# initial helper
helper = Helper()
# get the dict of users in group
g_m_d = helper.gen_group_member_dict(config.user_in_group_path)
# initial dataSet class
dataset = GDataset(config.user_dataset, config.group_dataset,
config.num_negatives)
# get group number~
num_group = len(g_m_d)
num_users, num_items = dataset.num_users, dataset.num_items
# build AGREE model
agree = AGREE(num_users, num_items, num_group, config.embedding_size,
g_m_d, config.drop_ratio)
# config information
print("AGREE at embedding size %d, run Iteration:%d, NDCG and HR at %d" %
(config.embedding_size, config.epoch, config.topK))
# train the model
for epoch in range(config.epoch):
agree.train()
# 开始训练时间
t1 = time()
training(agree, dataset.get_user_dataloader(config.batch_size), epoch,
config, 'user')
training(agree, dataset.get_group_dataloader(config.batch_size), epoch,
config, 'group')
print("user and group training time is: [%.1f s]" % (time() - t1))
# initial helper
helper = Helper()
# get the dict of users in group
g_m_d = helper.gen_group_member_dict(args.user_in_group_path)
# initial dataSet class
dataset = GDataset(args.user_dataset, args.group_dataset,
args.num_negatives, g_m_d)
# get group number
num_group = max(g_m_d.keys()) + 1
num_users, num_items = dataset.num_users, dataset.num_items
# print('num_users',num_users)
# build AGREE model
agree = AGREE(num_users, num_items, num_group, args.embedding_size, g_m_d,
args.drop_ratio).to(device)
# config information
print("AGREE at embedding size %d, run Iteration:%d" %
(args.embedding_size, args.epoch))
history = []
tops = [5, 10, 15, 20]
for epoch in range(args.epoch):
# training
agree.train()
t1 = time()
training(agree, dataset.get_user_dataloader(args.batch_size), epoch,
args, 'user')
training(agree, dataset.get_group_dataloader(128), epoch, args,
'group')
print("user and group training time is: [%.1f s]" % (time() - t1))
t2 = time()
g_m_d = helper.gen_group_member_dict(config.user_in_group_path2)
# 初始化数据类 在这里可以决定使用哪个数据
dataset = GDataset(config.user_dataset2, config.group_dataset2,
config.num_negatives)
# 获取群组的数目、训练集中用户的数目、训练集中物品的数目
num_group = len(g_m_d)
num_users, num_items = dataset.num_users, dataset.num_items
print("num_group is: " + str(num_group))
print("num_users is: " + str(num_users))
print("num_items is: " + str(num_items))
# 训练 AGREE 模型
agree = AGREE(num_users, num_items, num_group, config.num_follow,
config.embedding_size, g_m_d, u_f_d,
config.drop_ratio).to(device)
# 打印配置信息
print("AGREE 的Embedding 维度为: %d, 迭代次数为: %d, NDCG、HR评估选择topK: %d" %
(config.embedding_size, config.epoch, config.topK))
# 训练模型
for epoch in range(config.epoch):
agree.train()
# 开始训练时间
t1 = time()
for _ in range(config.balance):
training(agree, dataset.get_group_dataloader(config.batch_size),
epoch, config, 'group')
return hr, ndcg
if __name__ == '__main__':
config = Config()
helper = Helper()
# get the dict of users in group
g_m_d = helper.gen_group_member_dict(config.user_in_group_path)
dataset = GDataset(config.user_dataset, config.group_dataset, config.num_negatives)
# get group number
num_group = len(g_m_d)
num_users, num_items = dataset.num_users, dataset.num_items
# build AGREE model
agree = AGREE(num_users, num_items, num_group, config.embedding_size, g_m_d, config.drop_ratio, config.GPU_available)
# config information
print("AGREE at embedding size %d, run Iteration:%d, NDCG and HR at %d" %(config.embedding_size, config.epoch, config.topK))
# train the model
if config.GPU_available:
device = torch.device('cuda')
agree = agree.to(device)
for epoch in range(config.epoch):
agree.train()
training(agree, dataset.get_user_dataloader(config.batch_size), epoch, config, 'user')
training(agree, dataset.get_group_dataloader(config.batch_size), epoch, config, 'group')
u_hr, u_ndcg = evaluation(agree, helper, dataset.user_testRatings, dataset.user_testNegatives, config.topK, 'user')
hr, ndcg = evaluation(agree, helper, dataset.group_testRatings, dataset.group_tyestNegatives, config.topK, 'group')