User Base协同过滤的推荐系统，python实现

同学让帮忙写的，临时写出来的，可能有bug，贴着存档吧，不解释了

# -*- coding: utf-8 -*-
# test on python 2.7.11
import csv
import random
import math
# user base CF


def read_csv(file_name='train.csv'):
    '''
        read csv file to a matrix
    '''

    with open(file_name) as csvfile:
        matrix_reader = csv.reader(csvfile)
        matrix = [[int(x) for x in row] for row in matrix_reader]
    return matrix
#&&&&&&&


def avg(vector):
    '''
    calculate average skipping 0
    '''

    count = sum([int(a > 0) for a in vector])
    if count:
        return sum(vector) / count
    else:
        return 0


def common_items(x, y):
    '''
    x, y are list, return two list
    if x[i] & y[i] > 0: commx.append(x[i]) so as commy
    '''

    commx = [x[i] for i in range(len(x)) if x[i] > 0 and y[i] > 0]
    commy = [y[i] for i in range(len(x)) if x[i] > 0 and y[i] > 0]
    return commx, commy


def pearson(x, y):
    '''
    calculte similarity between x and y with pearson method.
    '''

    avgx = avg(x)
    avgy = avg(y)
    commx, commy = common_items(x, y)
    if len(commx) == 0:
        return 0
    dx = [x-avgx for x in commx]
    dy = [y-avgy for y in commy]
    up = sum([dx[i]*dy[i] for i in range(len(commx))])
    down = math.sqrt(sum([x**2 for x in dx])*sum([y**2 for y in dy]))
    if down:
        return up / down
    else:
        return 0

#&&&&&&&&&&&


def pearson_for_matrix(matrix):
    '''
    calculate similarity between any two user, return a paerson matrix
    '''

    col_length = row_length = len(matrix)
    pearson_matrix = [([0] * col_length) for i in range(row_length)]
    for i in range(row_length):
        for j in range(i+1, col_length):
            pearson_matrix[i][j] = pearson(matrix[i], matrix[j])
            pearson_matrix[j][i] = pearson_matrix[i][j]
    return pearson_matrix

#&&&&&&&


def k_sim_user(sim_matrix, u, k):
    '''
    get the k most similar users with user(u)
    return their id and similarity
    '''

    m = len(sim_matrix)
    tmp_list = [(sim_matrix[u][i], i) for i in range(m)]
    tmp_list.sort(reverse=True)
    tmp_list = [int(x[1]) for x in tmp_list]
    users_id_list = tmp_list[0:k]
    users_value = [sim_matrix[u][x] for x in users_id_list]
    return users_id_list, users_value


def knn_fill_rate_matrix(matrix, sim_matrix, k=3):
    '''
    use knn to fill rating matrix
    '''

    row_length = len(matrix)
    col_length = len(matrix[0])
    for i in range(row_length):
        simi_users, simi_users_value = k_sim_user(sim_matrix, i, k)
        z = sum(simi_users_value)
        user_i_avg = avg(matrix[i])
        for j in range(col_length):
            if matrix[i][j]==0:
                weight = [
                    sim_matrix[i][u]*(matrix[u][j]-avg(matrix[u])) for u in simi_users]
                if sum(simi_users_value) and z!=0:
                    matrix[i][j] = user_i_avg + sum(weight) / z
                else:
                    matrix[i][j] = user_i_avg
                if matrix[i][j]==0:
                    print('again')
    return matrix


def saveMatrix(matrix, file_name='matrix.csv'):
    '''
    write matrix to csv file
    if python3 with open(file_name, 'w', newline='') as csvfile:
    '''

    with open(file_name, 'wb') as csvfile:
        matrix_writer = csv.writer(csvfile)
        for line in matrix:
            # print(line)
            matrix_writer.writerow(line)

# test code below
rate_matrix = read_csv()
sim_matrix = pearson_for_matrix(rate_matrix)
filled = knn_fill_rate_matrix(rate_matrix, sim_matrix, k=3)
saveMatrix(filled, 'result.csv')