MindSpore实践图神经网络04之GCN模型实践

GCN介绍

• 图卷积网络（GCN）于2016年提出，旨在对图结构数据进行半监督学习。它提出了一种基于卷积神经网络有效变体的可扩展方法，可直接在图上操作。该模型在图边缘的数量上线性缩放，并学习隐藏层表示，这些表示编码了局部图结构和节点特征。

• GCN（图卷积神经网络) 类似CNN（卷积神经网络），只不过CNN用于二维数据结构，GCN用于图数据结构。GCN实际上跟CNN的作用一样，就是一个特征提取器，只不过它的对象是图数据。GCN精妙地设计了一种从图数据中提取特征的方法。

• GCN包含两个图卷积层。每一层以节点特征和邻接矩阵为输入，通过聚合相邻特征来更新节点特征。

环境配置

• 配置MindSpore环境

# 控制台安装mindspore 
conda create -n py39_ms18 python=3.9
conda activate py39_ms18

pip install https://ms-release.obs.cn-north-4.myhuaweicloud.com/1.8.1/MindSpore/cpu/x86_64/mindspore-1.8.1-cp39-cp39-linux_x86_64.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.tuna.tsinghua.edu.cn/simple

# 验证是否安装成功
python -c "import mindspore;mindspore.run_check()"

conda activate py39_ms18

• 配置python环境

conda activate py39_ms18

pip install numpy
pip install scipy
pip install sklearn
pip install pyyaml
# 缺包
pip  install matplotlib

算子开发

• 算子开发：Layer、Model

# 定义算子：Layer
class GraphConvolution(nn.Cell):
    def __init__(self,
                 feature_in_dim,
                 feature_out_dim,
                 dropout_ratio=None,
                 activation=None):
        super(GraphConvolution, self).__init__()
        self.in_dim = feature_in_dim
        self.out_dim = feature_out_dim
        self.weight_init = glorot([self.out_dim, self.in_dim])
        self.fc = nn.Dense(self.in_dim,
                           self.out_dim,
                           weight_init=self.weight_init,
                           has_bias=False)
        self.dropout_ratio = dropout_ratio
        if self.dropout_ratio is not None:
            self.dropout = nn.Dropout(keep_prob=1-self.dropout_ratio)
        self.dropout_flag = self.dropout_ratio is not None
        self.activation = get_activation(activation)
        self.activation_flag = self.activation is not None
        self.matmul = P.MatMul()

    def construct(self, adj, input_feature):
        """
        GCN graph convolution layer.
        """
        dropout = input_feature
        if self.dropout_flag:
            dropout = self.dropout(dropout)

        fc = self.fc(dropout)
        output_feature = self.matmul(adj, fc)

        if self.activation_flag:
            output_feature = self.activation(output_feature)
        return output_feature

# 定义模型：Model
class GCN(nn.Cell):
    def __init__(self, config, input_dim, output_dim):
        super(GCN, self).__init__()
        self.layer0 = GraphConvolution(input_dim, config.hidden1, activation="relu", dropout_ratio=config.dropout)
        self.layer1 = GraphConvolution(config.hidden1, output_dim, dropout_ratio=None)

    def construct(self, adj, feature):
        output0 = self.layer0(adj, feature)
        output1 = self.layer1(adj, output0)
        return output1

• 数据处理utils

# 归一化邻接矩阵
def normalize_adj(adj):
    """Symmetrically normalize adjacency matrix."""
    rowsum = np.array(adj.sum(1))
    d_inv_sqrt = np.power(rowsum, -0.5).flatten()
    d_inv_sqrt[np.isinf(d_inv_sqrt)] = 0.
    d_mat_inv_sqrt = sp.diags(d_inv_sqrt)
    return adj.dot(d_mat_inv_sqrt).transpose().dot(d_mat_inv_sqrt).tocoo()

