer_dimension is determined based on given `feature_columns`.
Raises:
ValueError: features and feature_columns have different lengths.
"""
local_cols_to_output_tensors = {}
input_layer = original_input_layer(
features=features,
feature_columns=feature_columns,
weight_collections=weight_collections,
trainable=trainable,
cols_to_vars=cols_to_vars,
cols_to_output_tensors=local_cols_to_output_tensors)
if cols_to_output_tensors is not None:
cols_to_output_tensors = local_cols_to_output_tensors
if mode and mode == model_fn.ModeKeys.PREDICT:
# Only annotate in PREDICT mode.
# Annotate features.
# These are the parsed Tensors, before embedding.
# Only annotate features used by FeatureColumns.
# We figure which ones are used by FeatureColumns by creating a parsing
# spec and looking at the keys.
spec = feature_column_lib.make_parse_example_spec(feature_columns)
for key in spec.keys():
tensor = features[key]
ops.add_to_collection(
LayerAnnotationsCollectionNames.keys(
LayerAnnotationsCollectionNames.UNPROCESSED_FEATURES), key)
ops.add_to_collection(
LayerAnnotationsCollectionNames.values(
LayerAnnotationsCollectionNames.UNPROCESSED_FEATURES),
_to_any_wrapped_tensor_info(tensor))
# Annotate feature columns.
for column in feature_columns:
# TODO(cyfoo): Find a better way to serialize and deserialize
# _FeatureColumn.
ops.add_to_collection(LayerAnnotationsCollectionNames.FEATURE_COLUMNS,
serialize_feature_column(column))
for column, tensor in local_cols_to_output_tensors.items():
ops.add_to_collection(
LayerAnnotationsCollectionNames.keys(
LayerAnnotationsCollectionNames.PROCESSED_FEATURES),
column.name)
ops.add_to_collection(
LayerAnnotationsCollectionNames.values(
LayerAnnotationsCollectionNames.PROCESSED_FEATURES),
_to_any_wrapped_tensor_info(tensor))
return input_layer
def test_complete_flow_with_mode(self, distribution, use_train_and_evaluate):
label_dimension = 2
input_dimension = label_dimension
batch_size = 10
data = np.linspace(0., 2., batch_size * label_dimension, dtype=np.float32)
data = data.reshape(batch_size, label_dimension)
train_input_fn = self.dataset_input_fn(
x={'x': data},
y=data,
batch_size=batch_size // len(distribution.worker_devices))
eval_input_fn = self.dataset_input_fn(
x={'x': data},
y=data,
batch_size=batch_size // len(distribution.worker_devices))
predict_input_fn = numpy_io.numpy_input_fn(
x={'x': data}, batch_size=batch_size, shuffle=False)
linear_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension,))
]
dnn_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension,))
]
feature_columns = linear_feature_columns + dnn_feature_columns
session_config = config_pb2.ConfigProto(
log_device_placement=True, allow_soft_placement=True)
estimator = dnn_linear_combined.DNNLinearCombinedRegressor(
linear_feature_columns=linear_feature_columns,
dnn_hidden_units=(2, 2),
dnn_feature_columns=dnn_feature_columns,
label_dimension=label_dimension,
model_dir=self._model_dir,
dnn_optimizer=adam.Adam(0.001),
linear_optimizer=adam.Adam(0.001),
config=run_config.RunConfig(
train_distribute=distribution,
eval_distribute=distribution,
session_config=session_config))
num_steps = 2
if use_train_and_evaluate:
scores, _ = training.train_and_evaluate(
estimator, training.TrainSpec(train_input_fn, max_steps=num_steps),
training.EvalSpec(eval_input_fn))
else:
estimator.train(train_input_fn, steps=num_steps)
scores = estimator.evaluate(eval_input_fn)
self.assertIn('loss', six.iterkeys(scores))
predictions = np.array([
x[prediction_keys.PredictionKeys.PREDICTIONS]
for x in estimator.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, label_dimension), predictions.shape)
feature_spec = feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = estimator.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _complete_flow_with_mode(self, mode):
n_classes = 3
input_dimension = 2
batch_size = 12
data = np.linspace(
0., n_classes - 1., batch_size * input_dimension, dtype=np.float32)
x_data = data.reshape(batch_size, input_dimension)
categorical_data = np.random.random_integers(
0, len(x_data), size=len(x_data))
y_data = np.reshape(self._as_label(data[:batch_size]), (batch_size, 1))
train_input_fn = numpy_io.numpy_input_fn(
x={'x': x_data,
'categories': categorical_data},
y=y_data,
batch_size=batch_size,
num_epochs=None,
shuffle=True)
eval_input_fn = numpy_io.numpy_input_fn(
x={'x': x_data,
'categories': categorical_data},
y=y_data,
batch_size=batch_size,
shuffle=False)
predict_input_fn = numpy_io.numpy_input_fn(
x={'x': x_data,
'categories': categorical_data},
batch_size=batch_size,
shuffle=False)
feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension,)),
feature_column.embedding_column(
feature_column.categorical_column_with_vocabulary_list(
'categories',
vocabulary_list=np.linspace(
0., len(x_data), len(x_data), dtype=np.int64)), 1)
]
estimator = dnn.DNNClassifier(
hidden_units=(2, 2),
feature_columns=feature_columns,
n_classes=n_classes,
model_dir=self._model_dir)
def optimizer_fn():
return optimizers.get_optimizer_instance('Adagrad', learning_rate=0.05)
if not mode: # Use the public `replicate_model_fn`.
model_fn = replicate_model_fn.replicate_model_fn(
estimator.model_fn,
optimizer_fn,
devices=['/gpu:0', '/gpu:1', '/gpu:2'])
else:
model_fn = replicate_model_fn._replicate_model_fn_with_mode(
estimator.model_fn,
optimizer_fn,
devices=['/gpu:0', '/gpu:1', '/gpu:2'],
mode=mode)
estimator = estimator_lib.Estimator(
model_fn=model_fn,
model_dir=estimator.model_dir,
config=estimator.config,
params=estimator.params)
num_steps = 10
estimator.train(train_input_fn, steps=num_steps)
scores = estimator.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops_lib.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
predicted_proba = np.array([
x[prediction_keys.PredictionKeys.PROBABILITIES]
for x in estimator.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, n_classes), predicted_proba.shape)
feature_spec = feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = estimator.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _get_exporter(self, name, fc):
feature_spec = feature_column.make_parse_example_spec(fc)
serving_input_receiver_fn = (
export_lib.build_parsing_serving_input_receiver_fn(feature_spec))
return exporter_lib.LatestExporter(
name, serving_input_receiver_fn=serving_input_receiver_fn)
def input_layer_with_layer_annotations(features,
feature_columns,
weight_collections=None,
trainable=True,
cols_to_vars=None,
scope=None,
cols_to_output_tensors=None,
from_template=False):
"""Returns a dense `Tensor` as input layer based on given `feature_columns`.
