GNN的第一个简单案例：Cora分类

最新推荐文章于 2025-05-23 09:50:51 发布

想成为风筝

最新推荐文章于 2025-05-23 09:50:51 发布

阅读量3.4k

点赞数 4

CC 4.0 BY-SA版权

分类专栏： Tensorflow/Pytorch核心知识文章标签： python 深度学习机器翻译

本文链接：https://blog.csdn.net/weixin_50918736/article/details/121462046

Tensorflow/Pytorch核心知识专栏收录该内容

7 篇文章

订阅专栏

该博客介绍了如何使用图卷积网络（GCN）、图注意力网络（GAT）、谱卷积网络（SGConv）、Chebyshev卷积（ChebConv）和图卷积网络（SAGEConv）在Cora数据集上进行节点分类任务。通过加载Cora数据集，定义并训练了各种GNN模型，并在500个训练周期后展示了模型在测试集上的准确率。最终，SGConv模型在Cora数据集上达到了0.794的测试准确率。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

GNN–Cora分类

Cora数据集是GNN中一个经典的数据集，将2708篇论文分为七类：1）基于案例、2）遗传算法、3）神经网络、4）概率方法、5）、强化学习、6）规则学习、7）理论。每一篇论文看作是一个节点，每个节点有1433个特征。

import os

import torch
import torch.nn.functional as F
import torch.nn as nn
from torch_geometric.datasets import Planetoid
import torch_geometric.nn as pyg_nn

#load Cora dataset
def get_data(root_dir='D:\Python\python_dataset\GNN_Dataset\Cora',data_name='Cora'):
    Cora_dataset = Planetoid(root=root_dir,name=data_name)
    print(Cora_dataset)
    return Cora_dataset
Cora_dataset = get_data()
print(Cora_dataset.num_classes,Cora_dataset.num_node_features,Cora_dataset.num_edge_features)
print(Cora_dataset.data)

Cora()
7 1433 0
Data(x=[2708, 1433], edge_index=[2, 10556], y=[2708], train_mask=[2708], val_mask=[2708], test_mask=[2708])

代码中给出GCN、GAT、SGConv、ChebConv、SAGEConv的简单实现

import os
import torch
import torch.nn.functional as F
import torch.nn as nn
from torch_geometric.datasets import Planetoid
import torch_geometric.nn as pyg_nn

#load Cora dataset
def get_data(root_dir='D:\Python\python_dataset\GNN_Dataset\Cora',data_name='Cora'):
    Cora_dataset = Planetoid(root=root_dir,name=data_name)
    print(Cora_dataset)
    return Cora_dataset

#create the Graph cnn model
"""
    2-GATConv
"""
# class GATConv(nn.Module):
#     def __init__(self,in_c,hid_c,out_c):
#         super(GATConv,self).__init__()
#         self.GATConv1 = pyg_nn.GATConv(in_channels=in_c,out_channels=hid_c)
#         self.GATConv2 = pyg_nn.GATConv(in_channels=hid_c, out_channels=hid_c)
#
#     def forward(self,data):
#         x = data.x
#         edge_index = data.edge_index
#         hid = self.GATConv1(x=x,edge_index=edge_index)
#         hid = F.relu(hid)
#
#         out = self.GATConv2(hid,edge_index=edge_index)
#         out = F.log_softmax(out,dim=1)
#
#         return out



"""
    2-SAGE 0.788
"""
# class SAGEConv(nn.Module):
#     def __init__(self,in_c,hid_c,out_c):
#         super(SAGEConv,self).__init__()
#         self.SAGEConv1 = pyg_nn.SAGEConv(in_channels=in_c,out_channels=hid_c)
#         self.SAGEConv2 = pyg_nn.SAGEConv(in_channels=hid_c, out_channels=hid_c)
#
#     def forward(self,data):
#         x = data.x
#         edge_index = data.edge_index
#         hid = self.SAGEConv1(x=x,edge_index=edge_index)
#         hid = F.relu(hid)
#
#         out = self.SAGEConv2(hid,edge_index=edge_index)
#         out = F.log_softmax(out,dim=1)
#
#         return out
"""
    2-SGConv  0.79
"""
class SGConv(nn.Module):
    def __init__(self,in_c,hid_c,out_c):
        super(SGConv,self).__init__()
        self.SGConv1 = pyg_nn.SGConv(in_channels=in_c,out_channels=hid_c)
        self.SGConv2 = pyg_nn.SGConv(in_channels=hid_c, out_channels=hid_c)

