def get_input_reciever_fn(self):
feature = tf.placeholder(dtype=tf.int32, shape=[None, None, None], name="feature_tensor")
pos_name = tf.placeholder(dtype=tf.int32, shape=[None, None, None], name = "pos_tensor")
receiver_tensors = {self.feature_name: feature, self.pos_name: pos_name}
return build_raw_serving_input_receiver_fn(receiver_tensors)
def serving_input_fn(params):
user = tf.placeholder(tf.int64, shape=[1])
item = tf.placeholder(tf.int64, shape=[1])
return build_raw_serving_input_receiver_fn({
USER_EMBEDDING_TENSOR_NAME: user,
ITEM_EMBEDDING_TENSOR_NAME: item
})()
def dummy_serving_receiver_fn():
feature_spec = {
'x':
array_ops.placeholder(dtype=dtypes.int64,
shape=(2, 1),
name='feature_x'),
}
return export.build_raw_serving_input_receiver_fn(feature_spec)
def test_build_raw_serving_input_receiver_fn_name(self):
"""Test case for issue #12755."""
f = {
"feature":
array_ops.placeholder(
name="feature", shape=[32], dtype=dtypes.float32)
}
serving_input_receiver_fn = export.build_raw_serving_input_receiver_fn(f)
v = serving_input_receiver_fn()
self.assertTrue(isinstance(v, export.ServingInputReceiver))
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
num_training_data = maybe_download_iris_data(IRIS_TRAINING,
IRIS_TRAINING_URL)
num_test_data = maybe_download_iris_data(IRIS_TEST, IRIS_TEST_URL)
# Build 3 layer DNN with 10, 20, 10 units respectively.
feature_columns = [
tf.feature_column.numeric_column(key, shape=1) for key in FEATURE_KEYS
]
classifier = tf.estimator.DNNClassifier(feature_columns=feature_columns,
hidden_units=[10, 20, 10],
n_classes=3)
# Train.
train_input_fn = input_fn(IRIS_TRAINING,
num_training_data,
batch_size=32,
is_training=True)
classifier.train(input_fn=train_input_fn, steps=400)
# Eval.
test_input_fn = input_fn(IRIS_TEST,
num_test_data,
batch_size=32,
is_training=False)
scores = classifier.evaluate(input_fn=test_input_fn)
print('Accuracy (tensorflow): {0:f}'.format(scores['accuracy']))
# Export the SavedModel file
import shutil
import tempfile
#savemodel_dir = classifier.export_savedmodel(tempfile.mkdtemp(), serving_input_fn = serving_input_fn, as_text = True)
from tensorflow.python.estimator.export import export
#feature_spec = {'MY_FEATURE': tf.constant(2.0, shape=[1, 1])}
# FEATURE_KEYS = ['sepal_length', 'sepal_width', 'petal_length', 'petal_width']
feature_spec = {
'sepal_length': tf.constant(2.0, shape=[1, 1]),
'sepal_width': tf.constant(2.0, shape=[1, 1]),
'petal_length': tf.constant(2.0, shape=[1, 1]),
'petal_width': tf.constant(2.0, shape=[1, 1])
}
serving_input_fn = export.build_raw_serving_input_receiver_fn(feature_spec)
savemodel_dir = classifier.export_savedmodel(tempfile.mkdtemp(),
serving_input_fn,
as_text=True)
savemodel_dir = savemodel_dir.decode("UTF-8")
name = "1"
if (os.path.isdir("savedmodel/" + name)):
shutil.rmtree("savedmodel/" + name)
shutil.move(savemodel_dir, "savedmodel/" + name)
def serving_input_fn(hyperparameters):
feature_spec = {
INPUT_TENSOR_PRICE: tf.placeholder(tf.float32, shape=[1]),
INPUT_TENSOR_INVENTORY: tf.placeholder(tf.float32, shape=[1])
}
# These should have () after them? I don't know. Most example for input_fn show it, but my simple example
# doesn't work with it.
# Used in RAW for parse_input
return build_raw_serving_input_receiver_fn(feature_spec)
def get_input_reciever_fn(self):
context = tf.placeholder(dtype=tf.string,
shape=[None, None, None],
name="context_tensor")
question = tf.placeholder(dtype=tf.string,
shape=[None, None],
name="query_tensor")
receiver_tensors = {
self.context_name: context,
self.question_name: question
}
return build_raw_serving_input_receiver_fn(receiver_tensors)
def test_build_raw_serving_input_receiver_fn_without_shape(self):
"""Test case for issue #21178."""
