UnlearnRec/make_noise_dataset.py at main · HKUDS/UnlearnRec · 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
import torch as t
import Utils.time_logger as logger
from Utils.time_logger import log
from params import args
# from model import LightGCN
from data_handler import DataHandler
import numpy as np
import pickle
import os
import setproctitle
from scipy.sparse import coo_matrix
import random
import torch_sparse as ts

os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
args.adversarial_attack = False

t.manual_seed(args.seed)
t.cuda.manual_seed_all(args.seed)
t.backends.cudnn.deterministic = True
np.random.seed(args.seed)
random.seed(args.seed)


handler = DataHandler()
handler.load_data(drop_rate=0.0, adv_attack=False)

# save_path = '../../datasets/ml-1m/adv_rd_mat.pkl'

# save_path = '../../datasets/sparse_yelp/adv_lightgcn_mat.pkl'
# save_path = '../../datasets/sparse_gowalla/adv_lightgcn_mat.pkl'
# save_path = '../../datasets/sparse_amazon/adv_lightgcn_mat.pkl'
save_path = '../../datasets/yelp2018/adv_simgcl0.6_mat.pkl'


# model_path = '../outModels/yelp/retrain/retrain_yelp_lightgcn_reg1e-6_lr1e-3_b4096_ep300_dim128_ly3.mod'
# model_path = '../outModels/gowalla/retrain/retrain_gowalla_lightgcn_reg1e-7_lr1e-3_b4096_ep300_dim128_ly3.mod'
# model_path = '../outModels/amazon/retrain/retrain_amazon_lightgcn_reg1e-8_lr1e-3_b4096_ep300_dim128_ly3.mod'
model_path = '../outModels/yelp2018/retrain/retrain_yelp2018_simgcl_reg1e-6_ssl1e-2_esp2e-1_t1e-1_v1_lr1e-3_b4096_ep300_dim128_ly3.mod'


def find_least_related_edges(model, handler, save_path):
    model.is_training = False
    usr_embeds, itm_embeds = model.forward(handler.torch_adj)
    least_related_edges = [[], []]
    for i in range(args.user):
        usr_embed = usr_embeds[i]
        preds = usr_embed @ itm_embeds.T
        j = t.argmin(preds).item()
        least_related_edges[0].append(i)
        least_related_edges[1].append(j)

    for j in range(args.item):
        itm_embed = itm_embeds[j]
        preds = itm_embed @ usr_embeds.T
        i = t.argmin(preds).item()
        least_related_edges.append((i, j))
        least_related_edges[0].append(i)
        least_related_edges[1].append(j)
    # rows = handler.trn_mat.row
    # cols = handler.trn_mat.col
    rows = handler.ori_trn_mat.row
    cols = handler.ori_trn_mat.col


    log('Original number of edges %d' % len(rows))
    rows = np.concatenate([rows, least_related_edges[0]])
    cols = np.concatenate([cols, least_related_edges[1]])
    vals = np.ones_like(rows)
    log('New number of edges %d' % len(rows))
    adv_adj = coo_matrix((vals, (rows, cols)), shape=[args.user, args.item])
    with open(save_path, 'wb') as fs:
        pickle.dump((adv_adj, least_related_edges), fs)


def find_least_related_edges_smp(model, handler, save_path, ratio=0.6):
    model.is_training = False
    usr_embeds, itm_embeds = model.forward(handler.torch_adj)
    least_related_edges = [[], []]

    rd_list = np.random.permutation(args.user)
    user_set = rd_list[:int(ratio*args.user)]

    rd_list = np.random.permutation(args.item)
    item_set = rd_list[:int(ratio*args.item)]


    for i in user_set:
        usr_embed = usr_embeds[i]
        preds = usr_embed @ itm_embeds.T
        j = t.argmin(preds).item()
        least_related_edges[0].append(i)
        least_related_edges[1].append(j)

    for j in item_set:
        itm_embed = itm_embeds[j]
        preds = itm_embed @ usr_embeds.T
        i = t.argmin(preds).item()
        least_related_edges.append((i, j))
        least_related_edges[0].append(i)
        least_related_edges[1].append(j)
    # rows = handler.trn_mat.row
    # cols = handler.trn_mat.col
    rows = handler.ori_trn_mat.row
    cols = handler.ori_trn_mat.col


    log('Original number of edges %d' % len(rows))
    rows = np.concatenate([rows, least_related_edges[0]])
    cols = np.concatenate([cols, least_related_edges[1]])
    vals = np.ones_like(rows)
    log('New number of edges %d' % len(rows))
    adv_adj = coo_matrix((vals, (rows, cols)), shape=[args.user, args.item])
    with open(save_path, 'wb') as fs:
        pickle.dump((adv_adj, least_related_edges), fs)

def load_model(load_model):
    ckp = t.load(load_model)
    model = ckp['model']
    return model


model = load_model(model_path)
find_least_related_edges_smp(model, handler, save_path, 0.6)
# find_least_related_edges(model, handler, save_path)