推荐经典算法实现之PMF(python+MovieLen)

最新推荐文章于 2024-07-20 10:30:46 发布

fjssharpsword

最新推荐文章于 2024-07-20 10:30:46 发布

阅读量2.6k

点赞数 3

CC 4.0 BY-SA版权

分类专栏： Algorithm

本文链接：https://blog.youkuaiyun.com/fjssharpsword/article/details/96823817

# -*- Encoding:UTF-8 -*-
'''
@author: Jason.F
@data: 2019.07.18
@function: Implementing PMF
           Dataset: Movielen Dataset(ml-1m) 
           Evaluating: hitradio,ndcg
           https://papers.nips.cc/paper/3208-probabilistic-matrix-factorization.pdf
           Matlab: http://www.utstat.toronto.edu/~rsalakhu/BPMF.html 
@reference: https://github.com/adamzjw/Probabilistic-matrix-factorization-in-Python
'''
import numpy as np
import pandas as pd
from numpy.random import RandomState
import copy
import heapq
import math
from numpy import linalg as LA
import random
#define class PMF
class PMF:
    def __init__(self, num_feat=8, epsilon=1, _lambda=0.1, momentum=0.8, maxepoch=20, num_batches=10, batch_size=1000):
        self.num_feat = num_feat
        self.epsilon = epsilon
        self._lambda = _lambda
        self.momentum = momentum
        self.maxepoch = maxepoch
        self.num_batches = num_batches
        self.batch_size = batch_size
        
        self.w_C = None
        self.w_I = None

        self.err_train = []
        self.err_val = []
        
    def fit(self, train_vec, val_vec):   
        # mean subtraction
        self.mean_inv = np.mean(train_vec[:,2])
        
        pairs_tr = train_vec.shape[0]
        pairs_va = val_vec.shape[0]
        
        # 1-p-i, 2-m-c
        num_inv = int(max(np.amax(train_vec[:,0]), np.amax(val_vec[:,0]))) + 1
        num_com = int(max(np.amax(train_vec[:,1]), np.amax(val_vec[:,1]))) + 1

        incremental = False
        if ((not incremental) or (self.w_C is None)):
            # initialize
            self.epoch = 0
            self.w_C = 0.1 * np.random.randn(num_com, self.num_feat)
            self.w_I = 0.1 * np.random.randn(num_inv, self.num_feat)
            
            self.w_C_inc = np.zeros((num_com, self.num_feat))
            self.w_I_inc = np.zeros((num_inv, se