LSIR/main.py at main · fs302/LSIR · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
import numpy as np
from parse import set_params
from dataloader import Loader
import warnings
import utils
from model import Inac_rec
import torch
import torch.nn as nn
from evaluation import Evaluation
from datetime import datetime


warnings.filterwarnings('ignore')
args = set_params()
if torch.cuda.is_available():
    args.device = torch.device("cuda")
else:
    args.device = torch.device("cpu")

class TrainFlow:
    def __init__(self, args):
        self.args = args
        self.own_str = args.dataset
        print(self.own_str)
        self.dataset = Loader(args)
        self.args.user_feat_dim = self.dataset.user_feat.shape[1]
        self.args.item_feat_dim = self.dataset.item_feat.shape[1]
        self.val_data = self.dataset.valDict
        self.test_data = self.dataset.testDict
        self.add_prob, self.dele_prob = utils.get_add_dele_prob(args, self.dataset.users_D)
        if args.edge_emb_flag:
            self.emb_initial_matrix = nn.Embedding(args.feat_dim, args.hidden_dim).cuda()
        self.model = Inac_rec(args, self.dataset.n_user, self.dataset.m_item, self.dataset.cluster_map).to(args.device)
        self.opt_encoder = torch.optim.Adam(self.model.encoder.parameters(), lr=args.lr_encoder)#, weight_decay=args.weight_decay)
        self.opt_gsl = torch.optim.Adam(self.model.GSL4uu.parameters(), lr=args.lr_gsl)#, weight_decay=args.weight_decay)
        # self.scheduler = torch.optim.lr_scheduler.ExponentialLR(self.optimizer, 0.999)

        self.eva_val = Evaluation(self.dataset.n_user, self.dataset.m_item, self.dataset.val_link, self.dataset.index_inact, self.args.eva_neg_num, \
            "./dataset/"+self.args.dataset+"/val_neg_links.npy")
        self.eva_test = Evaluation(self.dataset.n_user, self.dataset.m_item, self.dataset.test_link, self.dataset.index_inact, self.args.eva_neg_num, \
            "./dataset/"+self.args.dataset+"/test_neg_links.npy")

    def eva(self, data, eva, test_flag=False):
        self.model.eval()
        all_users = np.arange(self.dataset.n_user)
        all_user_emb, all_item_emb = self.model.get_emb(self.dataset.UI_Graph, self.dataset.user_feat, self.dataset.item_feat, all_users)
        all_user_emb = all_user_emb.data.cpu().numpy()
        all_item_emb = all_item_emb.data.cpu().numpy()
        results_10, results_inac_10 = eva.get_result(10, all_user_emb, all_item_emb)
        results_20, result_inacs_20 = eva.get_result(20, all_user_emb, all_item_emb)
        if test_flag:
            f = open(self.own_str+"_10.txt", "a")
            f.write(str(results_10[0])+"\t"+str(results_10[1])+"\t"+str(results_10[2])+"\t"+\
                str(results_inac_10[0])+"\t"+str(results_inac_10[1])+"\t"+str(results_inac_10[2])+"\n")
            f.close()
            f = open(self.own_str+"_20.txt", "a")
            f.write(str(results_20[0])+"\t"+str(results_20[1])+"\t"+str(results_20[2])+"\t"+\
                str(results_inac_20[0])+"\t"+str(results_inac_20[1])+"\t"+str(results_inac_20[2])+"\n")
            f.close()

    def train(self):
        self.model.uu_graph = self.dataset.UU_Graph
        for epoch in range(args.tot_epochs):
            ## Train encoder
            print("Train encoder")
            for inner_encoder in range(args.encoder_epochs):
                S_act, S_inact = utils.generate_train_data(self.dataset)
                act_batch_size = int(S_act.shape[0] / args.train_iters)
                inact_batch_size = int(S_inact.shape[0] / args.train_iters)
                act_temp = 0
                inact_temp = 0
                tot_loss = []
                print("New epoch!")
                terminal = True
                while terminal:
                    a=datetime.now()
                    self.model.train()
                    self.opt_encoder.zero_grad()
                    if args.train_iters == 1:
                        act_curr = S_act
                        inact_curr = S_inact
                        terminal = False
                    else:
                        if act_temp + act_batch_size < S_act.shape[0]:
                            act_curr = S_act[act_temp : act_temp + act_batch_size]
                            inact_curr = S_inact[inact_temp : inact_temp + inact_batch_size]
                            act_temp += act_batch_size
                            inact_temp += inact_batch_size
                        else:
                            act_curr = S_act[act_temp : ]
                            inact_curr = S_inact[inact_temp : ]
                            terminal = False
                    batch_user_pos_neg = np.vstack([act_curr, inact_curr])
                    batch_act_user = np.sort(list(set(act_curr[:, 0])))
                    batch_inact_user = np.sort(list(set(inact_curr[:, 0])))
                    batch_user = np.sort(list(set(batch_user_pos_neg[:, 0])))

