单层GNN完成cora数据集节点分类任务

悠悠 2023-02-26 10:24 20阅读 0赞

import torch
    import torch.nn as nn
    import torch.nn.functional as F
    
    import numpy as np
    import matplotlib.pyplot as plt
    #使用上一篇博客中用到数据集处理函数
    from coraDatasetsProcess import main
    node_nums,feature_dims,label_list,feat_Matrix,degree_list,cites,X_Node,X_Neis=main()
    
    #定义图卷积模型
    class gnnModel(torch.nn.Module):
        def __init__(self):
            super(gnnModel, self).__init__()
            self.lin1=nn.Linear(in_features=1433,out_features=7)
        def forward(self,x,dig_list,A):
            print(A.shape)
            N=len(x)
            I_list=[1 for i in range(N)]
            I=np.diag(I_list)#单位矩阵
            A=A+I+I#添加自循环
            diags=np.diag(dig_list**(-0.5))
            pre=np.dot(np.dot(np.dot(diags,A),diags),x)
            pre = pre.astype(np.float32)
            pre=torch.from_numpy(pre)#从numpy的ndarray格式转化为tensor张量
            x=self.lin1(pre)
            return x
    #数据预处理
    def processData():
        A=np.zeros((node_nums,node_nums))
        for i in range(len(X_Node)):
            source=X_Node[i].item()
            target=X_Neis[i].item()
            A[source][target]=1#构造邻接矩阵
        return A
    
    #实例化模型并训练
    def modelStart(A):
        net=gnnModel()#实例化图神经网络模型
        net.train()#训练模式
    
        #优化器
        optimizer=torch.optim.Adam(net.parameters(),lr=0.01)
        #损失函数
        loss_function=nn.CrossEntropyLoss()
        out=0
        lossList=[]
        for epoch in range(200):
            optimizer.zero_grad()
            out=net(feat_Matrix[:500],degree_list[:500],A[:500,:500])
            loss=loss_function(out,label_list[:500])
            lossList.append(loss)
            print("epoch:",epoch," loss:",loss)
            loss.backward()
            optimizer.step()
    
        with torch.no_grad():
            out = net(feat_Matrix, degree_list, A)
            max_value,max_index=torch.max(out.data,1)
            correct=max_index.eq(label_list).sum().item()
            print("the accuracy of node classification is:",correct/len(label_list))
        return lossList
    
    #plot画loss曲线
    def plotCora(lossList):
        N=len(lossList)
        y=lossList
        x=[i for i in range(N)]
        p=plt.plot(x,y)
        plt.xlabel("iterations")
        plt.ylabel("loss")
        plt.title("the loss curve for node classification")
        plt.show(p)
    def mainP():
        A=processData()
        lossList=modelStart(A)
        plotCora(lossList)
    if __name__ == '__main__':
        mainP()

![20200715112122338.png][] ，公式如图所示

使用前500个节点数据集做训练，用整个数据集做测试，最终经过200次迭代后，节点分类的准确率是75%

loss曲线如下：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2p1c3Rfc29fc29fZm5j_size_16_color_FFFFFF_t_70][]

[20200715112122338.png]: /images/20230209/ac7d0a24b1d246b0bc31fd7dbe80ef2e.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2p1c3Rfc29fc29fZm5j_size_16_color_FFFFFF_t_70]: /images/20230209/0b71c91f15ad4f6bb15ea96171c8a4af.png