def test_push_does_not_add_hypothesis_that_falls_below_beam_threshold(self):
# arrange
stack = _Stack(3, 0.5)
poor_hypothesis = _Hypothesis(0.01)
# act
stack.push(_Hypothesis(0.9)) # greatly superior hypothesis
stack.push(poor_hypothesis)
# assert
self.assertFalse(poor_hypothesis in stack)
def setUp(self):
root = _Hypothesis()
child = _Hypothesis(raw_score=0.5,
src_phrase_span=(3, 7),
trg_phrase=('hello', 'world'),
previous=root)
grandchild = _Hypothesis(raw_score=0.4,
src_phrase_span=(1, 2),
trg_phrase=('and', 'goodbye'),
previous=child)
self.hypothesis_chain = grandchild
def test_push_does_not_add_hypothesis_that_falls_below_beam_threshold(self):
# arrange
stack = _Stack(3, 0.5)
poor_hypothesis = _Hypothesis(0.01)
# act
stack.push(_Hypothesis(0.9)) # greatly superior hypothesis
stack.push(poor_hypothesis)
# assert
self.assertFalse(poor_hypothesis in stack)
def test_best_returns_the_best_hypothesis(self):
# arrange
stack = _Stack(3)
best_hypothesis = _Hypothesis(0.99)
# act
stack.push(_Hypothesis(0.0))
stack.push(best_hypothesis)
stack.push(_Hypothesis(0.5))
# assert
self.assertEqual(stack.best(), best_hypothesis)
def test_best_returns_the_best_hypothesis(self):
# arrange
stack = _Stack(3)
best_hypothesis = _Hypothesis(0.99)
# act
stack.push(_Hypothesis(0.0))
stack.push(best_hypothesis)
stack.push(_Hypothesis(0.5))
# assert
self.assertEqual(stack.best(), best_hypothesis)
def test_push_bumps_off_worst_hypothesis_when_stack_is_full(self):
# arrange
stack = _Stack(3)
poor_hypothesis = _Hypothesis(0.01)
# act
stack.push(_Hypothesis(0.2))
stack.push(poor_hypothesis)
stack.push(_Hypothesis(0.1))
stack.push(_Hypothesis(0.3))
# assert
self.assertFalse(poor_hypothesis in stack)
def test_push_bumps_off_worst_hypothesis_when_stack_is_full(self):
# arrange
stack = _Stack(3)
poor_hypothesis = _Hypothesis(0.01)
# act
stack.push(_Hypothesis(0.2))
stack.push(poor_hypothesis)
stack.push(_Hypothesis(0.1))
stack.push(_Hypothesis(0.3))
# assert
self.assertFalse(poor_hypothesis in stack)
def test_push_removes_hypotheses_that_fall_below_beam_threshold(self):
# arrange
stack = _Stack(3, 0.5)
poor_hypothesis = _Hypothesis(0.01)
worse_hypothesis = _Hypothesis(0.009)
# act
stack.push(poor_hypothesis)
stack.push(worse_hypothesis)
stack.push(_Hypothesis(0.9)) # greatly superior hypothesis
# assert
self.assertFalse(poor_hypothesis in stack)
self.assertFalse(worse_hypothesis in stack)
def test_push_removes_hypotheses_that_fall_below_beam_threshold(self):
# arrange
stack = _Stack(3, 0.5)
poor_hypothesis = _Hypothesis(0.01)
worse_hypothesis = _Hypothesis(0.009)
# act
stack.push(poor_hypothesis)
stack.push(worse_hypothesis)
stack.push(_Hypothesis(0.9)) # greatly superior hypothesis
# assert
self.assertFalse(poor_hypothesis in stack)
self.assertFalse(worse_hypothesis in stack)
def setUp(self):
root = _Hypothesis()
child = _Hypothesis(
raw_score=0.5,
src_phrase_span=(3, 7),
trg_phrase=("hello", "world"),
previous=root,
)
grandchild = _Hypothesis(
raw_score=0.4,
src_phrase_span=(1, 2),
trg_phrase=("and", "goodbye"),
previous=child,
)
self.hypothesis_chain = grandchild
def setUp(self):
root = _Hypothesis()
child = _Hypothesis(
raw_score=0.5,
src_phrase_span=(3, 7),
trg_phrase=('hello', 'world'),
previous=root
)
grandchild = _Hypothesis(
raw_score=0.4,
src_phrase_span=(1, 2),
trg_phrase=('and', 'goodbye'),
previous=child
)
self.hypothesis_chain = grandchild
def test_translation_so_far_for_empty_hypothesis(self):
# arrange
hypothesis = _Hypothesis()
# act
translation = hypothesis.translation_so_far()
# assert
self.assertEqual(translation, [])
def test_untranslated_spans_for_empty_hypothesis(self):
# arrange
hypothesis = _Hypothesis()
# act
untranslated_spans = hypothesis.untranslated_spans(10)
# assert
self.assertEqual(untranslated_spans, [(0, 10)])
