6.1 用户相关性推荐,python

单纯的练手项目,电影的时间没考虑,用户的口味随时间的变化之类的因素都没考虑,单纯的练手

import pandas as pd
import numpy as np
import copy

#这块是自己写的一个personr相关系数的算法,最好别用这个,我在学习,一堆坑
#from scipy.stats import pearsonr
# pearsonr(vector1,vector2)  建议用scipy框架的pearsonr相关系数
def get_pearsonr(b,o):
    b_s = np.sum(b)
    o_s = np.sum(o)
    b_ss = np.sum(np.square(b))
    o_ss = np.sum(np.square(o))
    t_bo = 0
    n = len(b)
    for i in range(len(b)):
        if not np.isnan(o[i]) :
            t_bo += b[i] * o[i]
    denominator = np.sqrt(b_ss - np.square(b_s)/n)*np.sqrt(o_ss - np.square(o_s)/n)
    if denominator:
        return (t_bo - (o_s * b_s)/n) / denominator
    else:
        return 0

def d_pearsonr(data_f):
    n_arr = data_f.values
    base_vector = n_arr[0,:]
    d_values = []
    for o_vector in n_arr:
        if base_vector is o_vector:
            continue
        d_values.append(get_pearsonr(base_vector,o_vector))
    if len(d_values) > 0:
        return pd.Series(d_values,index=data_f.index).sort_values()
    else:
        return -1

# emmm数据来源于https://grouplens.org/datasets/movielens/
# 算法分为根据用户相关系数推荐,和根据电影本身推荐

# 这是简单的根据用户相关系数的推荐算法,随机生成一个userId在1000以内的用户,里面包含了已经有数据的用户和没有数据的用户,
# 有数据的用户根据其它用户评分和该用户评分的相关系数生成推荐
# 没用户数据的推荐评分相对较高的热门电影.
# 毫无难度所以比较适合入门,没区分时间所以并不准,只适合了解推荐算法这个东西

def start_studio(data_f):
    result = d_pearsonr(data_f).index[0:3]
    return result

def newUser_reeccomendation(ratings):
    # ratings.userId = 1
    n_r = copy.deepcopy(ratings)
    n_r.userId = 1
    n_s = n_r.groupby('movieId').sum()
    # print(n_s)
    # print(n_s.sort_values(by='userId',ascending=False).iloc[0:100])
    # print(type(n_s.sort_values(by='userId',ascending=False).iloc[0:100]))
    u_movieId = n_s.sort_values(by='userId',ascending=False).iloc[0:100].index
    r_movieId = n_s.sort_values(by='rating',ascending=False).iloc[0:100].index
    return fetch_movieInfo(list((set(u_movieId) & set(r_movieId)))[0:20])


def fetch_movieInfo(movieIds):
    if len(movieIds) < 20:
        return -1
    movies = pd.read_csv('./ml-latest-small/movies.csv', delimiter=',')
    r_movies = []
    for m_id in movieIds[0:20]:
        name_str = ('电影名称:' + movies[movies['movieId'] == m_id]['title'])
        actor_str = ('演员' + movies[movies['movieId'] == m_id]['genres'])
        r_movies.append(name_str + actor_str)
    return r_movies


def push(userId):
    ratings = pd.read_csv('./ml-latest-small/ratings.csv', delimiter=',')
    all_user = set(ratings.userId)
    real_columns = set(ratings[ratings['userId'] == userId]['movieId'])
    if not len(real_columns):
        return newUser_reeccomendation(ratings)
    real_indexs = [userId]
    for o_userId in all_user:
        if o_userId == userId:
            continue
        u_movie = set(ratings[ratings['userId'] == o_userId]['movieId'])
        if len(u_movie&real_columns) > 26:
            real_indexs.append(o_userId)

    data_arr = np.array(ratings[ratings.userId == userId]['rating']).reshape((1,-1))
    for u_id in real_indexs:
        if u_id == userId:
            continue
        t_arr = []
        for m_id in real_columns:
            um_rank = ratings[ratings.userId == u_id][ratings.movieId == m_id]['rating']
            if not um_rank.empty:
                t_arr.append(um_rank.values[0])
            else:
                t_arr.append(np.NaN)
        data_arr = np.concatenate((data_arr,np.array(t_arr).reshape((1,-1))),axis=0)
    s_userIds = start_studio(pd.DataFrame(data=data_arr,index=real_indexs,columns=real_columns))
    allreccomendation = []
    for r_userId in s_userIds:
        allreccomendation.extend(ratings[ratings['userId'] == r_userId].sort_values(by='rating')['movieId'])
    real_movie = list(set(allreccomendation) - set(real_columns))

    return fetch_movieInfo(real_movie)

print(push(345))