Generally a single example in training data is described with
FeatureColumns.
At the first layer of the model, this column oriented data should be
converted
to a single `Tensor`.
This is like tf.feature_column.input_layer, except with added
Integrated-Gradient annotations.
Args:
features: A mapping from key to tensors. `_FeatureColumn`s look up via
these keys. For example `numeric_column('price')` will look at 'price'
key in this dict. Values can be a `SparseTensor` or a `Tensor` depends
on corresponding `_FeatureColumn`.
feature_columns: An iterable containing the FeatureColumns to use as
inputs to your model. All items should be instances of classes derived
from `_DenseColumn` such as `numeric_column`, `embedding_column`,
`bucketized_column`, `indicator_column`. If you have categorical
features, you can wrap them with an `embedding_column` or
`indicator_column`.
weight_collections: A list of collection names to which the Variable will
be added. Note that variables will also be added to collections
`tf.GraphKeys.GLOBAL_VARIABLES` and `ops.GraphKeys.MODEL_VARIABLES`.
trainable: If `True` also add the variable to the graph collection
`GraphKeys.TRAINABLE_VARIABLES` (see `tf.Variable`).
cols_to_vars: If not `None`, must be a dictionary that will be filled with
a mapping from `_FeatureColumn` to list of `Variable`s. For example,
after the call, we might have cols_to_vars = {_EmbeddingColumn(
categorical_column=_HashedCategoricalColumn( key='sparse_feature',
hash_bucket_size=5, dtype=tf.string), dimension=10): [<tf.Variable
'some_variable:0' shape=(5, 10), <tf.Variable 'some_variable:1'
shape=(5, 10)]} If a column creates no variables, its value will be an
empty list.
scope: A name or variable scope to use
cols_to_output_tensors: If not `None`, must be a dictionary that will be
filled with a mapping from '_FeatureColumn' to the associated output
`Tensor`s.
from_template: True if the method is being instantiated from a
`make_template`.
Returns:
A `Tensor` which represents input layer of a model. Its shape
is (batch_size, first_layer_dimension) and its dtype is `float32`.
first_layer_dimension is determined based on given `feature_columns`.
Raises:
ValueError: features and feature_columns have different lengths.
"""
local_cols_to_output_tensors = {}
input_layer = original_input_layer(
features=features,
feature_columns=feature_columns,
weight_collections=weight_collections,
trainable=trainable,
cols_to_vars=cols_to_vars,
scope=scope,
cols_to_output_tensors=local_cols_to_output_tensors,
from_template=from_template)
if cols_to_output_tensors is not None:
cols_to_output_tensors = local_cols_to_output_tensors
# Annotate features.
# These are the parsed Tensors, before embedding.
# Only annotate features used by FeatureColumns.
# We figure which ones are used by FeatureColumns by creating a parsing
# spec and looking at the keys.
spec = feature_column_lib.make_parse_example_spec(feature_columns)
for key in spec.keys():
tensor = ops.convert_to_tensor_or_indexed_slices(features[key])
ops.add_to_collection(
LayerAnnotationsCollectionNames.keys(
LayerAnnotationsCollectionNames.UNPROCESSED_FEATURES), key)
ops.add_to_collection(
LayerAnnotationsCollectionNames.values(
LayerAnnotationsCollectionNames.UNPROCESSED_FEATURES),
_to_any_wrapped_tensor_info(tensor))
# Annotate feature columns.
for column in feature_columns:
# TODO(cyfoo): Find a better way to serialize and deserialize
# _FeatureColumn.
ops.add_to_collection(
LayerAnnotationsCollectionNames.FEATURE_COLUMNS,
serialize_feature_column(column))
for column, tensor in local_cols_to_output_tensors.items():
ops.add_to_collection(
LayerAnnotationsCollectionNames.keys(
LayerAnnotationsCollectionNames.PROCESSED_FEATURES),
column.name)
ops.add_to_collection(
LayerAnnotationsCollectionNames.values(
LayerAnnotationsCollectionNames.PROCESSED_FEATURES),
_to_any_wrapped_tensor_info(tensor))
return input_layer
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
placeholders = {}
time_placeholder = array_ops.placeholder(
name=feature_keys.TrainEvalFeatures.TIMES,
dtype=dtypes.int64,
shape=[default_batch_size, default_series_length])
placeholders[feature_keys.TrainEvalFeatures.TIMES] = time_placeholder
# Values are only necessary when filtering. For prediction the default
# value will be ignored.
placeholders[feature_keys.TrainEvalFeatures.VALUES] = (
array_ops.placeholder_with_default(
name=feature_keys.TrainEvalFeatures.VALUES,
input=array_ops.zeros(
shape=[
default_batch_size
if default_batch_size else 0, default_series_length
if default_series_length else 0, self._model.num_features
],
dtype=self._model.dtype),
shape=(default_batch_size, default_series_length,
self._model.num_features)))
if self._model.exogenous_feature_columns:
with ops.Graph().as_default():
# Default placeholders have only an unknown batch dimension. Make them
# in a separate graph, then splice in the series length to the shapes
# and re-create them in the outer graph.
parsed_features = (
feature_column.make_parse_example_spec(
self._model.exogenous_feature_columns))
placeholder_features = parsing_ops.parse_example(
serialized=array_ops.placeholder(
shape=[None], dtype=dtypes.string),
features=parsed_features)
exogenous_feature_shapes = {
key: (value.get_shape(), value.dtype) for key, value
in placeholder_features.items()}
for feature_key, (batch_only_feature_shape, value_dtype) in (
exogenous_feature_shapes.items()):
batch_only_feature_shape = (
batch_only_feature_shape.with_rank_at_least(1).as_list())
feature_shape = ([default_batch_size, default_series_length]
+ batch_only_feature_shape[1:])
placeholders[feature_key] = array_ops.placeholder(
dtype=value_dtype, name=feature_key, shape=feature_shape)
