240722 knn 使用皮尔逊找到相似用户

import json
import numpy as np

# 计算皮尔逊系数
def pearson_score(dataset, user1, user2):
    if user1 not in dataset:
        raise TypeError('User ' + user1 + ' not present in the dataset')

    if user2 not in dataset:
        raise TypeError('User ' + user2 + ' not present in the dataset')

    # 提取两个用户都评分过的电影
    rated_by_both = {}

    for item in dataset[user1]:
        if item in dataset[user2]:
            rated_by_both[item] = 1

    num_ratings = len(rated_by_both) 

    # 都没有评分则是 0 
    if num_ratings == 0:
        return 0

    # 计算相同评分的平方值和
    user1_sum = np.sum([dataset[user1][item] for item in rated_by_both])
    user2_sum = np.sum([dataset[user2][item] for item in rated_by_both])

    # 计算所有相同评分电影的评分的平方和
    user1_squared_sum = np.sum([np.square(dataset[user1][item]) for item in rated_by_both])
    user2_squared_sum = np.sum([np.square(dataset[user2][item]) for item in rated_by_both])

    # 计算数据集乘积之和
    product_sum = np.sum([dataset[user1][item] * dataset[user2][item] for item in rated_by_both])

    # 计算皮尔逊相关度
    Sxy = product_sum - (user1_sum * user2_sum / num_ratings)
    Sxx = user1_squared_sum - np.square(user1_sum) / num_ratings
    Syy = user2_squared_sum - np.square(user2_sum) / num_ratings
    # 分母为0的处理
    if Sxx * Syy == 0:
        return 0

    return Sxy / np.sqrt(Sxx * Syy)

if __name__=='__main__':
    data_file = 'movie_ratings.json'

    with open(data_file, 'r') as f:
        data = json.loads(f.read())

    user1 = 'John Carson'
    user2 = 'Michelle Peterson'

    print "\nPearson score:"
    print pearson_score(data, user1, user2)