LLMGuard/train_edges.py at main · lvXiangwei/LLMGuard · 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
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import TensorDataset, DataLoader

from sklearn.model_selection import train_test_split

import argparse
import numpy as np

# from utils import time_logger

class EdgePredictor(nn.Module):
    def __init__(self, embedding_dim, hidden_dim):
        super(EdgePredictor, self).__init__()
        self.fc1 = nn.Linear(embedding_dim * 2, hidden_dim)
        self.relu = nn.ReLU()
        self.fc2 = nn.Linear(hidden_dim, 1)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        x = self.fc1(x)
        x = self.relu(x)
        x = self.fc2(x)
        x = self.sigmoid(x)

        return x

from utils.util import get_text_data
def preprocess(args):
    ood_data, id_data= get_text_data(dataname='cora')

    positive_edges = torch.load(f'./cora/{args.edge_type}_positive_edges.pt')
    negative_edges = torch.load(f'./cora/{args.edge_type}_negative_edges.pt')

    ood_embs = np.load('./cora/cora_ood_embs.npy')
    id_embs = np.load('./cora/cora_id_embs.npy')
    data_x = []
    data_y = []
    for i in range(len(positive_edges)):
        for j in positive_edges[i]:
            # paper_ood = ood_data[i]
            # paper_id = id_data[j]
            # import ipdb; ipdb.set_trace()
            data_x.append(torch.cat([
                torch.from_numpy(ood_embs[i]), torch.from_numpy(id_embs[j])]))
            data_y.append(1)

    for i in range(len(negative_edges)):
        for j in negative_edges[i]:
            # paper_ood = ood_data[i]
            # paper_id = id_data[j]
            data_x.append(torch.cat([
                torch.from_numpy(ood_embs[i]), torch.from_numpy(id_embs[j])]))
            data_y.append(0)
    # import ipdb; ipdb.set_trace()
    train_x, test_x, train_y, test_y = train_test_split(data_x, data_y, test_size=0.2, random_state=args.seed)
    train_x = torch.stack(train_x).to(torch.float32)  # 转换为 FloatTensor
    train_y = torch.FloatTensor(train_y)

    test_x = torch.stack(test_x).to(torch.float32)  # 转换为 FloatTensor
    test_y = torch.FloatTensor(test_y)
    return train_x, train_y, test_x, test_y

# @time_logger
def train(args):
    device = args.device
    model = EdgePredictor(args.embedding_dim, args.hidden_dim).to(device)

    train_x, train_y, test_x, test_y = preprocess(args)

    train_dataset = TensorDataset(train_x, train_y)
    train_loader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True)

    test_dataset = TensorDataset(test_x, test_y)
    test_loader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False)


    criterion = nn.BCELoss()
    optimizer = optim.Adam(model.parameters(), lr=args.lr)

    best_accuracy = 0
    epochs_no_improve = 0

    num_epochs = args.epochs
    for epoch in range(num_epochs):
        for i, (node_embedding_pairs, batch_labels) in enumerate(train_loader):
            node_embedding_pairs, batch_labels = node_embedding_pairs.to(device), batch_labels.to(device)

            outputs = model(node_embedding_pairs)
            outputs = outputs.squeeze()
            loss = criterion(outputs, batch_labels)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

        model.eval()
        total = 0
        correct = 0
        with torch.no_grad():
            for node_embedding_pairs, batch_labels in test_loader:
                node_embedding_pairs, batch_labels = node_embedding_pairs.to(device), batch_labels.to(device)
                outputs = model(node_embedding_pairs)
                predicted = (outputs.squeeze() > 0.5).float()
                total += batch_labels.size(0)
                correct += (predicted == batch_labels).sum().item()

        test_accuracy = 100 * correct / total

        print(f'Epoch {epoch+1}/{num_epochs}, Train Loss: {loss.item()}, Test Accuracy: {test_accuracy}%')

        if test_accuracy > best_accuracy:
            best_accuracy = test_accuracy
            epochs_no_improve = 0
            torch.save(model.state_dict(), 'cora/' + args.dataset + f'_{args.edge_type}.pth')
        else:
            epochs_no_improve += 1

        if epochs_no_improve >= args.early_stop:
            print(f'Early stopping triggered after {epoch + 1} epochs')
            break


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--device', type=str, default='cuda')
    parser.add_argument('--dataset', type=str, default='cora')
    parser.add_argument('--seed', type=int, default=42)
    parser.add_argument('--embedding_dim', type=int, default=4096)
    parser.add_argument('--hidden_dim', type=int, default=1024)
    parser.add_argument('--batch_size', type=int, default=512)
    parser.add_argument('--lr', type=float, default=0.00005)
    parser.add_argument('--epochs', type=int, default=200)
    parser.add_argument('--early_stop', type=int, default=20)

    parser.add_argument('--edge_type', type=str, default='ood_id')
    args = parser.parse_args()

    train(args)