f = {"feature_1": array_ops.placeholder(dtypes.float32),
"feature_2": array_ops.placeholder(dtypes.int32)}
serving_input_receiver_fn = export.build_raw_serving_input_receiver_fn(f)
v = serving_input_receiver_fn()
self.assertTrue(isinstance(v, export.ServingInputReceiver))
self.assertEqual(
tensor_shape.unknown_shape(),
v.receiver_tensors["feature_1"].shape)
self.assertEqual(
tensor_shape.unknown_shape(),
v.receiver_tensors["feature_2"].shape)
def test_build_raw_serving_input_receiver_fn_without_shape(self):
"""Test case for issue #21178."""
f = {"feature_1": array_ops.placeholder(dtypes.float32),
"feature_2": array_ops.placeholder(dtypes.int32)}
serving_input_receiver_fn = export.build_raw_serving_input_receiver_fn(f)
v = serving_input_receiver_fn()
self.assertTrue(isinstance(v, export.ServingInputReceiver))
self.assertEqual(
tensor_shape.unknown_shape(),
v.receiver_tensors["feature_1"].shape)
self.assertEqual(
tensor_shape.unknown_shape(),
v.receiver_tensors["feature_2"].shape)
def main(_):
if len(sys.argv) < 1 or sys.argv[-1].startswith('-'):
print('Usage: keras_vgg.py export_dir')
sys.exit(-1)
export_path_base = sys.argv[-1]
model = keras.applications.vgg16.VGG16(weights='imagenet')
model.compile(optimizer=keras.optimizers.SGD(lr=.01, momentum=.9),
loss='binary_crossentropy',
metrics=['accuracy'])
print(model.summary())
estimator = tf.keras.estimator.model_to_estimator(keras_model=model)
feature_spec = {'input_1': model.input}
serving_input_fn = export.build_raw_serving_input_receiver_fn(feature_spec)
estimator.export_savedmodel(export_path_base, serving_input_fn)
def test_build_raw_serving_input_receiver_fn(self):
features = {"feature_1": constant_op.constant(["hello"]),
"feature_2": constant_op.constant([42])}
serving_input_receiver_fn = export.build_raw_serving_input_receiver_fn(
features)
with ops.Graph().as_default():
serving_input_receiver = serving_input_receiver_fn()
self.assertEqual(set(["feature_1", "feature_2"]),
set(serving_input_receiver.features.keys()))
self.assertEqual(set(["feature_1", "feature_2"]),
set(serving_input_receiver.receiver_tensors.keys()))
self.assertEqual(
dtypes.string,
serving_input_receiver.receiver_tensors["feature_1"].dtype)
self.assertEqual(
dtypes.int32,
serving_input_receiver.receiver_tensors["feature_2"].dtype)
def export(self):
assert self.config.checkpoint_path is not None
model_dir = str(self.config.checkpoint_path)
def model_fn(features, labels, mode):
sentence = features['sentence']
model = Model(sentence, labels, self.params, mode)
return tf.estimator.EstimatorSpec(mode,
{'label': model.prediction})
estimator = tf.estimator.Estimator(model_fn, model_dir)
sentence = tf.placeholder(tf.string, [None], 'sentence')
serving_input_receiver_fn = build_raw_serving_input_receiver_fn(
{'sentence': sentence})
estimator.export_saved_model(
str(self.config.checkpoint_path / 'exported'),
serving_input_receiver_fn)
log.info("Export complete")
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
num_training_data = maybe_download_iris_data(
IRIS_TRAINING, IRIS_TRAINING_URL)
num_test_data = maybe_download_iris_data(IRIS_TEST, IRIS_TEST_URL)
