Graph Convolution Network图卷积网络（一）训练运行与代码概览

我就是我 2022-04-24 06:24 338阅读 1赞

**背景：**看懂并运行Graph Convolution Network的pytorch代码。

**代码地址**：[https://github.com/tkipf/pygcn][https_github.com_tkipf_pygcn]

**论文地址**：[https://arxiv.org/abs/1609.02907 ][https_arxiv.org_abs_1609.02907]Semi-Supervised Classification with Graph Convolutional Networks，ICLR 2017

**目录**

一、运行

1.1 搭建环境

1.2 运行

二、数据集概览

2.1 总览

2.2 content file

2.3 cites file

三、代码概览

2.1 train.py

2.2 models.py

2.3 layers.py

2.4 utils.py

--------------------

# 一、运行 #

## 1.1 搭建环境 ##

搭建python、torch及显卡环境，

[CentOS 6.3安装anaconda并配置pytorch与cuda ][CentOS 6.3_anaconda_pytorch_cuda]

## 1.2 运行 ##

运行很简单，直接环境搭建好直接就可运行，并且没有复杂的命令行。

`python setup.py install`

`python train.py`

`迭代200次，很快即可运行完毕。`

Epoch: 0198 loss_train: 0.4363 acc_train: 0.9571 loss_val: 0.7047 acc_val: 0.8233 time: 0.0076s
    Epoch: 0199 loss_train: 0.3899 acc_train: 0.9500 loss_val: 0.7030 acc_val: 0.8233 time: 0.0076s
    Epoch: 0200 loss_train: 0.4644 acc_train: 0.9071 loss_val: 0.7015 acc_val: 0.8200 time: 0.0076s
    Optimization Finished!
    Total time elapsed: 2.2366s
    Test set results: loss= 0.7510 accuracy= 0.8320

# 二、数据集概览 #

core数据集[  https://github.com/tkipf/pygcn/tree/master/data/cora][https_github.com_tkipf_pygcn_tree_master_data_cora]

## 2.1 总览 ##

该数据集就是图结构的数据集。机器学习的paper，表示用paper ID，共2708个paper。包含了下面这些类：

Case_Based
    		Genetic_Algorithms
    		Neural_Networks
    		Probabilistic_Methods
    		Reinforcement_Learning
    		Rule_Learning
    		Theory

共2708个paper，7个类，一共有1433个关键词，关键词频次小于10会被删掉。

## 2.2 content file ##

包含下面：

<paper_id> <word_attributes>+ <class_label>

最开始为paper\_id，后面为每一个单词是否出现，最后为类的标签。

第一列为paper ID，后面几列为每个单词出现与否，用0与1表示，最后一列为类别标签。

![20190418104517245.png][]  ![20190418104542994.png][]

## 2.3 cites file ##

表示论文之间的引用关系（由此看来这个图是有向图）

前面为被引论文的ID，后面为引用前面的论文的ID

887	6215
    887	64519
    887	87363
    887	976334
    906	1103979
    906	1105344
    906	1114352
    906	1136397

# 三、代码概览 #

看到论文中有大量公式及推导，但是实际代码量很少。相当于只有四个代码有实际语义层面的信息。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70][]

## 2.1 train.py ##

包含模型训练等等的信息。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 1][]

加载参数

args = parser.parse_args()
    args.cuda = not args.no_cuda and torch.cuda.is_available()

生成随机种子

np.random.seed(args.seed)
    torch.manual_seed(args.seed)
    if args.cuda:
        torch.cuda.manual_seed(args.seed)

加载数据

# Load data
    adj, features, labels, idx_train, idx_val, idx_test = load_data()

定义模型与optimizer

# Model and optimizer
    model = GCN(nfeat=features.shape[1],
                nhid=args.hidden,
                nclass=labels.max().item() + 1,
                dropout=args.dropout)
    optimizer = optim.Adam(model.parameters(),
                           lr=args.lr, weight_decay=args.weight_decay)

数据写入cuda，便于后续加速

if args.cuda:
        model.cuda()
        features = features.cuda()
        adj = adj.cuda()
        labels = labels.cuda()
        idx_train = idx_train.cuda()
        idx_val = idx_val.cuda()
        idx_test = idx_test.cuda()

定义训练函数

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 2][]

def train(epoch):
        t = time.time()
        model.train()
        optimizer.zero_grad()
        output = model(features, adj)
        loss_train = F.nll_loss(output[idx_train], labels[idx_train])
        acc_train = accuracy(output[idx_train], labels[idx_train])
        loss_train.backward()
        optimizer.step()
    
        if not args.fastmode:
            # Evaluate validation set performance separately,
            # deactivates dropout during validation run.
            model.eval()
            output = model(features, adj)
    
        loss_val = F.nll_loss(output[idx_val], labels[idx_val])
        acc_val = accuracy(output[idx_val], labels[idx_val])
        print('Epoch: {:04d}'.format(epoch+1),
              'loss_train: {:.4f}'.format(loss_train.item()),
              'acc_train: {:.4f}'.format(acc_train.item()),
              'loss_val: {:.4f}'.format(loss_val.item()),
              'acc_val: {:.4f}'.format(acc_val.item()),
              'time: {:.4f}s'.format(time.time() - t))

定义测试函数，相当于对已有的模型再测试集上运行对应的loss与accuracy

![20190418142957273.png][]

逐个epoch进行train，最后test

# Train model
    t_total = time.time()
    for epoch in range(args.epochs):
        train(epoch)
    print("Optimization Finished!")
    print("Total time elapsed: {:.4f}s".format(time.time() - t_total))
    
    # Testing
    test()

## 2.2 models.py ##

定义了模型GCN

## 2.3 layers.py ##

定义了模型如何卷积

## 2.4 utils.py ##

定义了加载数据等工具性的函数，这几个函数后面再详细解析。

[Graph Convolution Network图卷积网络（二）数据加载与网络结构定义 ][Graph Convolution Network_]

[https_github.com_tkipf_pygcn]: https://github.com/tkipf/pygcn
[https_arxiv.org_abs_1609.02907]: https://arxiv.org/abs/1609.02907
[CentOS 6.3_anaconda_pytorch_cuda]: https://blog.csdn.net/weixin_36474809/article/details/88799164
[https_github.com_tkipf_pygcn_tree_master_data_cora]: https://github.com/tkipf/pygcn/tree/master/data/cora
[20190418104517245.png]: https://img-blog.csdnimg.cn/20190418104517245.png
[20190418104542994.png]: /images/20220218/0b45507d9fa94a39b4fa01e65ef165ea.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70]: /images/20220218/9bc84a1da4aa4494969dcce789311192.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 1]: /images/20220218/4a1e2832832d48bda1e001de3b2e6c4a.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 2]: /images/20220218/f19bb16f01954c1582c7e9e7a8ea9be8.png
[20190418142957273.png]: /images/20220218/29193925df024c5a83f7ba55bb0e9db7.png
[Graph Convolution Network_]: https://blog.csdn.net/weixin_36474809/article/details/89379727