distance.py

# -*- coding:utf-8 -*-
import math
import numpy as np

def cosine_similarity_slow(v1, v2):
    # compute cosine similarity of v1 to v2: (v1 dot v2)/{||v1||*||v2||)
    sumxx, sumxy, sumyy = 0, 0, 0
    for i in range(len(v1)):
        x = v1[i]
        y = v2[i]
        sumxx += x * x
        sumyy += y * y
        sumxy += x * y
    return sumxy / math.sqrt(sumxx * sumyy)

def cosine_similarity(v1, v2):
    cosV12 = np.dot(v1, v2)/(np.linalg.norm(v1)*np.linalg.norm(v2))
    return cosV12

def cosine_distance(v1, v2):
    return 1 - cosine_similarity(v1, v2)

    
def L2_distance(v1, v2):
    return np.sqrt(np.sum(np.square(v1 - v2)))  

    
def SSD_distance(v1, v2):
    return np.sum(np.square(v1 - v2))

    
def get_distance(dist_type):
    loss_map = {'cosine':cosine_distance, 'L2':L2_distance, 'SSD':SSD_distance}
    return loss_map[dist_type]