# Models may not know the shape of their state without creating some
# variables/ops. Avoid polluting the default graph by making a new one. We
# use only static metadata from the returned Tensors.
with ops.Graph().as_default():
self._model.initialize_graph()
# Evaluate the initial state as same-dtype "zero" values. These zero
# constants aren't used, but are necessary for feeding to
# placeholder_with_default for the "cold start" case where state is not
# fed to the model.
def _zeros_like_constant(tensor):
return tensor_util.constant_value(array_ops.zeros_like(tensor))
start_state = nest.map_structure(
_zeros_like_constant, self._model.get_start_state())
batch_size_tensor = array_ops.shape(time_placeholder)[0]
for prefixed_state_name, state in ts_head_lib.state_to_dictionary(
start_state).items():
state_shape_with_batch = tensor_shape.TensorShape(
(default_batch_size,)).concatenate(state.shape)
default_state_broadcast = array_ops.tile(
state[None, ...],
multiples=array_ops.concat(
[batch_size_tensor[None],
array_ops.ones(len(state.shape), dtype=dtypes.int32)],
axis=0))
placeholders[prefixed_state_name] = array_ops.placeholder_with_default(
input=default_state_broadcast,
name=prefixed_state_name,
shape=state_shape_with_batch)
return export_lib.ServingInputReceiver(placeholders, placeholders)
def regressor_parse_example_spec(feature_columns,
label_key,
label_dtype=dtypes.float32,
label_default=None,
label_dimension=1,
weight_column=None):
"""Generates parsing spec for tf.parse_example to be used with regressors.
If users keep data in tf.Example format, they need to call tf.parse_example
with a proper feature spec. There are two main things that this utility helps:
* Users need to combine parsing spec of features with labels and weights
(if any) since they are all parsed from same tf.Example instance. This
utility combines these specs.
* It is difficult to map expected label by a regressor such as `DNNRegressor`
to corresponding tf.parse_example spec. This utility encodes it by getting
related information from users (key, dtype).
Example output of parsing spec:
```python
# Define features and transformations
feature_b = tf.feature_column.numeric_column(...)
feature_c_bucketized = tf.feature_column.bucketized_column(
tf.feature_column.numeric_column("feature_c"), ...)
feature_a_x_feature_c = tf.feature_column.crossed_column(
columns=["feature_a", feature_c_bucketized], ...)
feature_columns = [feature_b, feature_c_bucketized, feature_a_x_feature_c]
parsing_spec = tf.estimator.regressor_parse_example_spec(
feature_columns, label_key='my-label')
# For the above example, regressor_parse_example_spec would return the dict:
assert parsing_spec == {
"feature_a": parsing_ops.VarLenFeature(tf.string),
"feature_b": parsing_ops.FixedLenFeature([1], dtype=tf.float32),
"feature_c": parsing_ops.FixedLenFeature([1], dtype=tf.float32)
"my-label" : parsing_ops.FixedLenFeature([1], dtype=tf.float32)
}
```
Example usage with a regressor:
```python
feature_columns = # define features via tf.feature_column
estimator = DNNRegressor(
hidden_units=[256, 64, 16],
feature_columns=feature_columns,
weight_column='example-weight',
label_dimension=3)
# This label configuration tells the regressor the following:
# * weights are retrieved with key 'example-weight'
# * label is a 3 dimension tensor with float32 dtype.
# Input builders
def input_fn_train(): # Returns a tuple of features and labels.
features = tf.contrib.learn.read_keyed_batch_features(
file_pattern=train_files,
batch_size=batch_size,
# creates parsing configuration for tf.parse_example
features=tf.estimator.classifier_parse_example_spec(
feature_columns,
label_key='my-label',
label_dimension=3,
weight_column='example-weight'),
reader=tf.RecordIOReader)
labels = features.pop('my-label')
return features, labels
estimator.train(input_fn=input_fn_train)
```
Args:
feature_columns: An iterable containing all feature columns. All items
should be instances of classes derived from `_FeatureColumn`.
label_key: A string identifying the label. It means tf.Example stores labels
with this key.
label_dtype: A `tf.dtype` identifies the type of labels. By default it is
`tf.float32`.
label_default: used as label if label_key does not exist in given
tf.Example. By default default_value is none, which means
`tf.parse_example` will error out if there is any missing label.
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.
Returns:
A dict mapping each feature key to a `FixedLenFeature` or `VarLenFeature`
value.
Raises:
ValueError: If label is used in `feature_columns`.
ValueError: If weight_column is used in `feature_columns`.
ValueError: If any of the given `feature_columns` is not a `_FeatureColumn`
instance.
ValueError: If `weight_column` is not a `_NumericColumn` instance.
ValueError: if label_key is None.
