1 理论部分

交通预测论文翻译：Deep Learning on Traffic Prediction: Methods,Analysis and Future Directions_UQI-LIUWJ的博客-CSDN博客-4.1.2.1.1 ChebNet

2 类写法

CLASSChebConv(in_channels: int, out_channels: int, K: int, normalization: Optional[str] = 'sym', bias: bool = True, **kwargs)

3 参数说明

in_channels (int)

输入样本的通道数

out_channels (int)

输出样本的通道数

（在Cheb的源码中，每一阶切比雪夫多项式进行卷积之后，都会再过一个FC，这个就是给每一阶的切比雪夫多项式卷积修改维度、调整权重用的）

K (int)

几阶切比雪夫多项式近似

normalization (str, optional)

图拉普拉斯矩阵的归一化方法：默认是sym

None	没有归一化
"sym"	对称归一化
"rw"	随机游走归一化

需要将lambda_max参数提供给forward()方法，以防normalization是不对称的

lambda_max 需要时一个[batch_size]维度的Tensor

可以使用torch_geometric.transforms.LaplacianLambdaMax 方法事先计算lambda_max

bias

默认是True ，如果是False，那么这个ChebNet就不会有偏移量

4 forward 函数

forward(x,edge_index, edge_weight: Optional[torch.Tensor] = None, batch: Optional[torch.Tensor] = None, lambda_max: Optional[torch.Tensor] = None)

注：这里的batch是指torch_geometric笔记：数据集 ENZYMES &Minibatches_UQI-LIUWJ的博客-CSDN博客第2小节中说的batch

5 源码

这里处理得很高妙，它相当于把正则化拉普拉斯矩阵作为新图的邻接矩阵

from typing import Optional
from torch_geometric.typing import OptTensorimport torch
from torch.nn import Parameterfrom torch_geometric.nn.inits import zeros
from torch_geometric.utils import get_laplacian
from torch_geometric.nn.dense.linear import Linear
from torch_geometric.nn.conv import MessagePassing
from torch_geometric.utils import remove_self_loops, add_self_loopsclass ChebConv(MessagePassing):def __init__(self, in_channels: int, out_channels: int, K: int,normalization: Optional[str] = 'sym', bias: bool = True,**kwargs):kwargs.setdefault('aggr', 'add')super(ChebConv, self).__init__(**kwargs)#设置聚合方式（add，也就是将各层切比雪夫多项式近似求和）assert K > 0assert normalization in [None, 'sym', 'rw'], 'Invalid normalization'#两个断言，切比雪夫多项式近似的阶数大于0；在这三种normalization里面选择self.in_channels = in_channelsself.out_channels = out_channelsself.normalization = normalizationself.lins = torch.nn.ModuleList([Linear(in_channels, out_channels, bias=False,weight_initializer='glorot') for _ in range(K)])#各层切比雪夫多项式近似之后接的维度转换全连接层if bias:self.bias = Parameter(torch.Tensor(out_channels))else:self.register_parameter('bias', None)self.reset_parameters()def reset_parameters(self):#初始化参数for lin in self.lins:lin.reset_parameters()zeros(self.bias)def __norm__(self, edge_index, num_nodes: Optional[int],edge_weight: OptTensor, normalization: Optional[str],lambda_max, dtype: Optional[int] = None,batch: OptTensor = None):#这里处理得很高妙，它相当于把正则化拉普拉斯矩阵作为新图的邻接矩阵edge_index, edge_weight = remove_self_loops(edge_index, edge_weight)#去掉自环edge_index, edge_weight = get_laplacian(edge_index, edge_weight,normalization, dtype,num_nodes)#计算拉普拉斯矩阵if batch is not None and lambda_max.numel() > 1:lambda_max = lambda_max[batch[edge_index[0]]]edge_weight = (2.0 * edge_weight) / lambda_maxedge_weight.masked_fill_(edge_weight == float('inf'), 0)#图中所有原来边权重非零的边，权重全部乘以2/lambda_maxedge_index, edge_weight = add_self_loops(edge_index, edge_weight,fill_value=-1.,num_nodes=num_nodes)#由于归一化拉普拉斯矩阵还需要-I，所以所有的自环权重减一assert edge_weight is not Nonereturn edge_index, edge_weight#返回以拉普拉斯矩阵为邻接矩阵的“新图”def forward(self, x, edge_index, edge_weight: OptTensor = None,batch: OptTensor = None, lambda_max: OptTensor = None):""""""if self.normalization != 'sym' and lambda_max is None:raise ValueError('You need to pass `lambda_max` to `forward() in`''case the normalization is non-symmetric.')if lambda_max is None:lambda_max = torch.tensor(2.0, dtype=x.dtype, device=x.device)if not isinstance(lambda_max, torch.Tensor):lambda_max = torch.tensor(lambda_max, dtype=x.dtype,device=x.device)assert lambda_max is not Noneedge_index, norm = self.__norm__(edge_index, x.size(self.node_dim),edge_weight, self.normalization,lambda_max, dtype=x.dtype,batch=batch)#得到以拉普拉斯矩阵为邻接矩阵的“新图”Tx_0 = x#Z_1=Xout = self.lins[0](Tx_0)# propagate_type: (x: Tensor, norm: Tensor)if len(self.lins) > 1:Tx_1 = self.propagate(edge_index, x=x, norm=norm, size=None)#每一轮的propagate相当于对每个点，计算所有邻边的拉普拉斯矩阵权重*临近点,再求和【aggr=add】out = out + self.lins[1](Tx_1)#Z_2=LXfor lin in self.lins[2:]:Tx_2 = self.propagate(edge_index, x=Tx_1, norm=norm, size=None)#Tx_2=Z_k=L*Z_k-1Tx_2 = 2. * Tx_2 - Tx_0#Z_k=2*L*k-1-Z_k-2out = out + lin.forward(Tx_2)Tx_0, Tx_1 = Tx_1, Tx_2if self.bias is not None:out += self.biasreturn outdef message(self, x_j, norm):return norm.view(-1, 1) * x_j#就是对应的邻边权重*邻接点def __repr__(self):return '{}({}, {}, K={}, normalization={})'.format(self.__class__.__name__, self.in_channels, self.out_channels,len(self.lins), self.normalization)

