def test_padded_where_dynamic_shape(self):
"""Tests that padded_where can deal with dynamic shapes."""
with tf.Graph().as_default():
condition = tf.placeholder(tf.bool, [None, None])
indices, mask = tensor_utils.padded_where(condition, 3)
with tf.Session() as sess:
# Test for single batch with 8 elements.
results = sess.run(
[indices, mask],
feed_dict={condition: [[True] * 4 + [False] * 4]})
self.assertAllEqual(results[0], [[0, 1, 2]])
self.assertAllEqual(results[1], [[1, 1, 1]])
# Test for three batches with 6 elements each.
results = sess.run(
[indices, mask],
feed_dict={
condition: [[False, True, False, True, True, True],
[True, False, True, False, False, True],
[True, True, False, False, True, False]]
})
self.assertAllEqual(results[0],
[[1, 3, 4], [0, 2, 5], [0, 1, 4]])
self.assertAllEqual(results[1],
[[1, 1, 1], [1, 1, 1], [1, 1, 1]])
def test_padded_where_with_padding(self):
"""Basic padding test for padded_where."""
# Case where True condition values are in the left (beginning) of tensor.
condition = tf.constant([True] * 4 + [False] * 4)
indices, mask = tensor_utils.padded_where(condition, 6)
self.assertAllEqual(indices, [0, 1, 2, 3, 0, 0])
self.assertAllEqual(mask, [1, 1, 1, 1, 0, 0])
# Case where True condition values are in the righ (end) of tensor.
condition = tf.constant([False] * 4 + [True] * 4)
indices, mask = tensor_utils.padded_where(condition, 6)
self.assertAllEqual(indices, [4, 5, 6, 7, 0, 0])
self.assertAllEqual(mask, [1, 1, 1, 1, 0, 0])
# Case where True condition values non-contiguous.
condition = tf.constant(
[False, True, False, False, True, False, True, True])
indices, mask = tensor_utils.padded_where(condition, 6)
self.assertAllEqual(indices, [1, 4, 6, 7, 0, 0])
self.assertAllEqual(mask, [1, 1, 1, 1, 0, 0])
# Case with batch dimention in the condition.
condition = tf.constant([[False, True, False, True, True, True],
[True, False, True, False, False, True],
[True, False, False, False, True, False]])
indices, mask = tensor_utils.padded_where(condition, 6)
self.assertAllEqual(indices, [
[1, 3, 4, 5, 0, 0],
[0, 2, 5, 0, 0, 0],
[0, 4, 0, 0, 0, 0],
])
self.assertAllEqual(mask, [
[1, 1, 1, 1, 0, 0],
[1, 1, 1, 0, 0, 0],
[1, 1, 0, 0, 0, 0],
])
def test_padded_where_truncation(self):
"""Basic truncation test for padded_where."""
# Case where all condition values are True, in the left of the tensor.
condition = tf.constant([True] * 4 + [False] * 4)
indices, mask = tensor_utils.padded_where(condition, 3)
self.assertAllEqual(indices, [0, 1, 2])
self.assertAllEqual(mask, [1, 1, 1])
# Case where True condition values are right-shifted.
condition = tf.constant([False] * 4 + [True] * 4)
indices, mask = tensor_utils.padded_where(condition, 3)
self.assertAllEqual(indices, [4, 5, 6])
self.assertAllEqual(mask, [1, 1, 1])
# Case where True condition values non-contiguous.
condition = tf.constant(
[False, True, False, False, True, False, True, True])
indices, mask = tensor_utils.padded_where(condition, 3)
self.assertAllEqual(indices, [1, 4, 6])
self.assertAllEqual(mask, [1, 1, 1])
# Case with batch dimention in the condition.
condition = tf.constant([[False, True, False, True, True, True],
[True, False, True, False, False, True],
[True, True, False, False, True, False]])
indices, mask = tensor_utils.padded_where(condition, 3)
self.assertAllEqual(indices, [
[1, 3, 4],
[0, 2, 5],
[0, 1, 4],
])
self.assertAllEqual(mask, [
[1, 1, 1],
[1, 1, 1],
[1, 1, 1],
])