"""
parsing_spec = fc.make_parse_example_spec(feature_columns)
if label_key in parsing_spec:
raise ValueError('label should not be used as feature. '
'label_key: {}, features: {}'.format(
label_key, parsing_spec.keys()))
parsing_spec[label_key] = parsing_ops.FixedLenFeature(
(label_dimension,), label_dtype, label_default)
if weight_column is None:
return parsing_spec
if isinstance(weight_column, six.string_types):
weight_column = fc.numeric_column(weight_column)
if not isinstance(weight_column, fc._NumericColumn): # pylint: disable=protected-access
raise ValueError('weight_column should be an instance of '
'tf.feature_column.numeric_column. '
'Given type: {} value: {}'.format(
type(weight_column), weight_column))
if weight_column.key in parsing_spec:
raise ValueError('weight_column should not be used as feature. '
'weight_column: {}, features: {}'.format(
weight_column.key, parsing_spec.keys()))
parsing_spec.update(weight_column._parse_example_spec) # pylint: disable=protected-access
return parsing_spec
def testCreateFeatureSpec(self):
sparse_col = fc.sparse_column_with_hash_bucket(
"sparse_column", hash_bucket_size=100)
embedding_col = fc.embedding_column(
fc.sparse_column_with_hash_bucket(
"sparse_column_for_embedding", hash_bucket_size=10),
dimension=4)
str_sparse_id_col = fc.sparse_column_with_keys(
"str_id_column", ["marlo", "omar", "stringer"])
int32_sparse_id_col = fc.sparse_column_with_keys(
"int32_id_column", [42, 1, -1000], dtype=dtypes.int32)
int64_sparse_id_col = fc.sparse_column_with_keys(
"int64_id_column", [42, 1, -1000], dtype=dtypes.int64)
weighted_id_col = fc.weighted_sparse_column(str_sparse_id_col,
"str_id_weights_column")
real_valued_col1 = fc.real_valued_column("real_valued_column1")
real_valued_col2 = fc.real_valued_column("real_valued_column2", 5)
bucketized_col1 = fc.bucketized_column(
fc.real_valued_column("real_valued_column_for_bucketization1"), [0, 4])
bucketized_col2 = fc.bucketized_column(
fc.real_valued_column("real_valued_column_for_bucketization2", 4),
[0, 4])
a = fc.sparse_column_with_hash_bucket("cross_aaa", hash_bucket_size=100)
b = fc.sparse_column_with_hash_bucket("cross_bbb", hash_bucket_size=100)
cross_col = fc.crossed_column(set([a, b]), hash_bucket_size=10000)
one_hot_col = fc.one_hot_column(fc.sparse_column_with_hash_bucket(
"sparse_column_for_one_hot", hash_bucket_size=100))
scattered_embedding_col = fc.scattered_embedding_column(
"scattered_embedding_column", size=100, dimension=10, hash_key=1)
feature_columns = set([
sparse_col, embedding_col, weighted_id_col, int32_sparse_id_col,
int64_sparse_id_col, real_valued_col1, real_valued_col2,
bucketized_col1, bucketized_col2, cross_col, one_hot_col,
scattered_embedding_col
])
expected_config = {
"sparse_column":
parsing_ops.VarLenFeature(dtypes.string),
"sparse_column_for_embedding":
parsing_ops.VarLenFeature(dtypes.string),
"str_id_column":
parsing_ops.VarLenFeature(dtypes.string),
"int32_id_column":
parsing_ops.VarLenFeature(dtypes.int32),
"int64_id_column":
parsing_ops.VarLenFeature(dtypes.int64),
"str_id_weights_column":
parsing_ops.VarLenFeature(dtypes.float32),
"real_valued_column1":
parsing_ops.FixedLenFeature(
[1], dtype=dtypes.float32),
"real_valued_column2":
parsing_ops.FixedLenFeature(
[5], dtype=dtypes.float32),
"real_valued_column_for_bucketization1":
parsing_ops.FixedLenFeature(
[1], dtype=dtypes.float32),
"real_valued_column_for_bucketization2":
parsing_ops.FixedLenFeature(
[4], dtype=dtypes.float32),
"cross_aaa":
parsing_ops.VarLenFeature(dtypes.string),
"cross_bbb":
parsing_ops.VarLenFeature(dtypes.string),
"sparse_column_for_one_hot":
parsing_ops.VarLenFeature(dtypes.string),
"scattered_embedding_column":
parsing_ops.VarLenFeature(dtypes.string),
}
config = fc.create_feature_spec_for_parsing(feature_columns)
self.assertDictEqual(expected_config, config)
# Tests that contrib feature columns work with core library:
config_core = fc_core.make_parse_example_spec(feature_columns)
self.assertDictEqual(expected_config, config_core)
# Test that the same config is parsed out if we pass a dictionary.
feature_columns_dict = {
str(i): val
for i, val in enumerate(feature_columns)
}
config = fc.create_feature_spec_for_parsing(feature_columns_dict)
self.assertDictEqual(expected_config, config)
def test_complete_flow_with_mode(self, distribution,
use_train_and_evaluate):
label_dimension = 2
input_dimension = label_dimension
batch_size = 10
data = np.linspace(0.,
2.,
batch_size * label_dimension,
dtype=np.float32)
data = data.reshape(batch_size, label_dimension)
train_input_fn = self.dataset_input_fn(
x={'x': data},
y=data,
batch_size=batch_size // len(distribution.worker_devices))
eval_input_fn = self.dataset_input_fn(x={'x': data},
y=data,
batch_size=batch_size //
len(distribution.worker_devices))
predict_input_fn = numpy_io.numpy_input_fn(x={'x': data},
batch_size=batch_size,
shuffle=False)
linear_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension, ))
]
dnn_feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension, ))
]
feature_columns = linear_feature_columns + dnn_feature_columns
session_config = config_pb2.ConfigProto(log_device_placement=True,
allow_soft_placement=True)
estimator = dnn_linear_combined.DNNLinearCombinedRegressor(
linear_feature_columns=linear_feature_columns,
dnn_hidden_units=(2, 2),
dnn_feature_columns=dnn_feature_columns,
label_dimension=label_dimension,
model_dir=self._model_dir,
dnn_optimizer=adam.Adam(0.001),
linear_optimizer=adam.Adam(0.001),
config=run_config.RunConfig(train_distribute=distribution,
eval_distribute=distribution,
session_config=session_config))
num_steps = 2
if use_train_and_evaluate:
scores, _ = training.train_and_evaluate(
estimator,
training.TrainSpec(train_input_fn, max_steps=num_steps),
training.EvalSpec(eval_input_fn))
else:
estimator.train(train_input_fn, steps=num_steps)
scores = estimator.evaluate(eval_input_fn)
self.assertIn('loss', six.iterkeys(scores))
predictions = np.array([
x[prediction_keys.PredictionKeys.PREDICTIONS]
for x in estimator.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, label_dimension), predictions.shape)
feature_spec = feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = estimator.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def classifier_parse_example_spec(feature_columns,
label_key,
label_dtype=dtypes.int64,
label_default=None,
weight_column=None):
"""Generates parsing spec for tf.parse_example to be used with classifiers.
If users keep data in tf.Example format, they need to call tf.parse_example
with a proper feature spec. There are two main things that this utility helps:
* Users need to combine parsing spec of features with labels and weights
(if any) since they are all parsed from same tf.Example instance. This
utility combines these specs.
* It is difficult to map expected label by a classifier such as
`DNNClassifier` to corresponding tf.parse_example spec. This utility encodes
it by getting related information from users (key, dtype).
