使用 scikit-learn 高效计算余弦相似度

Question

在预处理和转换（BOW、TF-IDF）数据之后，我需要计算其与数据集的每个其他元素的余弦相似度。目前，我这样做：

cs_title = [cosine_similarity(a, b) for a in tr_title for b in tr_title]
cs_abstract = [cosine_similarity(a, b) for a in tr_abstract for b in tr_abstract]
cs_mesh = [cosine_similarity(a, b) for a in pre_mesh for b in pre_mesh]
cs_pt = [cosine_similarity(a, b) for a in pre_pt for b in pre_pt]

在此示例中，每个输入变量（例如 ）tr_title是一个 SciPy 稀疏矩阵。然而，这段代码运行速度极其缓慢。我可以做些什么来优化代码，使其运行得更快？

Answer 1

为了提高性能，您应该用矢量化代码替换列表推导式。这可以通过 Numpy 轻松实现pdist，squareform如下面的代码片段所示：

import numpy as np
from sklearn.feature_extraction.text import CountVectorizer
from scipy.spatial.distance import pdist, squareform

titles = [
    'A New Hope',
    'The Empire Strikes Back',
    'Return of the Jedi',
    'The Phantom Menace',
    'Attack of the Clones',
    'Revenge of the Sith',
    'The Force Awakens',
    'A Star Wars Story',
    'The Last Jedi',
    ]

vectorizer = CountVectorizer()
X = vectorizer.fit_transform(titles)
cs_title = squareform(pdist(X.toarray(), 'cosine'))

演示：

In [87]: X
Out[87]: 
<9x21 sparse matrix of type '<type 'numpy.int64'>'
    with 30 stored elements in Compressed Sparse Row format>

In [88]: X.toarray()          
Out[88]: 
array([[0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0],
       [0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 1, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0],
       [1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 1, 0, 0, 0, 1, 0],
       [0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1],
       [0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0]], dtype=int64)

In [89]: vectorizer.get_feature_names()
Out[89]: 
[u'attack',
 u'awakens',
 u'back',
 u'clones',
 u'empire',
 u'force',
 u'hope',
 u'jedi',
 u'last',
 u'menace',
 u'new',
 u'of',
 u'phantom',
 u'return',
 u'revenge',
 u'sith',
 u'star',
 u'story',
 u'strikes',
 u'the',
 u'wars']

In [90]: np.set_printoptions(precision=2)

In [91]: print(cs_title)
[[ 0.    1.    1.    1.    1.    1.    1.    1.    1.  ]
 [ 1.    0.    0.75  0.71  0.75  0.75  0.71  1.    0.71]
 [ 1.    0.75  0.    0.71  0.5   0.5   0.71  1.    0.42]
 [ 1.    0.71  0.71  0.    0.71  0.71  0.67  1.    0.67]
 [ 1.    0.75  0.5   0.71  0.    0.5   0.71  1.    0.71]
 [ 1.    0.75  0.5   0.71  0.5   0.    0.71  1.    0.71]
 [ 1.    0.71  0.71  0.67  0.71  0.71  0.    1.    0.67]
 [ 1.    1.    1.    1.    1.    1.    1.    0.    1.  ]
 [ 1.    0.71  0.42  0.67  0.71  0.71  0.67  1.    0.  ]]

请注意，X.toarray().shapeyield是(9L, 21L)因为在上面的玩具示例中，有 9 个标题和 21 个不同的单词，而cs_title是一个9 x 9数组。