6 举例

from torch_geometric.nn import ChebConvdata
#Batch(x=[9893, 1], edge_index=[2, 34637], y=[9893, 1], batch=[9893], ptr=[2])conv1 = ChebConv(1, 32, 2)x = conv1(data.x, data.edge_index)type(x)
#torch.Tensorx.shape
#torch.Size([9893, 32]) 每个点的维度是[9893,32]

torch_geometric 笔记：nn.ChebNet相关推荐

torch_geometric笔记：nn. graclus （图点分类）
torch_geometric.nn.graclus(edge_index, weight: Optional[torch.Tensor] = None, num_nodes: Optional[in ...
torch_geometric 笔记：TORCH_GEOMETRIC.UTILS（更新中）
1 torch_geometric.utils.add_self_loops add_self_loops(edge_index, edge_weight: Optional[torch.Tensor ...
torch_geometric笔记：max_pool 与max_pool_x
1 max_pool 1.1 函数介绍 torch_geometric.nn.max_pool(cluster, data,transform=None) 对由torch_geometricy .da ...
torch_geometric 笔记：数据集Cora 简易 GNN
1 获取数据集该数据集用于semi-supervised的节点分类任务 from torch_geometric.datasets import Planetoiddataset = Planeto ...
torch_geometric笔记：数据集 ENZYMES Minibatches
Pytorch Geometric中包含大量的常见基准数据集.在初始化数据集的时候,框架会自动下载数据集的原始文件,并将其处理为Data对象.例如要下载ENZYMES数据集(由600个graph划分为 ...
Hadoop学习笔记-NN与2NN
几个概念 NameNode被格式化之后,将在/opt/module/hadoop-3.1.3/data/tmp/dfs/name/current目录中产生如下文件: fsimage_000000000 ...
torch_geometric 笔记：global_mean_pool
对全图的点嵌入(node embedding)进行池化操作,返回一个图嵌入(graph embedding) global_mean_pool(x, batch, size=None)[ x (tor ...
pytorch 学习笔记目录
1 部分内容 pytorch笔记 pytorch模型中的parameter与buffer_刘文巾的博客-CSDN博客 pytorch学习笔记 torchnn.ModuleList_刘文巾的博客-CSD ...
Pytorch使用笔记
Pytorch使用笔记 nn.Module CNN torch.nn.Conv1d torch.nn.Conv2d torch.nn.ConvTranspose1d RNN LSTM self.reg ...

torch_geometric 笔记：nn.ChebNet