Example output of parsing spec:
```python
# Define features and transformations
feature_b = tf.feature_column.numeric_column(...)
feature_c_bucketized = tf.feature_column.bucketized_column(
tf.feature_column.numeric_column("feature_c"), ...)
feature_a_x_feature_c = tf.feature_column.crossed_column(
columns=["feature_a", feature_c_bucketized], ...)
feature_columns = [feature_b, feature_c_bucketized, feature_a_x_feature_c]
parsing_spec = tf.estimator.classifier_parse_example_spec(
feature_columns, label_key='my-label', label_dtype=tf.string)
# For the above example, classifier_parse_example_spec would return the dict:
assert parsing_spec == {
"feature_a": parsing_ops.VarLenFeature(tf.string),
"feature_b": parsing_ops.FixedLenFeature([1], dtype=tf.float32),
"feature_c": parsing_ops.FixedLenFeature([1], dtype=tf.float32)
"my-label" : parsing_ops.FixedLenFeature([1], dtype=tf.string)
}
```
Example usage with a classifier:
```python
feature_columns = # define features via tf.feature_column
estimator = DNNClassifier(
n_classes=1000,
feature_columns=feature_columns,
weight_column='example-weight',
label_vocabulary=['photos', 'keep', ...],
hidden_units=[256, 64, 16])
# This label configuration tells the classifier the following:
# * weights are retrieved with key 'example-weight'
# * label is string and can be one of the following ['photos', 'keep', ...]
# * integer id for label 'photos' is 0, 'keep' is 1, ...
# Input builders
def input_fn_train(): # Returns a tuple of features and labels.
features = tf.contrib.learn.read_keyed_batch_features(
file_pattern=train_files,
batch_size=batch_size,
# creates parsing configuration for tf.parse_example
features=tf.estimator.classifier_parse_example_spec(
feature_columns,
label_key='my-label',
label_dtype=tf.string,
weight_column='example-weight'),
reader=tf.RecordIOReader)
labels = features.pop('my-label')
return features, labels
estimator.train(input_fn=input_fn_train)
```
Args:
feature_columns: An iterable containing all feature columns. All items
should be instances of classes derived from `_FeatureColumn`.
label_key: A string identifying the label. It means tf.Example stores labels
with this key.
label_dtype: A `tf.dtype` identifies the type of labels. By default it is
`tf.int64`. If user defines a `label_vocabulary`, this should be set as
`tf.string`. `tf.float32` labels are only supported for binary
classification.
label_default: used as label if label_key does not exist in given
tf.Example. An example usage: let's say `label_key` is 'clicked' and
tf.Example contains clicked data only for positive examples in following
format `key:clicked, value:1`. This means that if there is no data with
key 'clicked' it should count as negative example by setting
`label_deafault=0`. Type of this value should be compatible with
`label_dtype`.
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.
Returns:
A dict mapping each feature key to a `FixedLenFeature` or `VarLenFeature`
value.
Raises:
ValueError: If label is used in `feature_columns`.
ValueError: If weight_column is used in `feature_columns`.
ValueError: If any of the given `feature_columns` is not a `_FeatureColumn`
instance.
ValueError: If `weight_column` is not a `_NumericColumn` instance.
ValueError: if label_key is None.
"""
parsing_spec = fc.make_parse_example_spec(feature_columns)
if label_key in parsing_spec:
raise ValueError('label should not be used as feature. '
'label_key: {}, features: {}'.format(
label_key, parsing_spec.keys()))
parsing_spec[label_key] = parsing_ops.FixedLenFeature((1,), label_dtype,
label_default)
if weight_column is None:
return parsing_spec
if isinstance(weight_column, six.string_types):
weight_column = fc.numeric_column(weight_column)
if not isinstance(weight_column, fc._NumericColumn): # pylint: disable=protected-access
raise ValueError('weight_column should be an instance of '
'tf.feature_column.numeric_column. '
'Given type: {} value: {}'.format(
type(weight_column), weight_column))
if weight_column.key in parsing_spec:
raise ValueError('weight_column should not be used as feature. '
'weight_column: {}, features: {}'.format(
weight_column.key, parsing_spec.keys()))
parsing_spec.update(weight_column._parse_example_spec) # pylint: disable=protected-access
return parsing_spec
def regressor_parse_example_spec(feature_columns,
label_key,
label_dtype=dtypes.float32,
label_default=None,
label_dimension=1,
weight_column=None):
"""Generates parsing spec for tf.parse_example to be used with regressors.
If users keep data in tf.Example format, they need to call tf.parse_example
with a proper feature spec. There are two main things that this utility helps:
* Users need to combine parsing spec of features with labels and weights
(if any) since they are all parsed from same tf.Example instance. This
utility combines these specs.
* It is difficult to map expected label by a regressor such as `DNNRegressor`
to corresponding tf.parse_example spec. This utility encodes it by getting
related information from users (key, dtype).
Example output of parsing spec:
```python
# Define features and transformations
feature_b = tf.feature_column.numeric_column(...)
feature_c_bucketized = tf.feature_column.bucketized_column(
tf.feature_column.numeric_column("feature_c"), ...)
feature_a_x_feature_c = tf.feature_column.crossed_column(
columns=["feature_a", feature_c_bucketized], ...)
feature_columns = [feature_b, feature_c_bucketized, feature_a_x_feature_c]
parsing_spec = tf.estimator.regressor_parse_example_spec(
feature_columns, label_key='my-label')
# For the above example, regressor_parse_example_spec would return the dict:
assert parsing_spec == {
"feature_a": parsing_ops.VarLenFeature(tf.string),
"feature_b": parsing_ops.FixedLenFeature([1], dtype=tf.float32),
"feature_c": parsing_ops.FixedLenFeature([1], dtype=tf.float32)
"my-label" : parsing_ops.FixedLenFeature([1], dtype=tf.float32)
}
```
Example usage with a regressor:
```python
feature_columns = # define features via tf.feature_column
estimator = DNNRegressor(
hidden_units=[256, 64, 16],
feature_columns=feature_columns,
weight_column='example-weight',
label_dimension=3)
# This label configuration tells the regressor the following:
# * weights are retrieved with key 'example-weight'
# * label is a 3 dimension tensor with float32 dtype.
# Input builders
def input_fn_train(): # Returns a tuple of features and labels.
features = tf.contrib.learn.read_keyed_batch_features(
file_pattern=train_files,
batch_size=batch_size,
# creates parsing configuration for tf.parse_example
features=tf.estimator.classifier_parse_example_spec(
feature_columns,
label_key='my-label',
label_dimension=3,
weight_column='example-weight'),
reader=tf.RecordIOReader)
labels = features.pop('my-label')
return features, labels
estimator.train(input_fn=input_fn_train)
```
Args:
feature_columns: An iterable containing all feature columns. All items
should be instances of classes derived from `_FeatureColumn`.
label_key: A string identifying the label. It means tf.Example stores labels
with this key.
label_dtype: A `tf.dtype` identifies the type of labels. By default it is
`tf.float32`.
label_default: used as label if label_key does not exist in given
tf.Example. By default default_value is none, which means
`tf.parse_example` will error out if there is any missing label.
