def test_encode_listwise_features(self):
with tf.Graph().as_default():
# Batch size = 2, list_size = 2.
features = {
"query_length":
tf.convert_to_tensor(value=[[1], [2]]),
"utility":
tf.convert_to_tensor(value=[[[1.0], [0.0]], [[0.0], [1.0]]]),
"unigrams":
tf.SparseTensor(
indices=[[0, 0, 0], [0, 1, 0], [1, 0, 0], [1, 1, 0]],
values=["ranking", "regression", "classification", "ordinal"],
dense_shape=[2, 2, 1])
}
context_feature_columns = {
"query_length":
feature_column.numeric_column(
"query_length", shape=(1,), default_value=0, dtype=tf.int64)
}
example_feature_columns = {
"utility":
feature_column.numeric_column(
"utility", shape=(1,), default_value=0.0, dtype=tf.float32),
"unigrams":
feature_column.embedding_column(
feature_column.categorical_column_with_vocabulary_list(
"unigrams",
vocabulary_list=[
"ranking", "regression", "classification", "ordinal"
]),
dimension=10)
}
with self.assertRaisesRegexp(
ValueError,
r"2nd dimension of tensor must be equal to input size: 3, but found .*"
):
feature_lib.encode_listwise_features(
features,
input_size=3,
context_feature_columns=context_feature_columns,
example_feature_columns=example_feature_columns)
context_features, example_features = feature_lib.encode_listwise_features(
features,
input_size=2,
context_feature_columns=context_feature_columns,
example_feature_columns=example_feature_columns)
self.assertAllEqual(["query_length"], sorted(context_features))
self.assertAllEqual(["unigrams", "utility"], sorted(example_features))
self.assertAllEqual([2, 2, 10],
example_features["unigrams"].get_shape().as_list())
with tf.compat.v1.Session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
sess.run(tf.compat.v1.tables_initializer())
context_features, example_features = sess.run(
[context_features, example_features])
self.assertAllEqual([[1], [2]], context_features["query_length"])
self.assertAllEqual([[[1.0], [0.0]], [[0.0], [1.0]]],
example_features["utility"])
def _multiply_by_two_transform_fn(features, mode):
for feature, tensor in six.iteritems(features):
features[feature] = 2 * tensor
context, example = feature_lib.encode_listwise_features(
features=features,
context_feature_columns=_context_feature_columns(),
example_feature_columns=_example_feature_columns(),
mode=mode)
return context, example
def _transform_fn(features, mode):
"""Splits the features into context and per-example features."""
for k in features:
print(features[k].shape)
context_features, example_features = feature_lib.encode_listwise_features(
features,
input_size=input_size,
context_feature_columns=context_feature_columns,
example_feature_columns=example_feature_columns,
mode=mode)
for k in example_features:
print(k)
print(example_features[k].shape)
for k in context_features:
print(k)
print(context_features[k].shape)
return context_features, example_features
def _transform_fn(self, features, mode):
"""Defines the transform fn."""
if self._transform_function is not None:
return self._transform_function(features=features, mode=mode)
if mode == tf.estimator.ModeKeys.PREDICT:
return feature.encode_pointwise_features(
features=features,
context_feature_columns=self._context_feature_columns,
example_feature_columns=self._example_feature_columns,
mode=mode,
scope="transform_layer")
else:
return feature.encode_listwise_features(
features=features,
context_feature_columns=self._context_feature_columns,
example_feature_columns=self._example_feature_columns,
mode=mode,
scope="transform_layer")
