Pairwise similarity/similarity 矩阵计算优化

Pairwise similarity/similarity matrix calculation optimization

问题定义


问题

如何优化计算大量向量的成对余弦相似度(估计适合)?

正式定义

对于包含向量的两个集合 (A, B) - 需要为每个 a 和 b 生成成对余弦相似度 sim(a_i, b_j)。 (余弦相似度矩阵也适用,因为它很容易从矩阵转换为成对矩阵。)


我为什么要寻求帮助


这看起来是一个普遍的问题,因为在计算生物学、推荐系统等中需要计算这样的距离。但是我还没有找到一些合理的解决方案。

我无法解决的问题

根据定义,此问题的复杂度为 O(len_A * len_B * O(similarity_function)),因此 A 和 B 集合中的 10^6 个向量趋于巨大 运行时间

我对未来方向的假设

看起来,我们在这里做了很多无用的工作,因为相似性不是独立的(如果我们为百万个向量计算了 a_i 的相似性,并且 b_j 非常相似a_i - 我们计算了 900k 个向量的 b_j 相似度,我们可以估计与其余 100k 个向量的 b_j 相似度)。我假设这里可以使用索引之类的东西。



其他详细信息


  1. A 和 B 不相交。
  2. 向量维数已经减少到最小的合理值。
  3. 不需要简单的for循环优化。简而言之 - 这是优化这个的简短 guide - 最简单的循环给出了算法的清晰说明。
  4. 如果有一种算法也可以进行估计,我很感兴趣,所以如果我们的相似度足够接近但与真实的不完全相同也没关系。
  5. 不需要并行化。
  6. 我知道生成的相似度矩阵会很大。
  7. 我也很感兴趣,如果这是一种算法,它只允许从集合 B 中为集合 A 中的每个向量获取最相似的向量。

感谢您的参与。


代码示例


要求

python==3.6
pandas==0.25.0
scikit-learn==0.21.3
numpy==1.17.1

正在生成虚拟数据

import pandas as pd
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity 

df_1 = pd.DataFrame({'object_id_1': range(10),
                   'feature_0': np.random.uniform(0,1,10),
                   'feature_1': np.random.uniform(0,1,10),
                   'feature_2': np.random.uniform(0,1,10),
                   'feature_3':np.random.uniform(0,1,10)})

df_2 = pd.DataFrame({'object_id_2': range(10,20),
                   'feature_0': np.random.uniform(0,1,10),
                   'feature_1': np.random.uniform(0,1,10),
                   'feature_2': np.random.uniform(0,1,10),
                   'feature_3':np.random.uniform(0,1,10)})

相似性生成函数

def get_similarities(df_1: pd.DataFrame, df_2: pd.DataFrame, meaningful_features:list) -> pd.DataFrame:
    '''
    This function generates features based similarity scores, between two groups of objects
    
    Parameters
    ----------
    df_1: pandas.DataFrame
        DataFrame with features, and id_s of objects
    df_2: pandas.DataFrame
        DataFrame with features, and id_s of objects which has no id_s same to df_1
    meaningful_features: list
        Features columns to calculate similarity on
        
    Returns
    ----------
        similarities_of_objects: pandas.DataFrame
            DataFrame, with columns 'object_id_1', 'object_id_2', 'similarity', 
            where we have features similarity, for each object_1-object_2 pair. 
            Similarity - symmetric.  
    '''

    objects_1 = [] #  list of all objects from df_1
    objects_2 = [] #  list of all objects from df_2
    similarities = [] #  list of scores for object_1-object_2 pairs

    for object_1 in df_1['object_id_1'].unique():
        features_vector_1 = df_1[df_1['object_id_1'] == object_1][meaningful_features] # object_1 features vector
        
        for object_2 in df_2['object_id_2'].unique():
            features_vector_2 = df_2[df_2['object_id_2'] == object_2][meaningful_features] # object_2 features vector
            
            objects_1.append(object_1)
            objects_2.append(object_2)
            similarities.append(cosine_similarity(X = np.array(features_vector_1)
                                    ,Y = np.array(features_vector_2)).item()) # similarities of vectors 
    
    sim_o1_to_o2 = pd.DataFrame()

    sim_o1_to_o2['objects_1']= objects_1
    sim_o1_to_o2['objects_2']= objects_2
    sim_o1_to_o2['similarity']= similarities

    return sim_o1_to_o2

产生相似性

get_similarities(df_1,df_2, ['feature_0', 'feature_1', 'feature_2'])