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.
Returns:
A dict mapping each feature key to a `FixedLenFeature` or `VarLenFeature`
value.
Raises:
ValueError: If label is used in `feature_columns`.
ValueError: If weight_column is used in `feature_columns`.
ValueError: If any of the given `feature_columns` is not a `_FeatureColumn`
instance.
ValueError: If `weight_column` is not a `_NumericColumn` instance.
ValueError: if label_key is None.
"""
parsing_spec = fc.make_parse_example_spec(feature_columns)
if label_key in parsing_spec:
raise ValueError('label should not be used as feature. '
'label_key: {}, features: {}'.format(
label_key, parsing_spec.keys()))
parsing_spec[label_key] = parsing_ops.FixedLenFeature(
(label_dimension,), label_dtype, label_default)
if weight_column is None:
return parsing_spec
if isinstance(weight_column, six.string_types):
weight_column = fc.numeric_column(weight_column)
if not isinstance(weight_column, fc._NumericColumn): # pylint: disable=protected-access
raise ValueError('weight_column should be an instance of '
'tf.feature_column.numeric_column. '
'Given type: {} value: {}'.format(
type(weight_column), weight_column))
if weight_column.key in parsing_spec:
raise ValueError('weight_column should not be used as feature. '
'weight_column: {}, features: {}'.format(
weight_column.key, parsing_spec.keys()))
parsing_spec.update(weight_column._parse_example_spec) # pylint: disable=protected-access
return parsing_spec
def classifier_parse_example_spec(feature_columns,
label_key,
label_dtype=dtypes.int64,
label_default=None,
weight_column=None):
"""Generates parsing spec for tf.parse_example to be used with classifiers.
If users keep data in tf.Example format, they need to call tf.parse_example
with a proper feature spec. There are two main things that this utility helps:
* Users need to combine parsing spec of features with labels and weights
(if any) since they are all parsed from same tf.Example instance. This
utility combines these specs.
* It is difficult to map expected label by a classifier such as
`DNNClassifier` to corresponding tf.parse_example spec. This utility encodes
it by getting related information from users (key, dtype).
Example output of parsing spec:
```python
# Define features and transformations
feature_b = tf.feature_column.numeric_column(...)
feature_c_bucketized = tf.feature_column.bucketized_column(
tf.feature_column.numeric_column("feature_c"), ...)
feature_a_x_feature_c = tf.feature_column.crossed_column(
columns=["feature_a", feature_c_bucketized], ...)
feature_columns = [feature_b, feature_c_bucketized, feature_a_x_feature_c]
parsing_spec = tf.estimator.classifier_parse_example_spec(
feature_columns, label_key='my-label', label_dtype=tf.string)
# For the above example, classifier_parse_example_spec would return the dict:
assert parsing_spec == {
"feature_a": parsing_ops.VarLenFeature(tf.string),
"feature_b": parsing_ops.FixedLenFeature([1], dtype=tf.float32),
"feature_c": parsing_ops.FixedLenFeature([1], dtype=tf.float32)
"my-label" : parsing_ops.FixedLenFeature([1], dtype=tf.string)
}
```
Example usage with a classifier:
```python
feature_columns = # define features via tf.feature_column
estimator = DNNClassifier(
n_classes=1000,
feature_columns=feature_columns,
weight_column='example-weight',
label_vocabulary=['photos', 'keep', ...],
hidden_units=[256, 64, 16])
# This label configuration tells the classifier the following:
# * weights are retrieved with key 'example-weight'
# * label is string and can be one of the following ['photos', 'keep', ...]
# * integer id for label 'photos' is 0, 'keep' is 1, ...
# Input builders
def input_fn_train(): # Returns a tuple of features and labels.
features = tf.contrib.learn.read_keyed_batch_features(
file_pattern=train_files,
batch_size=batch_size,
# creates parsing configuration for tf.parse_example
features=tf.estimator.classifier_parse_example_spec(
feature_columns,
label_key='my-label',
label_dtype=tf.string,
weight_column='example-weight'),
reader=tf.RecordIOReader)
labels = features.pop('my-label')
return features, labels
estimator.train(input_fn=input_fn_train)
```
Args:
feature_columns: An iterable containing all feature columns. All items
should be instances of classes derived from `_FeatureColumn`.
label_key: A string identifying the label. It means tf.Example stores labels
with this key.
label_dtype: A `tf.dtype` identifies the type of labels. By default it is
`tf.int64`. If user defines a `label_vocabulary`, this should be set as
`tf.string`. `tf.float32` labels are only supported for binary
classification.
label_default: used as label if label_key does not exist in given
tf.Example. An example usage: let's say `label_key` is 'clicked' and
tf.Example contains clicked data only for positive examples in following
format `key:clicked, value:1`. This means that if there is no data with
key 'clicked' it should count as negative example by setting
`label_deafault=0`. Type of this value should be compatible with
`label_dtype`.
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.
Returns:
A dict mapping each feature key to a `FixedLenFeature` or `VarLenFeature`
value.
Raises:
ValueError: If label is used in `feature_columns`.
ValueError: If weight_column is used in `feature_columns`.
ValueError: If any of the given `feature_columns` is not a `_FeatureColumn`
instance.
ValueError: If `weight_column` is not a `_NumericColumn` instance.
ValueError: if label_key is None.
