def test_calculation_opposite(self):
vector1 = [0, 1]
vector2 = [0, -1]
cosine_similarity = CosineSimilarity()
assert cosine_similarity.compute(vector1, vector2) == -1
def test_calculation_identical(self):
vector1 = [0, 1]
vector2 = [0, 2]
cosine_similarity = CosineSimilarity()
assert cosine_similarity.compute(vector1, vector2) == 1
def test_calculation_orthogonal(self):
vector1 = [0, 1]
vector2 = [1, 0]
cosine_similarity = CosineSimilarity()
assert cosine_similarity.compute(vector1, vector2) == 0
def test_is_better_than_noise(self):
embedding_file = self._random_test_file()
with open(embedding_file, "w+") as file_output:
print("2 2\n" + "paris 0.0 3.0\n" + "berlin 1.0 0.0",
file=file_output)
embedding = EmbeddingFileParser.create_from_file(embedding_file)
assert embedding.random_cosine_noise == 0
cosine_similarity = CosineSimilarity()
assert cosine_similarity.is_better_than_noise(1, embedding)
assert not cosine_similarity.is_better_than_noise(0, embedding)
assert not cosine_similarity.is_better_than_noise(-1, embedding)
def test_equality(self):
assert EuclideanDistance() == EuclideanDistance()
assert EuclideanDistance() is not None
assert not EuclideanDistance() == CosineSimilarity()
def __init__(self, name, test_set):
super(CosineOutlierDetectionTask,
self).__init__(name, test_set, CosineSimilarity())
def test_batch_compute(self):
cosine_similarity = CosineSimilarity()
vectors = [[0.0, 4.32], [1.0, 0.0], [0.0, -1312.3], [-4.3, 0.0]]
result = cosine_similarity.batch_compute(vectors)
assert result == -(1 / 3)
def __init__(self, name, test_set):
super(CosineNeighborhoodTask, self).__init__(name, test_set,
CosineSimilarity())
def _calculate_random_cosine_noise(vectors):
sampled_vectors = Embedding._select_random_vectors(vectors)
return CosineSimilarity().batch_compute(sampled_vectors)