【代码解析（2）】Secure Federated Matrix Factorization

DistributeMF_Part.py

import time
import copy
import numpy as np


from load_data import ratings_dict, item_id_list, user_id_list
from shared_parameter import *


"""
Part & Full
full: 用户上传所有项目的梯度,如果一个用户没有对一个item评价,梯度为零
part: 用户只上传已评项目的梯度
part会泄露已评项目的信息但是会有更高的计算效率
full不会泄露信息但是需要更多的计算时间

"""


def user_update(single_user_vector, user_rating_list, item_vector):

    gradient = {}  # 这里full和part确实不一样

    for item_id, rate, _ in user_rating_list:
        error = rate - np.dot(single_user_vector, item_vector[item_id])
        single_user_vector = single_user_vector - lr * (-2 * error * item_vector[item_id] + 2 * reg_u * single_user_vector)
        gradient[item_id] = error * single_user_vector
        '''
            gradient只包含用户评价的item,其他的不考虑为零
        '''
    return single_user_vector, gradient


def mse():
    loss = []
    for i in range(len(user_id_list)):
        for r in range(len(ratings_dict[user_id_list[i]])):
            item_id, rate, _ = ratings_dict[user_id_list[i]][r]
            error = (rate - np.dot(user_vector[i], item_vector[item_id])) ** 2
            loss.append(error)
    return np.mean(loss)


if __name__ == '__main__':
    
    # Init process
    user_vector = np.random.normal(size=[len(user_id_list), hidden_dim])
    item_vector = np.random.normal(size=[len(item_id_list), hidden_dim])
    '''
        i = 0
        for vector in item_vector:
            for e in vector:
                print(e)
                i = i + 1
                print('===============')
            print(i)
        print('================')
        print(len(item_vector[0]))第一行有多少个元素(列)
    '''

    start_time = time.time()

    for iteration in range(max_iteration):

        print('###################')
        t = time.time()

        # Step 2 User updates
        gradient_from_user = []
        for i in range(len(user_id_list)):
            user_vector[i], gradient = user_update(user_vector[i], ratings_dict[user_id_list[i]], item_vector)
            gradient_from_user.append(gradient)

        # Step 3 Server update
        tmp_item_vector = copy.deepcopy(item_vector)
        for g in gradient_from_user:
            for item_id in g:
                '''
                    full:
                        item_vector = item_vector - lr * (-2 * g + 2 * reg_u * item_vector)
                    part:
                        只是用户已评的项目,不是整个矩阵了
                '''
                item_vector[item_id] = item_vector[item_id] - lr * (-2 * g[item_id] + 2 * reg_v * item_vector[item_id])

        if np.mean(np.abs(item_vector - tmp_item_vector)) < 1e-4:
            print('Converged')
            break

        print('Time', time.time() - t, 's')
        print('loss', mse())

    print('Converged using', time.time() - start_time)