"""
parsing_spec = fc.make_parse_example_spec(feature_columns)
if label_key in parsing_spec:
raise ValueError('label should not be used as feature. '
'label_key: {}, features: {}'.format(
label_key, parsing_spec.keys()))
parsing_spec[label_key] = parsing_ops.FixedLenFeature((1,), label_dtype,
label_default)
if weight_column is None:
return parsing_spec
if isinstance(weight_column, six.string_types):
weight_column = fc.numeric_column(weight_column)
if not isinstance(weight_column, fc._NumericColumn): # pylint: disable=protected-access
raise ValueError('weight_column should be an instance of '
'tf.feature_column.numeric_column. '
'Given type: {} value: {}'.format(
type(weight_column), weight_column))
if weight_column.key in parsing_spec:
raise ValueError('weight_column should not be used as feature. '
'weight_column: {}, features: {}'.format(
weight_column.key, parsing_spec.keys()))
parsing_spec.update(weight_column._parse_example_spec) # pylint: disable=protected-access
return parsing_spec
def _get_exporter(self, name, fc):
feature_spec = feature_column.make_parse_example_spec(fc)
serving_input_receiver_fn = (
export_lib.build_parsing_serving_input_receiver_fn(feature_spec))
return exporter_lib.LatestExporter(
name, serving_input_receiver_fn=serving_input_receiver_fn)
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
placeholders = {}
time_placeholder = array_ops.placeholder(
name=feature_keys.TrainEvalFeatures.TIMES,
dtype=dtypes.int64,
shape=[default_batch_size, default_series_length])
placeholders[feature_keys.TrainEvalFeatures.TIMES] = time_placeholder
# Values are only necessary when filtering. For prediction the default
# value will be ignored.
placeholders[feature_keys.TrainEvalFeatures.VALUES] = (
array_ops.placeholder_with_default(
name=feature_keys.TrainEvalFeatures.VALUES,
input=array_ops.zeros(
shape=[
default_batch_size
if default_batch_size else 0, default_series_length
if default_series_length else 0, self._model.num_features
],
dtype=self._model.dtype),
shape=(default_batch_size, default_series_length,
self._model.num_features)))
if self._model.exogenous_feature_columns:
with ops.Graph().as_default():
# Default placeholders have only an unknown batch dimension. Make them
# in a separate graph, then splice in the series length to the shapes
# and re-create them in the outer graph.
parsed_features = (
feature_column.make_parse_example_spec(
self._model.exogenous_feature_columns))
placeholder_features = parsing_ops.parse_example(
serialized=array_ops.placeholder(
shape=[None], dtype=dtypes.string),
features=parsed_features)
exogenous_feature_shapes = {
key: (value.get_shape(), value.dtype) for key, value
in placeholder_features.items()}
for feature_key, (batch_only_feature_shape, value_dtype) in (
exogenous_feature_shapes.items()):
batch_only_feature_shape = (
batch_only_feature_shape.with_rank_at_least(1).as_list())
feature_shape = ([default_batch_size, default_series_length]
+ batch_only_feature_shape[1:])
placeholders[feature_key] = array_ops.placeholder(
dtype=value_dtype, name=feature_key, shape=feature_shape)
# Models may not know the shape of their state without creating some
# variables/ops. Avoid polluting the default graph by making a new one. We
# use only static metadata from the returned Tensors.
with ops.Graph().as_default():
self._model.initialize_graph()
# Evaluate the initial state as same-dtype "zero" values. These zero
# constants aren't used, but are necessary for feeding to
# placeholder_with_default for the "cold start" case where state is not
# fed to the model.
def _zeros_like_constant(tensor):
return tensor_util.constant_value(array_ops.zeros_like(tensor))
start_state = nest.map_structure(
_zeros_like_constant, self._model.get_start_state())
batch_size_tensor = array_ops.shape(time_placeholder)[0]
for prefixed_state_name, state in ts_head_lib.state_to_dictionary(
start_state).items():
state_shape_with_batch = tensor_shape.TensorShape(
(default_batch_size,)).concatenate(state.shape)
default_state_broadcast = array_ops.tile(
state[None, ...],
multiples=array_ops.concat(
[batch_size_tensor[None],
array_ops.ones(len(state.shape), dtype=dtypes.int32)],
axis=0))
placeholders[prefixed_state_name] = array_ops.placeholder_with_default(
input=default_state_broadcast,
name=prefixed_state_name,
shape=state_shape_with_batch)
return export_lib.ServingInputReceiver(placeholders, placeholders)
def test_complete_flow(self):
n_classes = 3
input_dimension = 2
batch_size = 12
data = np.linspace(0.,
n_classes - 1.,
batch_size * input_dimension,
dtype=np.float32)
x_data = data.reshape(batch_size, input_dimension)
y_data = np.reshape(self._as_label(data[:batch_size]), (batch_size, 1))
train_input_fn = numpy_io.numpy_input_fn(x={'x': x_data},
y=y_data,
batch_size=batch_size,
num_epochs=None,
shuffle=True)
eval_input_fn = numpy_io.numpy_input_fn(x={'x': x_data},
y=y_data,
batch_size=batch_size,
shuffle=False)
predict_input_fn = numpy_io.numpy_input_fn(x={'x': x_data},
batch_size=batch_size,
shuffle=False)
feature_columns = [
feature_column.numeric_column('x', shape=(input_dimension, ))
]
estimator = dnn.DNNClassifier(hidden_units=(2, 2),
feature_columns=feature_columns,
n_classes=n_classes,
model_dir=self._model_dir)
def optimizer_fn():
return optimizers.get_optimizer_instance('Adagrad',
learning_rate=0.05)
# TODO(isaprykin): Switch Estimator to use allow_soft_placement=True
# during export_savedmodel and then switch this test to replicate over
# GPUs instead of CPUs.
estimator = estimator_lib.Estimator(
model_fn=replicate_model_fn.replicate_model_fn(
estimator.model_fn,
optimizer_fn,
devices=['/cpu:0', '/cpu:0', '/cpu:0']),
model_dir=estimator.model_dir,
config=estimator.config,
params=estimator.params)
num_steps = 10
estimator.train(train_input_fn, steps=num_steps)
scores = estimator.evaluate(eval_input_fn)
self.assertEqual(num_steps, scores[ops_lib.GraphKeys.GLOBAL_STEP])
self.assertIn('loss', six.iterkeys(scores))
predicted_proba = np.array([
x[prediction_keys.PredictionKeys.PROBABILITIES]
for x in estimator.predict(predict_input_fn)
])
self.assertAllEqual((batch_size, n_classes), predicted_proba.shape)
feature_spec = feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = export.build_parsing_serving_input_receiver_fn(
feature_spec)
export_dir = estimator.export_savedmodel(tempfile.mkdtemp(),
serving_input_receiver_fn)
self.assertTrue(gfile.Exists(export_dir))
def _serving_input_receiver_fn():
"""A receiver function to be passed to export_savedmodel."""