# Build 3 layer DNN with 10, 20, 10 units respectively.
feature_columns = [
tf.feature_column.numeric_column(key, shape=1) for key in FEATURE_KEYS]
classifier = tf.estimator.DNNClassifier(
feature_columns=feature_columns, hidden_units=[10, 20, 10], n_classes=3)
# Train.
train_input_fn = input_fn(IRIS_TRAINING, num_training_data, batch_size=32,
is_training=True)
classifier.train(input_fn=train_input_fn, steps=400)
# Eval.
test_input_fn = input_fn(IRIS_TEST, num_test_data, batch_size=32,
is_training=False)
scores = classifier.evaluate(input_fn=test_input_fn)
print('Accuracy (tensorflow): {0:f}'.format(scores['accuracy']))
# Export the SavedModel file
import shutil
import tempfile
#savemodel_dir = classifier.export_savedmodel(tempfile.mkdtemp(), serving_input_fn = serving_input_fn, as_text = True)
from tensorflow.python.estimator.export import export
#feature_spec = {'MY_FEATURE': tf.constant(2.0, shape=[1, 1])}
# FEATURE_KEYS = ['sepal_length', 'sepal_width', 'petal_length', 'petal_width']
feature_spec = {'sepal_length': tf.constant(2.0, shape=[1, 1]), 'sepal_width': tf.constant(2.0, shape=[1, 1]), 'petal_length': tf.constant(2.0, shape=[1, 1]), 'petal_width': tf.constant(2.0, shape=[1, 1])}
serving_input_fn = export.build_raw_serving_input_receiver_fn(feature_spec)
savemodel_dir = classifier.export_savedmodel(tempfile.mkdtemp(), serving_input_fn, as_text = True)
savemodel_dir = savemodel_dir.decode("UTF-8")
name = "1"
if(os.path.isdir("savedmodel/" + name)):
shutil.rmtree("savedmodel/" + name)
shutil.move(savemodel_dir, "savedmodel/" + name)
fc1 = tf.layers.dense(fc1, 1024)
# Apply Dropout (if is_training is False, dropout is not applied)
fc1 = tf.layers.dropout(fc1, rate=dropout, training=is_training)
# Output layer, class prediction
out = tf.layers.dense(fc1, n_classes)
return out
from tensorflow.python.estimator.export import export
with tf.Session() as sess:
# Build the Estimator
feature_spec = {'images': tf.constant(mnist.train.images)}
serving_input_fn = export.build_raw_serving_input_receiver_fn(feature_spec)
# Train the Model
# Evaluate the Model
# Define the input function for evaluating
input_fn = tf.estimator.inputs.numpy_input_fn(
x={'images': mnist.test.images},
y=mnist.test.labels,
batch_size=batch_size,
shuffle=False)
# Define a scope for reusing the variables
# TF Estimator input is a dict, in case of multiple inputs
x = mnist.test.images
is_training = False
n_classes = 10
def serving_input_fn(params):
tensor = tf.placeholder(tf.float32, shape=[1, 7])
return build_raw_serving_input_receiver_fn({INPUT_TENSOR_NAME: tensor})()
def serving_input_fn(params):
inputs = tf.placeholder(tf.int32, shape=[None, 7])
tensors = {'inputs': inputs}
return build_raw_serving_input_receiver_fn(tensors)()
def dummy_serving_receiver_fn():
feature_spec = {'x': array_ops.placeholder(
dtype=dtypes.int64, shape=(2, 1), name='feature_x'),}
return export.build_raw_serving_input_receiver_fn(feature_spec)
def serving_input_fn(params):
inputs = tf.placeholder(tf.int32, shape=[None, 7])
tensors = {'inputs': inputs}
return build_raw_serving_input_receiver_fn(tensors)()
def serving_input_fn(params):
tensor = tf.placeholder(tf.float32, shape=[1, 7])
return build_raw_serving_input_receiver_fn({INPUT_TENSOR_NAME: tensor})()
def serving_input_fn(params):
inputs = tf.convert_to_tensor(X_test)
return build_raw_serving_input_receiver_fn({INPUT_TENSOR_NAME: inputs})()
