def cosine_similarity(vec1, vec2):
vec1 = vec1.reshape(1, -1)
vec2 = vec2.reshape(1, -1)
alpha = sk_cosine_similarity(vec1, vec2)
alpha = round(alpha[0][0], 5)
# print("Custom sim: ", _cosine_similarity(vec1, vec2), "Sk sim:", alpha)
# Custom cosine function and sk learn are almost similar
assert alpha >= 0.0 and alpha <= 1.0, print("alpha is ", alpha)
return alpha
def test_cosine_similarity(self):
v1 = [
-0.046193234622478485,
-0.09216824918985367,
0.023753443732857704,
-0.03982221707701683,
0.030631808564066887,
0.06340867280960083,
-0.09439295530319214,
-0.1576867550611496,
0.459428995847702,
-0.22166694700717926,
0.21970123052597046,
0.19883397221565247,
-0.19289985299110413,
-0.157765731215477,
0.0013831154210492969,
0.29028451442718506,
0.18202221393585205,
0.14411108195781708,
0.43273560404777527,
-0.31332970261573792,
]
v2 = [
0.23711472749710083,
0.0747479647397995,
0.20933881402015686,
-0.1695360243320465,
0.2809278070926666,
0.2502232491970062,
-0.0907953605055809,
0.07467399537563324,
-0.04727679118514061,
-0.028494318947196007,
-0.0278947614133358,
0.2525108754634857,
-0.06464426219463348,
0.18594379723072052,
0.13334108889102936,
0.3466702401638031,
0.30664315819740295,
0.10267733037471771,
0.04714057222008705,
0.1208021491765976,
]
similarity = round(cosine_similarity(v1, v2), 5)
compare = round(sk_cosine_similarity([v1], [v2])[0][0], 5)
print(similarity)
self.assertEqual(similarity, compare)
def cosine_similarity(vec1, vec2):
vec1 = vec1.reshape(1, -1)
vec2 = vec2.reshape(1, -1)
alpha = sk_cosine_similarity(vec1, vec2)
alpha = alpha[0][0]
return alpha
def fill_similarity_matrix(self, src_filename, dst_filename):
self.centered_training_coo = sparse.load_npz(src_filename)
similarities_sparse = sk_cosine_similarity(
self.centered_training_coo.tocsr(), dense_output=False)
sparse.save_npz(dst_filename, similarities_sparse)
return similarities_sparse