def test_pad_word_sequence_handles_words_and_characters(self):
instance = IndexedTrueFalseInstance([[1, 2], [3, 1, 2]], True)
padded = instance.pad_word_sequence(instance.word_indices, {
'word_sequence_length': 3,
'word_character_length': 4
})
assert padded == [[0, 0, 0, 0], [1, 2, 0, 0], [3, 1, 2, 0]]
def test_as_training_data_produces_correct_numpy_arrays(self):
instance = IndexedTrueFalseInstance([1, 2, 3, 4], True)
inputs, label = instance.as_training_data()
assert numpy.all(label == numpy.asarray([0, 1]))
assert numpy.all(inputs == numpy.asarray([1, 2, 3, 4]))
instance.label = False
_, label = instance.as_training_data()
assert numpy.all(label == numpy.asarray([1, 0]))
def test_pad_word_sequence_handles_words_and_characters_greater(self):
instance = IndexedTrueFalseInstance([[1, 2], [3, 1, 2]], True)
padded = instance.pad_word_sequence(instance.word_indices, {
'num_sentence_words': 5,
'num_word_characters': 4
})
assert padded == [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0],
[1, 2, 0, 0], [3, 1, 2, 0]]
def setUp(self):
self.base_instance = IndexedTrueFalseInstance([1, 2], True)
self.background_instances = [
IndexedTrueFalseInstance([2, 3, 4], None),
IndexedTrueFalseInstance([4, 5], None)
]
self.qa_instance = IndexedQuestionAnswerInstance([1, 2, 3],
[[2, 3], [4], [5, 6]],
1)
def setUp(self):
# We'll just test with underlying IndexedTrueFalseInstances for most of these, because it's
# simpler.
self.instance = IndexedMultipleTrueFalseInstance([
IndexedTrueFalseInstance([1], False),
IndexedTrueFalseInstance([2, 3, 4], False),
IndexedTrueFalseInstance([5, 6], True),
IndexedTrueFalseInstance([7, 8], False)
], 2)
def test_get_lengths_works_with_words_and_characters(self):
instance = IndexedTrueFalseInstance([[1, 2], [3, 1, 2]], True)
assert instance.get_lengths() == {
'word_sequence_length': 2,
'word_character_length': 3
}
def test_as_training_data_produces_correct_numpy_arrays_with_background_instances(
self):
instance = IndexedMultipleTrueFalseInstance([
IndexedBackgroundInstance(IndexedTrueFalseInstance([1, 2], False),
[
IndexedTrueFalseInstance([2], None),
IndexedTrueFalseInstance([3], None)
]),
IndexedBackgroundInstance(IndexedTrueFalseInstance([3, 4], False),
[
IndexedTrueFalseInstance([5], None),
IndexedTrueFalseInstance([6], None)
]),
IndexedBackgroundInstance(IndexedTrueFalseInstance([5, 6], False),
[
IndexedTrueFalseInstance([8], None),
IndexedTrueFalseInstance([9], None)
]),
IndexedBackgroundInstance(IndexedTrueFalseInstance([7, 8], True), [
IndexedTrueFalseInstance([11], None),
IndexedTrueFalseInstance([12], None)
]),
], 3)
(word_arrays, background_arrays), label = instance.as_training_data()
assert numpy.all(label == numpy.asarray([0, 0, 0, 1]))
assert numpy.all(
word_arrays == numpy.asarray([[1, 2], [3, 4], [5, 6], [7, 8]]))
assert numpy.all(background_arrays == numpy.asarray(
[[[2], [3]], [[5], [6]], [[8], [9]], [[11], [12]]]))
def test_pad_adds_zeros_on_left(self):
instance = IndexedTrueFalseInstance([1, 2, 3, 4], True)
instance.pad({'word_sequence_length': 5})
assert instance.word_indices == [0, 1, 2, 3, 4]
def test_get_lengths_returns_length_of_word_indices(self):
instance = IndexedTrueFalseInstance([1, 2, 3, 4], True)
assert instance.get_lengths() == {'word_sequence_length': 4}
def test_pad_truncates_from_right(self):
instance = IndexedTrueFalseInstance([1, 2, 3, 4], True)
instance.pad({'word_sequence_length': 3})
assert instance.word_indices == [2, 3, 4]