    def forward(self,data):
        x = data.x
        edge_index = data.edge_index
        hid = self.SGConv1(x=x,edge_index=edge_index)
        hid = F.relu(hid)

        out = self.SGConv2(hid,edge_index=edge_index)
        out = F.log_softmax(out,dim=1)

        return out

"""
    2-ChebConv
"""
# class ChebConv(nn.Module):
#     def __init__(self,in_c,hid_c,out_c):
#         super(ChebConv,self).__init__()
#
#         self.ChebConv1 = pyg_nn.ChebConv(in_channels=in_c,out_channels=hid_c,K=1)
#         self.ChebConv2 = pyg_nn.ChebConv(in_channels=hid_c,out_channels=out_c,K=1)
#
#     def forward(self,data):
#         x = data.x
#         edge_index = data.edge_index
#         hid = self.ChebConv1(x=x,edge_index=edge_index)
#         hid = F.relu(hid)
#
#         out = self.ChebConv2(hid,edge_index=edge_index)
#         out = F.log_softmax(out,dim=1)
#
#         return out
"""
    2-GCN
"""
# class GraphCNN(nn.Module):
#     def __init__(self, in_c,hid_c,out_c):
#         super(GraphCNN,self).__init__()
#
#         self.conv1 = pyg_nn.GCNConv(in_channels=in_c,out_channels=hid_c)
#         self.conv2 = pyg_nn.GCNConv(in_channels=hid_c,out_channels=out_c)
#
#     def forward(self,data):
#         #data.x,data.edge_index
#         x = data.x       # [N,C]
#         edge_index = data.edge_index  # [2,E]
#         hid = self.conv1(x=x,edge_index=edge_index)  #[N,D]
#         hid = F.relu(hid)
#
#         out = self.conv2(hid,edge_index=edge_index)  # [N,out_c]
#
#         out = F.log_softmax(out,dim=1)
#
#         return out

def main():
    os.environ["CUDA_VISIBLE_DEVICES"] = '0'
    Cora_dataset = get_data()

    #my_net = GATConv(in_c=Cora_dataset.num_node_features, hid_c=100, out_c=Cora_dataset.num_classes)

    #my_net = SAGEConv(in_c=Cora_dataset.num_node_features, hid_c=40, out_c=Cora_dataset.num_classes)
    my_net = SGConv(in_c=Cora_dataset.num_node_features,hid_c=100,out_c=Cora_dataset.num_classes)
    #my_net = ChebConv(in_c=Cora_dataset.num_node_features,hid_c=20,out_c=Cora_dataset.num_classes)
    # my_net = GraphCNN(in_c=Cora_dataset.num_node_features,hid_c=12,out_c=Cora_dataset.num_classes)
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

    my_net = my_net.to(device)
    data = Cora_dataset[0].to(device)

    optimizer = torch.optim.Adam(my_net.parameters(),lr=1e-3)

    #model train
    my_net.train()

    for epoch in range(500):
        optimizer.zero_grad()

        output = my_net(data)
        loss = F.nll_loss(output[data.train_mask],data.y[data.train_mask])
        loss.backward()
        optimizer.step()
        print("Epoch",epoch+1,"Loss",loss.item())

    #model test
    my_net.eval()
    _,prediction = my_net(data).max(dim=1)

    target = data.y
    test_correct = prediction[data.test_mask].eq(target[data.test_mask]).sum().item()
    test_number = data.test_mask.sum().item()

    print("Accuracy of Test Sample:",test_correct/test_number)
if __name__ == '__main__':
    main()

Cora()
Epoch 1 Loss 4.600048542022705
Epoch 2 Loss 4.569146156311035
Epoch 3 Loss 4.535804271697998
Epoch 4 Loss 4.498434543609619
Epoch 5 Loss 4.456351280212402
Epoch 6 Loss 4.409425258636475
Epoch 7 Loss 4.357522964477539
Epoch 8 Loss 4.3007612228393555
Epoch 9 Loss 4.2392096519470215
Epoch 10 Loss 4.172731876373291
Epoch 11 Loss 4.101400375366211
Epoch 12 Loss 4.025243282318115
...............
Epoch 494 Loss 0.004426263272762299
Epoch 495 Loss 0.004407935775816441
Epoch 496 Loss 0.004389731213450432
Epoch 497 Loss 0.004371633753180504
Epoch 498 Loss 0.004353662021458149
Epoch 499 Loss 0.0043357922695577145
Epoch 500 Loss 0.004318032879382372
Accuracy of Test Sample: 0.794