python协同过滤算法（基于用户）的demo

0. 背景

有个协同过滤项目需要处理。从网上查阅资料，发现可用的资料不多。
发现一个简书的博客可以用，原文地址传送门, 数据地址传送门
此处只谈代码，不考虑原理。

1. 经过

本着尝试的心态打开了代码，发现原文代码运行效率极低。原文博主认为是算法本身所导致。
我略经修改，主要利用numpy函数和一些自带函数，优化了原文博主语句，使得运算效果大大提高。

2. 代码


import numpy as np
import pandas as pd
from scipy.spatial.distance import pdist, squareform, cosine
from sklearn.model_selection import train_test_split

npa = np.array
pd.set_option('display.max_rows', 500)
pd.set_option('display.max_columns', 500)
pd.set_option('display.width', 1000)

moviesPath = ".\\ml-latest-small\\movies.csv"
ratingsPath = ".\\ml-latest-small\\ratings.csv"
moviesDF = pd.read_csv(moviesPath, index_col=None)
ratingsDF = pd.read_csv(ratingsPath, index_col=None)

trainRatingsDF, testRatingsDF = train_test_split(ratingsDF, test_size=0.2)
print("total_movie_count:" + str(len(set(ratingsDF['movieId'].values.tolist()))))
print("total_user_count:" + str(len(set(ratingsDF['userId'].values.tolist()))))
print("train_movie_count:" + str(len(set(trainRatingsDF['movieId'].values.tolist()))))
print("test_movie_count:" + str(len(set(testRatingsDF['movieId'].values.tolist()))))
print("train_user_count:" + str(len(set(trainRatingsDF['userId'].values.tolist()))))
print("test_user_count:" + str(len(set(testRatingsDF['userId'].values.tolist()))))

trainRatingsPivotDF = pd.pivot_table(trainRatingsDF[['userId', 'movieId', 'rating']], columns=['movieId'],
                                     index=['userId'], values='rating', fill_value=0)

# enumerate返回穷举序列号与值
# 8981部电影
moviesMap = dict(enumerate(list(trainRatingsPivotDF.columns)))
# 610个用户
usersMap = dict(enumerate(list(trainRatingsPivotDF.index)))
# 矩阵变成list 每一行变成list的一个值 长度为610 每个值大小为8981
ratingValues = trainRatingsPivotDF.values


def calCosineSimilarity(l1, l2):
    return np.sum(l1 * l2) / np.sqrt(np.sum(l1 ** 2) * np.sum(l2 ** 2))


# 根据用户对电影的评分，来判断每个用户间相似度

userSimMatrixArray = pdist(ratingValues, calCosineSimilarity)
userSimMatrix = squareform(userSimMatrixArray)

# 找到与每个用户最相近的前K个用户
userMostSimDict = {i: sorted(enumerate(list(userSimMatrix[i])), key=lambda x: x[1], reverse=True)[:10] for i, v in
                   enumerate(ratingValues)}

# 用这K个用户的喜好中目标用户没有看过的电影进行推荐
userRecommendValues = np.zeros((len(ratingValues), len(ratingValues[0])), dtype=np.float32)  # 610*8981

ind_i, ind_j = np.where(ratingValues == 0)
for i, j in zip(ind_i, ind_j):
    val_array = (ratingValues[user][j] * sim for (user, sim) in userMostSimDict[i])
    userRecommendValues[i, j] = sum(val_array)

userRecommendDict = {i: sorted(enumerate(list(userRecommendValues[i])), key=lambda x: x[1], reverse=True)[:10] for i, v
                     in enumerate(ratingValues)}

# 将一开始的索引转换为原来用户id与电影id
userRecommendList = []
for key, value in userRecommendDict.items():
    user = usersMap[key]
    for (movieId, _) in value:
        userRecommendList.append([user, moviesMap[movieId]])

# 将推荐结果的电影id转换成对应的电影名
recommendDF = pd.DataFrame(userRecommendList, columns=['userId', 'movieId'])
recommendDF = pd.merge(recommendDF, moviesDF[['movieId', 'title']], on='movieId', how='inner')
print(recommendDF.tail(10))