def test_namespaces(self):
data_indexer = DataIndexer()
initial_vocab_size = data_indexer.get_vocab_size()
word_index = data_indexer.add_word_to_index("word", namespace='1')
assert "word" in data_indexer.words_in_index(namespace='1')
assert data_indexer.get_word_index("word", namespace='1') == word_index
assert data_indexer.get_word_from_index(word_index,
namespace='1') == "word"
assert data_indexer.get_vocab_size(
namespace='1') == initial_vocab_size + 1
# Now add it again, in a different namespace and a different word, and make sure it's like
# new.
word2_index = data_indexer.add_word_to_index("word2", namespace='2')
word_index = data_indexer.add_word_to_index("word", namespace='2')
assert "word" in data_indexer.words_in_index(namespace='2')
assert "word2" in data_indexer.words_in_index(namespace='2')
assert data_indexer.get_word_index("word", namespace='2') == word_index
assert data_indexer.get_word_index("word2",
namespace='2') == word2_index
assert data_indexer.get_word_from_index(word_index,
namespace='2') == "word"
assert data_indexer.get_word_from_index(word2_index,
namespace='2') == "word2"
assert data_indexer.get_vocab_size(
namespace='2') == initial_vocab_size + 2
def test_unknown_token(self):
# pylint: disable=protected-access
# We're putting this behavior in a test so that the behavior is documented. There is
# solver code that depends in a small way on how we treat the unknown token, so any
# breaking change to this behavior should break a test, so you know you've done something
# that needs more consideration.
data_indexer = DataIndexer()
oov_token = data_indexer._oov_token
oov_index = data_indexer.get_word_index(oov_token)
assert oov_index == 1
assert data_indexer.get_word_index("unseen word") == oov_index
def test_add_word_to_index_gives_consistent_results(self):
data_indexer = DataIndexer()
initial_vocab_size = data_indexer.get_vocab_size()
word_index = data_indexer.add_word_to_index("word")
assert "word" in data_indexer.words_in_index()
assert data_indexer.get_word_index("word") == word_index
assert data_indexer.get_word_from_index(word_index) == "word"
assert data_indexer.get_vocab_size() == initial_vocab_size + 1
# Now add it again, and make sure nothing changes.
data_indexer.add_word_to_index("word")
assert "word" in data_indexer.words_in_index()
assert data_indexer.get_word_index("word") == word_index
assert data_indexer.get_word_from_index(word_index) == "word"
assert data_indexer.get_vocab_size() == initial_vocab_size + 1
def test_set_from_file(self):
# pylint: disable=protected-access
vocab_filename = self.TEST_DIR + 'vocab_file'
with codecs.open(vocab_filename, 'w', 'utf-8') as vocab_file:
vocab_file.write('<S>\n')
vocab_file.write('</S>\n')
vocab_file.write('<UNK>\n')
vocab_file.write('a\n')
vocab_file.write('word\n')
vocab_file.write('another\n')
data_indexer = DataIndexer()
data_indexer.set_from_file(vocab_filename, oov_token="<UNK>")
assert data_indexer._oov_token == "<UNK>"
assert data_indexer.get_word_index("random string") == 3
assert data_indexer.get_word_index("<S>") == 1
assert data_indexer.get_word_index("</S>") == 2
assert data_indexer.get_word_index("<UNK>") == 3
assert data_indexer.get_word_index("a") == 4
assert data_indexer.get_word_index("word") == 5
assert data_indexer.get_word_index("another") == 6
assert data_indexer.get_word_from_index(
0) == data_indexer._padding_token
assert data_indexer.get_word_from_index(1) == "<S>"
assert data_indexer.get_word_from_index(2) == "</S>"
assert data_indexer.get_word_from_index(3) == "<UNK>"
assert data_indexer.get_word_from_index(4) == "a"
assert data_indexer.get_word_from_index(5) == "word"
assert data_indexer.get_word_from_index(6) == "another"
def test_get_embedding_layer_initializes_unseen_words_randomly_not_zero(self):
data_indexer = DataIndexer()
data_indexer.add_word_to_index("word2")
embeddings_filename = self.TEST_DIR + "embeddings.gz"
with gzip.open(embeddings_filename, 'wb') as embeddings_file:
embeddings_file.write("word 1.0 2.3 -1.0\n".encode('utf-8'))
embedding_layer = PretrainedEmbeddings.get_embedding_layer(embeddings_filename, data_indexer)
word_vector = embedding_layer._initial_weights[0][data_indexer.get_word_index("word2")]
assert not numpy.allclose(word_vector, numpy.asarray([0.0, 0.0, 0.0]))
def test_get_embedding_layer_actually_initializes_word_vectors_correctly(self):
data_indexer = DataIndexer()
data_indexer.add_word_to_index("word")
embeddings_filename = self.TEST_DIR + "embeddings.gz"
with gzip.open(embeddings_filename, 'wb') as embeddings_file:
embeddings_file.write("word 1.0 2.3 -1.0\n".encode('utf-8'))
embedding_layer = PretrainedEmbeddings.get_embedding_layer(embeddings_filename, data_indexer)
word_vector = embedding_layer._initial_weights[0][data_indexer.get_word_index("word")]
assert numpy.allclose(word_vector, numpy.asarray([1.0, 2.3, -1.0]))
def test_to_indexed_instance_converts_correctly(self):
instance = SnliInstance("a b", "d e f", "entails")
data_indexer = DataIndexer()
a_index = data_indexer.add_word_to_index("a")
d_index = data_indexer.add_word_to_index("d")
oov_index = data_indexer.get_word_index(data_indexer._oov_token) # pylint: disable=protected-access
indexed_instance = instance.to_indexed_instance(data_indexer)
assert indexed_instance.first_sentence_indices == [a_index, oov_index]
assert indexed_instance.second_sentence_indices == [d_index, oov_index, oov_index]
assert indexed_instance.label == instance.label
def test_to_indexed_instance_converts_correctly(self):
instance = QuestionAnswerInstance("a A b", ["d", "e f D"], 1)
data_indexer = DataIndexer()
a_index = data_indexer.add_word_to_index("a")
d_index = data_indexer.add_word_to_index("d")
oov_index = data_indexer.get_word_index(data_indexer._oov_token) # pylint: disable=protected-access
indexed_instance = instance.to_indexed_instance(data_indexer)
assert indexed_instance.question_indices == [a_index, a_index, oov_index]
assert len(indexed_instance.option_indices) == 2
assert indexed_instance.option_indices[0] == [d_index]
assert indexed_instance.option_indices[1] == [oov_index, oov_index, d_index]
def test_get_embedding_layer_skips_inconsistent_lines(self):
data_indexer = DataIndexer()
data_indexer.add_word_to_index("word1")
data_indexer.add_word_to_index("word2")
embeddings_filename = self.TEST_DIR + "embeddings.gz"
with gzip.open(embeddings_filename, 'wb') as embeddings_file:
embeddings_file.write("word1 1.0 2.3 -1.0\n".encode('utf-8'))
embeddings_file.write("word2 0.1 0.4 \n".encode('utf-8'))
embedding_layer = PretrainedEmbeddings.get_embedding_layer(embeddings_filename, data_indexer)
word_vector = embedding_layer.initial_weights[0][data_indexer.get_word_index("word2")]
assert not numpy.allclose(word_vector[:2], numpy.asarray([0.1, 0.4]))
