def testInputLayer(self):
with tf.Graph().as_default():
text_a = tf.SparseTensor(
values=["hello world", "pair-programming", "hello world"],
indices=[[0, 0], [0, 1], [1, 0]],
dense_shape=[2, 2])
text_b = tf.SparseTensor(
values=["hello world", "oov token"],
indices=[[0, 0], [0, 1]],
dense_shape=[2, 3])
features = {
"text_a": text_a,
"text_b": text_b,
}
feature_columns = [
hub.sparse_text_embedding_column(
"text_a",
self.spec,
combiner="mean",
default_value="__UNKNOWN__",
trainable=False),
hub.sparse_text_embedding_column(
"text_b",
self.spec,
combiner="mean",
default_value="__UNKNOWN__",
trainable=False),
]
input_layer = tf_v1.feature_column.input_layer(features, feature_columns)
with tf_v1.train.MonitoredSession() as sess:
output = sess.run(input_layer)
self.assertAllEqual(
output,
[[3, 3.5, 4, 4.5, 0.5, 1, 1.5, 2], [1, 2, 3, 4, 0, 0, 0, 0]])
def testTrainableEmbeddingColumn(self):
feature_columns = [
hub.sparse_text_embedding_column(
"text",
self.spec,
combiner="mean",
default_value=None,
trainable=True),
]
with tf.Graph().as_default():
text = tf.SparseTensor(
values=["hello world", "pair-programming"],
indices=[[0, 0], [1, 0]],
dense_shape=[2, 2])
target = [[1, 1, 1, 1], [4, 3, 2, 1]]
input_layer = tf_v1.feature_column.input_layer({"text": text},
feature_columns)
loss = tf_v1.losses.mean_squared_error(input_layer, target)
optimizer = tf_v1.train.GradientDescentOptimizer(learning_rate=0.97)
train_op = optimizer.minimize(loss)
with tf_v1.train.MonitoredSession() as sess:
self.assertAllEqual(sess.run(input_layer), [[1, 2, 3, 4], [5, 5, 5, 5]])
for _ in range(10):
sess.run(train_op)
self.assertAllClose(sess.run(input_layer), target, atol=0.5)
def testEmptySparseTensorBatch(self):
feature_columns = [
hub.sparse_text_embedding_column("text",
self.spec,
combiner="mean",
default_value="default",
trainable=True),
]
with tf.Graph().as_default():
text = tf.SparseTensor(values=tf_v1.constant([],
dtype=tf_v1.string,
shape=[0]),
indices=tf_v1.constant([],
dtype=tf_v1.int64,
shape=[0, 2]),
dense_shape=[3, 0])
input_layer = tf_v1.feature_column.input_layer({"text": text},
feature_columns)
with tf_v1.train.MonitoredSession() as sess:
embeddings = sess.run(input_layer)
self.assertAllEqual(embeddings,
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]])
def testParents(self):
text_column = hub.sparse_text_embedding_column("text",
self.spec,
"sum",
"",
trainable=False)
self.assertEqual(["text"], text_column.parents)
def testVariableShape(self):
text_column = hub.sparse_text_embedding_column("text",
self.spec,
combiner="mean",
default_value=None,
trainable=False)
self.assertEqual(text_column._variable_shape, [4])
def testMakeParseExampleSpec(self):
text_column = hub.sparse_text_embedding_column(
"text", self.spec, combiner="mean", default_value=None, trainable=False)
parsing_spec = tf.compat.v1.feature_column.make_parse_example_spec(
[text_column])
self.assertEqual(
parsing_spec, {"text": tf.compat.v1.VarLenFeature(tf.string)})