def test_beam_width_changes():
def the_lm(s):
if s == 'A':
return 0.5
return 1
labels = ['_', 'A', ' ']
samples = np.array([[0.8, 0.2, 0], [0.7, 0.3, 0], [0.6, 0.4, 0],
[0.0, 0.0, 1]])
res = prefix_beam_search(samples,
labels,
lm=the_lm,
return_weights=False,
k=25,
alpha=1,
beta=0)
res2 = prefix_beam_search(samples,
labels,
lm=the_lm,
return_weights=False,
k=1,
alpha=1,
beta=0)
assert res == ' '
assert res2 == 'A '
def test_sanity():
sample = np.zeros((10, len(english_labels)))
sample[0, 2] = 0.5
sample[1, 20] = 0.5
sample[2, 19] = 0.5
sample[3:, 0] = 0.5
res = prefix_beam_search(sample, english_labels)
assert res == 'ASR'
def test_beam_is_not_greedy():
'''
Example from https://towardsdatascience.com/beam-search-decoding-in-ctc-trained-neural-networks-51889a3d85a7
Shows that beam search can find a path that greedy decoding can not.
'''
labels = ['_', 'A', 'B']
samples = np.array([0.8, 0.2, 0], [0.6, 0.4, 0])
res = prefix_beam_search(samples,
labels,
blank_index=0,
return_weights=True)
assert res == ('A', 0.52)
greedy_res = greedy_decode(samples, labels)
assert greedy_res == ''
def test_beam_is_not_greedy():
'''
Example from https://towardsdatascience.com/beam-search-decoding-in-ctc-trained-neural-networks-51889a3d85a7
Shows that beam search can find a path that greedy decoding can not.
'''
labels = ['_', 'A', 'B', ' ']
samples = np.array([[0.8, 0.2, 0, 0], [0.6, 0.4, 0, 0]])
res = prefix_beam_search(samples,
labels,
blank_index=0,
return_weights=True)
assert res == ('A', 0.52)
greedy_decoder = GreedyDecoder(labels, blank_index=0)
greedy_res = greedy_decoder.decode(torch.FloatTensor(samples).unsqueeze(0),
sizes=None)
assert greedy_res == ['']
def test_inconsistent_sizes():
sample = np.zeros((10, len(english_labels) - 1))
with pytest.raises(AssertionError) as exc_info:
_ = prefix_beam_search(sample, english_labels)
assert exc_info is not None