使用Faiss

import faiss

dimension = 100

value1 = np.random.random((n, dimension)).astype('float32')
index = faiss.IndexFlatL2(d)
index.add(value1)

xq = value2
k= len(value1)
D, I = index.search(xq, k) 

注意这里D是距离,I是值的Index

此外,value1 和 value2 只是 NumPy 数组。

PS: 先安装faiss

pip install faiss

如何从欧氏距离得到余弦相似度


仅针对最相似的向量


,还有计算欧几里得距离的替代方法,尤其是在您只需要顶部相似向量而不是整个相似矩阵的情况下。

用@Abhik Sarka提出的方法解决


这是我发布的确切问题的解决方案,使用@Abhik Sarkar 提出的方法。要具有余弦相似性,请确保您的向量之前已归一化。 该解决方案还允许您根据需要生成尽可能多的相似点,而不是必需的完整矩阵。

免责声明:解决方案侧重于可读性,而不是性能。

要求

python==3.6
pandas==0.25.0
numpy==1.17.1
faiss==1.5.3 

正在生成虚拟数据

import pandas as pd
import numpy as np
import faiss 

df_1 = pd.DataFrame({'object_id_1': range(10),
                   'feature_0': np.random.uniform(0,1,10),
                   'feature_1': np.random.uniform(0,1,10),
                   'feature_2': np.random.uniform(0,1,10),
                   'feature_3':np.random.uniform(0,1,10)})

df_2 = pd.DataFrame({'object_id_2': range(10,20),
                   'feature_0': np.random.uniform(0,1,10),
                   'feature_1': np.random.uniform(0,1,10),
                   'feature_2': np.random.uniform(0,1,10),
                   'feature_3':np.random.uniform(0,1,10)})

相似性生成函数

def get_similarities(df_1: pd.DataFrame, 
                     df_2: pd.DataFrame, 
                     meaningful_features:list, 
                     n_neighbors:int = df_2.shape[0])->pd.DataFrame:
    '''
    This function generates features based similarity scores, between to groups of reviews
    
    Parameters
    ----------
    df_1: pandas.DataFrame
        DataFrame with features, and id_s of objects
    df_2: pandas.DataFrame
        DataFrame with features, and id_s of objects which has no id_s same to df_1
    meaningful_features: list
        Features columns to calculate similarity on
    n_neighbors: int
        Number of most similar objects_2 for every object_1. By default - full similarity matrix generated.
        (default = df_2.shape[0]) 
    
    Returns
    ----------
        similarities_of_objects: pandas.DataFrame
            DataFrame, with columns 'object_id_1', 'object_id_2', 'similarity', 
            where we have features similarity, for each object_1-object_2 pair. 
            Similarity - symmetric.  
    '''
    d = len(meaningful_features) #  dimensionality
    
    res = np.empty(shape=[1, 3]) #  res initialization
    
    xb = np.float32(df_1[meaningful_features].values)
    xb = np.ascontiguousarray(xb)
    
    xq = np.float32(df_2[meaningful_features].values)
    xq = np.ascontiguousarray(xq)

    index = faiss.IndexFlatL2(d) #  build the index
    index.add(xb)                #  add vectors to the index
    
    D, I = index.search(xq, n_neighbors)     # actual search
    
    for i in range(I.shape[0]): 
        object_id_1_v = [df_1["object_id_1"].iloc[i]]*n_neighbors
        object_id_2_v = df_2["object_id_2"].iloc[I[i]]
        similarities = 1-D[i]/2
        
        neighbors_scores_for_target = np.stack((object_id_1_v, object_id_2_v, similarities), axis=-1)
        res = np.concatenate((res, neighbors_scores_for_target))
        
    res = res[1:] #  remove line we've created during res initialization
    
    resulting_df = pd.DataFrame({'object_id_1': res[:, 0], 
                                 'object_id_2': res[:, 1],
                                 'similarity':  res[:, 2] })

    
    return resulting_df

产生相似性

get_similarities(df_1,df_2, ['feature_0', 'feature_1', 'feature_2'])