def test_untranslated_spans_for_empty_hypothesis(self):
# arrange
hypothesis = _Hypothesis()
# act
untranslated_spans = hypothesis.untranslated_spans(10)
# assert
self.assertEqual(untranslated_spans, [(0, 10)])
def test_translation_so_far_for_empty_hypothesis(self):
# arrange
hypothesis = _Hypothesis()
# act
translation = hypothesis.translation_so_far()
# assert
self.assertEqual(translation, [])
def test_distortion_score_of_first_expansion(self):
# arrange
stack_decoder = StackDecoder(None, None)
stack_decoder.distortion_factor = 0.5
hypothesis = _Hypothesis()
# act
score = stack_decoder.distortion_score(hypothesis, (8, 10))
# assert
# expansion from empty hypothesis always has zero distortion cost
self.assertEqual(score, 0.0)
def test_valid_phrases(self):
# arrange
hypothesis = _Hypothesis()
# mock untranslated_spans method
hypothesis.untranslated_spans = lambda _: [(0, 2), (3, 6)]
all_phrases_from = [[1, 4], [2], [], [5], [5, 6, 7], [], [7]]
# act
phrase_spans = StackDecoder.valid_phrases(all_phrases_from, hypothesis)
# assert
self.assertEqual(phrase_spans, [(0, 1), (1, 2), (3, 5), (4, 5), (4, 6)])
def test_valid_phrases(self):
# arrange
hypothesis = _Hypothesis()
# mock untranslated_spans method
hypothesis.untranslated_spans = lambda _: [(0, 2), (3, 6)]
all_phrases_from = [[1, 4], [2], [], [5], [5, 6, 7], [], [7]]
# act
phrase_spans = StackDecoder.valid_phrases(all_phrases_from, hypothesis)
# assert
self.assertEqual(phrase_spans, [(0, 1), (1, 2), (3, 5), (4, 5), (4, 6)])
def test_distortion_score_of_first_expansion(self):
# arrange
stack_decoder = StackDecoder(None, None)
stack_decoder.distortion_factor = 0.5
hypothesis = _Hypothesis()
# act
score = stack_decoder.distortion_score(hypothesis, (8, 10))
# assert
# expansion from empty hypothesis always has zero distortion cost
self.assertEqual(score, 0.0)
def test_distortion_score(self):
# arrange
stack_decoder = StackDecoder(None, None)
stack_decoder.distortion_factor = 0.5
hypothesis = _Hypothesis()
hypothesis.src_phrase_span = (3, 5)
# act
score = stack_decoder.distortion_score(hypothesis, (8, 10))
# assert
expected_score = log(stack_decoder.distortion_factor) * (8 - 5)
self.assertEqual(score, expected_score)
def test_distortion_score(self):
# arrange
stack_decoder = StackDecoder(None, None)
stack_decoder.distortion_factor = 0.5
hypothesis = _Hypothesis()
hypothesis.src_phrase_span = (3, 5)
# act
score = stack_decoder.distortion_score(hypothesis, (8, 10))
# assert
expected_score = log(stack_decoder.distortion_factor) * (8 - 5)
self.assertEqual(score, expected_score)
def test_future_score(self):
# arrange: sentence with 8 words; words 2, 3, 4 already translated
hypothesis = _Hypothesis()
hypothesis.untranslated_spans = lambda _: [(0, 2), (5, 8)] # mock
future_score_table = defaultdict(lambda: defaultdict(float))
future_score_table[0][2] = 0.4
future_score_table[5][8] = 0.5
stack_decoder = StackDecoder(None, None)
# act
future_score = stack_decoder.future_score(hypothesis, future_score_table, 8)
# assert
self.assertEqual(future_score, 0.4 + 0.5)
def test_future_score(self):
# arrange: sentence with 8 words; words 2, 3, 4 already translated
hypothesis = _Hypothesis()
hypothesis.untranslated_spans = lambda _: [(0, 2), (5, 8)] # mock
future_score_table = defaultdict(lambda: defaultdict(float))
future_score_table[0][2] = 0.4
future_score_table[5][8] = 0.5
stack_decoder = StackDecoder(None, None)
# act
future_score = stack_decoder.future_score(hypothesis, future_score_table, 8)
# assert
self.assertEqual(future_score, 0.4 + 0.5)
def setUp(self):
root = _Hypothesis()
child = _Hypothesis(raw_score=0.5, src_phrase_span=(3, 7), trg_phrase=("hello", "world"), previous=root)
grandchild = _Hypothesis(raw_score=0.4, src_phrase_span=(1, 2), trg_phrase=("and", "goodbye"), previous=child)
self.hypothesis_chain = grandchild