# 加载数据集  : Cora
def get_adj_features_labels(data_dir):
    """Get adjacency matrix, node features and labels from dataset."""
    g = ds.GraphData(data_dir)
    nodes = g.get_all_nodes(0)
    nodes_list = nodes.tolist()
    row_tensor = g.get_node_feature(nodes_list, [1, 2])
    features = row_tensor[0]
    labels = row_tensor[1]

    nodes_num = labels.shape[0]
    class_num = labels.max() + 1
    labels_onehot = np.eye(nodes_num, class_num)[labels].astype(np.float32)

    neighbor = g.get_all_neighbors(nodes_list, 0)
    node_map = {node_id: index for index, node_id in enumerate(nodes_list)}
    adj = np.zeros([nodes_num, nodes_num], dtype=np.float32)
    for index, value in np.ndenumerate(neighbor):
        # The first column of neighbor is node_id, second column to last column are neighbors of the first column.
        # So we only care index[1] > 1.
        # If the node does not have that many neighbors, -1 is padded. So if value < 0, we will not deal with it.
        if value >= 0 and index[1] > 0:
            adj[node_map[neighbor[index[0], 0]], node_map[value]] = 1
    adj = sp.coo_matrix(adj)
    adj = adj + adj.T.multiply(adj.T > adj) + sp.eye(nodes_num)
    nor_adj = normalize_adj(adj)
    nor_adj = np.array(nor_adj.todense())
    return nor_adj, features, labels_onehot, labels

# 数据集划分
def get_mask(total, begin, end):
    """Generate mask."""
    mask = np.zeros([total]).astype(np.float32)
    mask[begin:end] = 1
    return mask

Windows环境跑脚本报错（1）

问题描述

/mnt/d/mindspore_gallery/models/gnn/gcn/data
cora
data_mr exist
scripts/run_process_data.sh: line 46: cd: ../../../utils/graph_to_mindrecord: No such file or directory

根因分析

• 由报错信息可以看出可能是数据集存放路径不对，或者windows下脚本和Linux不一致

解决办法

• 修改路径,改为如下路径

../../utils/graph_to_mindrecord

• 改到Linux环境，如果没有Linux环境可以安装WSL2，创建Ubuntu环境
  

Windows环境跑脚本报错（2）

问题描述

{'data_dir': 'Dataset directory', 'train_nodes_num': 'Nodes numbers for training', 'eval_nodes_num': 'Nodes numbers for evaluation', 'test_nodes_num': 'Nodes numbers for test', 'save_TSNE': 'Whether to save t-SNE graph'}
Traceback (most recent call last):
  File "D:\mindspore_gallery\models\gnn\gcn\train.py", line 196, in <module>
    run_train()
  File "D:\mindspore_gallery\models\gnn\gcn\model_utils\moxing_adapter.py", line 105, in wrapped_func
    run_func(*args, **kwargs)
  File "D:\mindspore_gallery\models\gnn\gcn\train.py", line 114, in run_train
    context.set_context(mode=context.GRAPH_MODE,
  File "C:\Users\sunxiaobei\.conda\envs\py39_ms18\lib\site-packages\mindspore\_checkparam.py", line 1210, in wrapper
    return func(*args, **kwargs)
  File "C:\Users\sunxiaobei\.conda\envs\py39_ms18\lib\site-packages\mindspore\_checkparam.py", line 1179, in wrapper
    return func(*args, **kwargs)
  File "C:\Users\sunxiaobei\.conda\envs\py39_ms18\lib\site-packages\mindspore\context.py", line 911, in set_context
    raise ValueError(f"For 'context.set_context', package type {__package_name__} support 'device_target' "
ValueError: For 'context.set_context', package type mindspore support 'device_target' type cpu, but got Ascend.

根因分析

• 从log上不难看出，是代码指定的设备不一致，当前设备只有CPU，但是指定的是Ascent , 需要指定和实际环境一致的设备

解决办法

• 修改代码，指定CPU

    context.set_context(mode=context.GRAPH_MODE,
                        device_target="CPU", save_graphs=False)  # CPU  Ascend  GPU

运行代码

python train.py --data_dir=./data_mr/citeseer --train_nodes_num=120