07_LFM--梯度下降法--实现基于模型的协同过滤

LFM–梯度下降法–实现基于模型的协同过滤

0.引入依赖

import numpy as np # 数值计算、矩阵运算、向量运算
import pandas as pd # 数值分析、科学计算

1.数据准备

# 定义评分矩阵 R
R = np.array([[4, 0, 2, 0, 1],
              [0, 2, 3, 0, 0],
              [1, 0, 2, 4, 0],
              [5, 0, 0, 3, 1],
              [0, 0, 1, 5, 1],
              [0, 3, 2, 4, 1],
             ])
# R.shape # (6, 5)
# R.shape[0] # 6
# R.shape[1] # 5
# len(R) # 6
# len(R[0]) # 5

2.算法的实现

"""
@输入参数：
R：M*N 的评分矩阵
K：隐特征向量维度
max_iter: 最大迭代次数
alpha：步长
lamda：正则化系数

@输出：
分解之后的 P，Q
P：初始化用户特征矩阵 M*K
Q：初始化物品特征矩阵 N*K，Q 的转置是 K*N
"""

# 给定超参数
K = 5
max_iter = 5000
alpha = 0.0002
lamda = 0.004

# 核心算法
def LMF_grad_desc(R, K=2, max_iter=1000, alpha=0.0001, lamda=0.002):
    # 定义基本维度参数
    M = len(R)
    N = len(R[0])
    
    # P、Q 的初始值随机生成
    P = np.random.rand(M, K)
    Q = np.random.rand(N, K)
    Q = Q.T
    
    # 开始迭代
    for steps in range(max_iter):
        # 对所有的用户 u，物品 i 做遍历，然后对对应的特征向量 Pu、Qi 做梯度下降
        for u in range(M):
            for i in range(N):
                # 对于每一个大于 0 的评分，求出预测评分误差 e_ui
                if R[u][i] > 0:
                    e_ui = np.dot(P[u,:], Q[:,i]) - R[u][i]
                    # 代入公式，按照梯度下降算法更新当前的 Pu、Qi
                    for k in range(K):
                        P[u][k] = P[u][k] - alpha * (2 * e_ui * Q[k][i] + 2 * lamda * P[u][k])
                        Q[k][i] = Q[k][i] - alpha * (2 * e_ui * P[u][k] + 2 * lamda * Q[k][i])
        
        # u,i 遍历完成，所有的特征向量更新完成，可以得到 P、Q，可以计算预测评分矩阵
        predR = np.dot(P, Q)
        
        # 计算当前损失函数（所有的预测误差平方后求和）
        cost = 0
        for u in range(M):
            for i in range(N):
                # 对于每一个大于 0 的评分，求出预测评分误差后，将所有的预测误差平方后求和
                if R[u][i] > 0:
                    cost += (np.dot(P[u,:], Q[:,i]) - R[u][i]) ** 2
                    # 加上正则化项
                    for k in range(K):
                        cost += lamda * (P[u][k] ** 2 + Q[k][i] ** 2)
        if cost < 0.0001:
            # 当前损失函数小于给定的值，退出迭代
            break
    
    return P, Q.T, cost

3.测试

P, Q, cost = LMF_grad_desc(R, K, max_iter, alpha, lamda)

print(P)
print(Q)
print(cost)

predR = P.dot(Q.T)

print(R)
predR

当 K = 2 时，输出结果如下：

[[1.44372596 1.29573962]
 [1.82185633 0.0158696 ]
 [1.5331521  0.16327061]
 [0.31364667 1.9008297 ]
 [1.03622742 2.03603634]
 [1.34107967 0.93406796]]
[[ 0.4501051   2.55477489]
 [ 1.18869845  1.20910294]
 [ 1.54255106 -0.23514326]
 [ 2.33556583  1.21026575]
 [ 0.43753164  0.34555928]]
1.0432768290554293
[[4 0 2 0 1]
 [0 2 3 0 0]
 [1 0 2 4 0]
 [5 0 0 3 1]
 [0 0 1 5 1]
 [0 3 2 4 1]]
 
array([[3.96015147, 3.2828374 , 1.92233657, 4.9401063 , 1.07943065],
       [0.86057008, 2.18482578, 2.80657478, 4.27427181, 0.80260368],
       [1.10719924, 2.0198665 , 2.32657341, 3.77837848, 0.72722223],
       [4.99736596, 2.6711301 , 0.03684871, 3.03305153, 0.79407969],
       [5.66802576, 3.69353946, 1.11967348, 4.8843224 , 1.15695354],
       [2.98996017, 2.72352365, 1.84904408, 4.2626503 , 0.90954065]])

当 K = 5 时，输出结果如下：

[[ 0.77991893  0.95803701  0.75945903  0.74581653  0.58070622]
 [ 1.51777367  0.66949331  0.89818609  0.23566984  0.56583223]
 [ 0.03567022  0.58391558  1.42477223  0.87262652 -0.52553017]
 [ 1.24101793  0.86257736  0.73772417  0.18181617  0.97014545]
 [ 0.58789616  0.53522492  0.48830352  1.80622908  0.81202167]
 [ 1.08640318  0.87660384  0.68935314  0.84506882  0.92284071]]
[[ 1.64469428  1.10535565  0.56686066  0.38656745  1.56519511]
 [ 0.61680687  0.57188343  0.49729111  0.9623455   0.43969708]
 [ 0.99260822  0.6007452   1.14768173 -0.16998497 -0.14094479]
 [ 0.47070988  0.85347655  1.43546859  1.8185161   0.29759968]
 [ 0.07923314  0.49412497  0.53285806  0.23753882 -0.05146021]]
0.7478305665280703
[[4 0 2 0 1]
 [0 2 3 0 0]
 [1 0 2 4 0]
 [5 0 0 3 1]
 [0 0 1 5 1]
 [0 3 2 4 1]]

array([[3.9694342 , 2.37968507, 2.01268221, 3.8040546 , 1.08714641],
       [4.72218838, 2.2412959 , 2.81976984, 3.17210672, 0.95653992],
       [1.02652007, 1.67315396, 1.94711343, 3.99085212, 1.28488146],
       [5.0014878 , 2.22716585, 2.42906339, 2.99867943, 0.91091753],
       [3.80452512, 3.00679363, 1.04401937, 4.96078887, 0.95850804],
       [4.91762916, 2.73324389, 2.1224277 , 4.06049468, 1.03980543]])