import torch

from torch.nn import TransformerEncoder, TransformerEncoderLayer

d_model = 64

nhead = 4

num_layers = 2

batch_size = 16

seq_len = 20

# 定义编码器层

encoder_layer = TransformerEncoderLayer(d_model=d_model, nhead=nhead, batch_first=True)

encoder = TransformerEncoder(encoder_layer, num_layers=num_layers)

# 输入张量 (batch_size, seq_len, d_model)

src = torch.randn(batch_size, seq_len, d_model)

output = encoder(src)

print(output.shape)

import torch

import torch.nn.functional as F

class GCNLayer(torch.nn.Module):

   def init(self, in_feat, out_feat):

       super().__init__()

       self.weight = torch.nn.Linear(in_feat, out_feat)

   def forward(self, x, adj):

       x = self.weight(x)

       x = _______  # 邻接矩阵与特征矩阵相乘

       return F.relu(x)

import torch

def normalize_adj(adj):

   I = torch.eye(adj.size(0))

   adj = adj + I  # 添加自环

   # 计算度矩阵的逆平方根

   degree = torch.sum(adj, dim=1)

   degree_inv_sqrt = torch.pow(degree, -0.5)

   degree_inv_sqrt[degree_inv_sqrt == float('inf')] = 0.0

   degree_matrix = torch.diag(degree_inv_sqrt)

   adj_normalized = _______

   return adj_normalized