def test_word2idx_return_unknown_tag(vocabulary: dp.Vocabulary) -> None:
# given
word = "testnonexistingword"
expected_result = vocabulary.word2idx("<UNK>")
# when
result = vocabulary.word2idx(word)
# then
assert result == expected_result
def test_idx2word_raises_index_error(vocabulary: dp.Vocabulary) -> None:
# given
index = -1
expected_message = "No word is mapped to -1"
# when / then
with pytest.raises(IndexError) as result:
vocabulary.idx2word(index)
assert str(result.value) == expected_message
def test_decode_caption(vocabulary: dp.Vocabulary) -> None:
# given
caption = [vocabulary.word2idx("a"), vocabulary.word2idx("man")]
expected_result = "a man"
# when
result = dp.TextPipeline.decode_caption(vocabulary, caption)
# then
assert result == expected_result
def plot_embeddings(embeddings: np.array,
words: List[str],
vocab: dp.Vocabulary = dp.Vocabulary(),
out_path: str = None) -> None:
"""Plot word embeddings in 2D.
Args:
embeddings (np.array): Embeddings matrix
words (List[str]): List of words to be plotted
vocab (dp.Vocabulary, optional): Vocabulary. Defaults to dp.Vocabulary().
"""
E_reduced = reduce_to_k_dim(embeddings)
x = []
y = []
for word in words:
word_x, word_y = E_reduced[vocab.word2idx(word.lower())]
x.append(word_x)
y.append(word_y)
plt.annotate(word, (word_x + 0.01, word_y + 0.01),
color="blue",
fontsize=12)
plt.scatter(x, y, c="red")
if out_path is not None:
plt.savefig(out_path, dpi=100)
plt.show()
def test_word2idx(vocabulary: dp.Vocabulary) -> None:
# given
word = "a"
expected_result = 0
# when
result = vocabulary.word2idx(word)
# then
assert result == expected_result
def test_idx2word(vocabulary: dp.Vocabulary) -> None:
# given
index = 0
expected_result = "a"
# when
result = vocabulary.idx2word(index)
# then
assert result == expected_result