距离与相似度计算

一.余弦相似度

加速计算参考这篇文章

from math import *

def square_rooted(x):
    return round(sqrt(sum([a*a for a in x])), 3)

def cosine_similarity(x,y):
   numerator = sum(a*b for a, b in zip(x,y))
   denominator = square_rooted(x)*square_rooted(y)
   return round(numerator/float(denominator),3)

res = cosine_similarity([1, 0], [0,1])
print('==res:', res)

二.欧式距离

from math import *
def euclidean_distance(x, y):
    return sqrt(sum(pow(a - b, 2) for a, b in zip(x, y)))


res = euclidean_distance([0, 1], [1, 0])
print('res:', res)

三.曼哈顿距离

from math import *

def manhattan_distance(x,y):
    return sum(abs(a-b) for a, b in zip(x,y))
print(manhattan_distance([1, 0], [0, 1]))

四.汉明距离

两个等长字符串在对应位置上不同字符的数目。

def hamming_distance(s1, s2):
    """Return the Hamming distance between equal-length sequences"""
    if len(s1) != len(s2):
        raise ValueError("Undefined for sequences of unequal length")
    return sum(a != b for a, b in zip(s1, s2))

res = hamming_distance('12','13')
print('res:', res)

五.切比雪夫距离

切比雪夫距离起源于国际象棋中国王的走法，国际象棋中国王每次只能往周围的8格中走一步，那么如果要从棋盘中A格(x1,y1)走到B格(x2,y2)最少需要走几步？你会发现最少步数总是max(|x2-x1|,|y2-y1|)步。有一种类似的一种距离度量方法叫切比雪夫距离。

def chebyshev_distance(p, q):
    assert len(p) == len(q)
    return max([abs(x - y) for x, y in zip(p, q)])
res = chebyshev_distance([0,0], [1,3])
print('res:', res)

六.兰氏距离

def canberra_distance(p, q):
    n = len(p)
    distance = 0
    for i in range(n):
        if p[i] == 0 and q[i] == 0:
            distance += 0
        else:
            distance += abs(p[i] - q[i]) / (abs(p[i]) + abs(q[i]))
    return distance

res = canberra_distance([1,0], [0,1])
print('res:', res)

七.闵可夫斯基距离

p=2即为欧氏距离，而p=1时则为曼哈顿距离。当p取无穷时的极限情况下，可以得到切比雪夫距离：

def minkowski_distance(p, q, n):
    assert len(p) == len(q)
    return sum([abs(x - y) ** n for x, y in zip(p, q)]) ** (1. / n)


res = minkowski_distance([1, 0], [0, 1], n=2.)
print('res:', res)

八.编辑距离

编辑距离，又称Levenshtein距离（莱文斯坦距离也叫做Edit Distance），是指两个字串之间，由一个转成另一个所需的最少编辑操作次数，如果它们的距离越大，说明它们越是不同。许可的编辑操作包括将一个字符替换成另一个字符，插入一个字符，删除一个字符。

方法1:调包

import Levenshtein
texta = '者記聞新'
textb = '浪(第'
print(Levenshtein.distance(texta, textb))

方法2:动态规划

import os
import numpy as np
def edit_distance(S1,S2):
    #S1列　S2行
    mat = [[0] *(len(S1)+1) for i in range(len(S2)+1)]
    # print('mat:', mat)
    for i in range(len(S2)):
        mat[i+1][0] = mat[i][0]+1
    # print('mat:', mat)
    for i in range(len(S1)):
        mat[0][i+1] = mat[0][i]+1
    # print('mat:\n', np.array(mat))
    #相等就为０　不想等加１
    for i in range(len(S2)):
        for j in range(len(S1)):
            if S2[i] == S1[j]:
                # print('S2[i]:', S2[i])
                mat[i + 1][j + 1] = min(mat[i][j] + 0, mat[i + 1][j]+1, mat[i][j + 1]+1)
            else:
                mat[i + 1][j + 1] = min(mat[i][j] + 1, mat[i + 1][j]+1, mat[i][j + 1]+1)
    # print('mat:\n', np.array(mat))
    dis = mat[-1][-1]
    print('dis:', dis)
    return dis
# S1 = 'iva1'
# S2 = 'iva'
S2 = '者記聞新'
S1 = '浪(第'
dis = edit_distance(S1, S2)
print('dis:', dis)

九.杰卡德相似度

def jaccard_sim(a, b):
    unions = len(set(a).union(set(b)))
    intersections = len(set(a).intersection(set(b)))
    return intersections / unions

a = ['1', '0']
b = ['1', '1', '1']
res = jaccard_sim(a, b)
print('res:', res)

十.Dice距离

def dice_coefficient(a, b):
    """dice coefficient 2nt/na + nb."""
    intersections = len(set(a).intersection(set(b)))
    return intersections * 2.0/(len(set(a)) + len(set(b)))

res =  dice_coefficient(a = [1, 0], b =[0, 1])
print('===res:',res)