placeholders = {}
placeholders[feature_keys.TrainEvalFeatures.TIMES] = (
array_ops.placeholder(
name=feature_keys.TrainEvalFeatures.TIMES,
dtype=dtypes.int64,
shape=[default_batch_size, default_series_length]))
# Values are only necessary when filtering. For prediction the default
# value will be ignored.
placeholders[feature_keys.TrainEvalFeatures.VALUES] = (
array_ops.placeholder_with_default(
name=feature_keys.TrainEvalFeatures.VALUES,
input=array_ops.zeros(shape=[
default_batch_size if default_batch_size else 0,
default_series_length if default_series_length else 0,
self._model.num_features
],
dtype=self._model.dtype),
shape=(default_batch_size, default_series_length,
self._model.num_features)))
if self._model.exogenous_feature_columns:
with ops.Graph().as_default():
# Default placeholders have only an unknown batch dimension. Make them
# in a separate graph, then splice in the series length to the shapes
# and re-create them in the outer graph.
parsed_features = (feature_column.make_parse_example_spec(
self._model.exogenous_feature_columns))
placeholder_features = parsing_ops.parse_example(
serialized=array_ops.placeholder(shape=[None],
dtype=dtypes.string),
features=parsed_features)
exogenous_feature_shapes = {
key: (value.get_shape(), value.dtype)
for key, value in placeholder_features.items()
}
for feature_key, (batch_only_feature_shape, value_dtype) in (
exogenous_feature_shapes.items()):
batch_only_feature_shape = (
batch_only_feature_shape.with_rank_at_least(
1).as_list())
feature_shape = (
[default_batch_size, default_series_length] +
batch_only_feature_shape[1:])
placeholders[feature_key] = array_ops.placeholder(
dtype=value_dtype,
name=feature_key,
shape=feature_shape)
# Models may not know the shape of their state without creating some
# variables/ops. Avoid polluting the default graph by making a new one. We
# use only static metadata from the returned Tensors.
with ops.Graph().as_default():
self._model.initialize_graph()
model_start_state = self._model.get_start_state()
for prefixed_state_name, state_tensor in ts_head_lib.state_to_dictionary(
model_start_state).items():
state_shape_with_batch = tensor_shape.TensorShape(
(default_batch_size, )).concatenate(
state_tensor.get_shape())
placeholders[prefixed_state_name] = array_ops.placeholder(
name=prefixed_state_name,
shape=state_shape_with_batch,
dtype=state_tensor.dtype)
return export_lib.ServingInputReceiver(placeholders, placeholders)
def testCreateFeatureSpec(self):
sparse_col = fc.sparse_column_with_hash_bucket(
"sparse_column", hash_bucket_size=100)
embedding_col = fc.embedding_column(
fc.sparse_column_with_hash_bucket(
"sparse_column_for_embedding", hash_bucket_size=10),
dimension=4)
str_sparse_id_col = fc.sparse_column_with_keys(
"str_id_column", ["marlo", "omar", "stringer"])
int32_sparse_id_col = fc.sparse_column_with_keys(
"int32_id_column", [42, 1, -1000], dtype=dtypes.int32)
int64_sparse_id_col = fc.sparse_column_with_keys(
"int64_id_column", [42, 1, -1000], dtype=dtypes.int64)
weighted_id_col = fc.weighted_sparse_column(str_sparse_id_col,
"str_id_weights_column")
real_valued_col1 = fc.real_valued_column("real_valued_column1")
real_valued_col2 = fc.real_valued_column("real_valued_column2", 5)
bucketized_col1 = fc.bucketized_column(
fc.real_valued_column("real_valued_column_for_bucketization1"), [0, 4])
bucketized_col2 = fc.bucketized_column(
fc.real_valued_column("real_valued_column_for_bucketization2", 4),
[0, 4])
a = fc.sparse_column_with_hash_bucket("cross_aaa", hash_bucket_size=100)
b = fc.sparse_column_with_hash_bucket("cross_bbb", hash_bucket_size=100)
cross_col = fc.crossed_column(set([a, b]), hash_bucket_size=10000)
one_hot_col = fc.one_hot_column(fc.sparse_column_with_hash_bucket(
"sparse_column_for_one_hot", hash_bucket_size=100))
scattered_embedding_col = fc.scattered_embedding_column(
"scattered_embedding_column", size=100, dimension=10, hash_key=1)
feature_columns = set([
sparse_col, embedding_col, weighted_id_col, int32_sparse_id_col,
int64_sparse_id_col, real_valued_col1, real_valued_col2,
bucketized_col1, bucketized_col2, cross_col, one_hot_col,
scattered_embedding_col
])
expected_config = {
"sparse_column":
parsing_ops.VarLenFeature(dtypes.string),
"sparse_column_for_embedding":
parsing_ops.VarLenFeature(dtypes.string),
"str_id_column":
parsing_ops.VarLenFeature(dtypes.string),
"int32_id_column":
parsing_ops.VarLenFeature(dtypes.int32),
"int64_id_column":
parsing_ops.VarLenFeature(dtypes.int64),
"str_id_weights_column":
parsing_ops.VarLenFeature(dtypes.float32),
"real_valued_column1":
parsing_ops.FixedLenFeature(
[1], dtype=dtypes.float32),
"real_valued_column2":
parsing_ops.FixedLenFeature(
[5], dtype=dtypes.float32),
"real_valued_column_for_bucketization1":
parsing_ops.FixedLenFeature(
[1], dtype=dtypes.float32),
"real_valued_column_for_bucketization2":
parsing_ops.FixedLenFeature(
[4], dtype=dtypes.float32),
"cross_aaa":
parsing_ops.VarLenFeature(dtypes.string),
"cross_bbb":
parsing_ops.VarLenFeature(dtypes.string),
"sparse_column_for_one_hot":
parsing_ops.VarLenFeature(dtypes.string),
"scattered_embedding_column":
parsing_ops.VarLenFeature(dtypes.string),
}
config = fc.create_feature_spec_for_parsing(feature_columns)
self.assertDictEqual(expected_config, config)
# Tests that contrib feature columns work with core library:
config_core = fc_core.make_parse_example_spec(feature_columns)
self.assertDictEqual(expected_config, config_core)
# Test that the same config is parsed out if we pass a dictionary.
feature_columns_dict = {
str(i): val
for i, val in enumerate(feature_columns)
}
config = fc.create_feature_spec_for_parsing(feature_columns_dict)
self.assertDictEqual(expected_config, config)