                    loss = self.model.train_encoder(batch_user_pos_neg, self.dataset.UI_Graph, self.dataset.user_feat, self.dataset.item_feat)
                    loss.backward()
                    self.opt_encoder.step()
                    tot_loss.append(loss.cpu().data.numpy())
                b=datetime.now()
                print("tot_epochs ", epoch, "\tencoder_epochs ", inner_encoder, "\tLoss ", np.array(tot_loss).mean(), "\tseconds ", (b-a).seconds)

            ## Get user item emb
            print("Get user item emb")
            all_users = np.arange(self.dataset.n_user)
            user_emb, item_emb, user_emb_ego = self.model.get_emb(self.dataset.UI_Graph, self.dataset.user_feat, self.dataset.item_feat, all_users, temp_flag=True)
            ## Train gsl
            print("Train gsl")
            for inner_gsl in range(args.gsl_epochs):
                S_act, S_inact = utils.generate_train_data(self.dataset)
                act_batch_size = int(S_act.shape[0] / args.train_iters)
                inact_batch_size = int(S_inact.shape[0] / args.train_iters)
                act_temp = 0
                inact_temp = 0
                tot_loss = []
                print("New epoch!")
                terminal = True
                while terminal:
                    a=datetime.now()
                    self.model.train()
                    self.opt_gsl.zero_grad()
                    if args.train_iters == 1:
                        act_curr = S_act
                        inact_curr = S_inact
                        terminal = False
                    else:
                        if act_temp + act_batch_size < S_act.shape[0]:
                            act_curr = S_act[act_temp : act_temp + act_batch_size]
                            inact_curr = S_inact[inact_temp : inact_temp + inact_batch_size]
                            act_temp += act_batch_size
                            inact_temp += inact_batch_size
                        else:
                            act_curr = S_act[act_temp : ]
                            inact_curr = S_inact[inact_temp : ]
                            terminal = False
                    batch_user_pos_neg = np.vstack([act_curr, inact_curr])
                    batch_act_user = np.sort(list(set(act_curr[:, 0])))
                    batch_inact_user = np.sort(list(set(inact_curr[:, 0])))
                    batch_user = np.sort(list(set(batch_user_pos_neg[:, 0])))
                    loss = self.model.train_graph_generator(user_emb_ego, batch_user_pos_neg, batch_act_user, batch_inact_user, self.dataset.uu_dict, user_emb, item_emb, add_prob=self.add_prob, dele_prob=self.dele_prob)
                    loss.backward()
                    self.opt_gsl.step()
                    tot_loss.append(loss.cpu().data.numpy())
                b=datetime.now()
                print("tot_epochs ", epoch, "\tgsl_epochs ", inner_gsl, "\tLoss ", np.array(tot_loss).mean(), "\tseconds ", (b-a).seconds)
            ## Get whole structure
            with torch.no_grad():
                print("Get whole structure")
                all_users = torch.arange(self.dataset.n_user)
                final_dele_indices, final_dele_sim, final_add_indices, final_add_sim = self.model.get_stru(user_emb_ego, all_users, user_emb, self.dataset.uu_dict, self.add_prob, self.dele_prob)

                self.model.uu_graph =self.model.get_whole_stru(all_users, final_dele_indices, final_dele_sim, final_add_indices, final_add_sim, self.dataset.n_user)
                print("Get whole structure finish")


            print("Evaluation")
            self.eva(self.test_data, self.eva_test, test_flag=True)
            # torch.save(self.model.state_dict(), self.own_str+'.pkl')

        ## test ##
        self.eva(self.test_data, self.eva_test, test_flag=True)

if __name__ == '__main__':
    train = TrainFlow(args)
    train.train()