代码链接：RIPGeo代码实现
├── lib # 包含模型（model）实现文件
│ |── layers.py # 注意力机制的代码。
│ |── model.py # TrustGeo的核心源代码。
│ |── sublayers.py # layer.py的支持文件。
│ |── utils.py # 辅助函数。
一、导入各种模块和神经网络类
from .layers import *
import torch
import torch.nn as nn这段代码是一个 Python 模块，包含了一些导入语句和定义了一个神经网络模型的类。
from .layers import *：导入了当前模块所在目录中的 layers 模块中的所有内容。* 表示导入所有的内容。
二、RIPGeo类定义（NN模型）
class RIPGeo(nn.Module):def __init__(self, dim_in, dim_z, dim_med, dim_out, collaborative_mlp=True):super(RIPGeo, self).__init__()# RIPGeoself.att_attribute = SimpleAttention1(temperature=dim_z ** 0.5,d_q_in=dim_in,d_k_in=dim_in,d_v_in=dim_in + 2,d_q_out=dim_z,d_k_out=dim_z,d_v_out=dim_z)if collaborative_mlp:self.pred = SimpleAttention2(temperature=dim_z ** 0.5,d_q_in=dim_in * 3 + 4,d_k_in=dim_in,d_v_in=2,d_q_out=dim_z,d_k_out=dim_z,d_v_out=2,drop_last_layer=False)else:self.pred = nn.Sequential(nn.Linear(dim_z, dim_med),nn.ReLU(),nn.Linear(dim_med, dim_out))self.collaborative_mlp = collaborative_mlp# calculate Aself.gamma_1 = nn.Parameter(torch.ones(1, 1))self.gamma_2 = nn.Parameter(torch.ones(1, 1))self.gamma_3 = nn.Parameter(torch.ones(1, 1))self.alpha = nn.Parameter(torch.ones(1, 1))self.beta = nn.Parameter(torch.zeros(1, 1))# transform in Graphself.w_1 = nn.Linear(dim_in + 2, dim_in + 2)self.w_2 = nn.Linear(dim_in + 2, dim_in + 2)def forward(self, lm_X, lm_Y, tg_X, tg_Y, lm_delay, tg_delay):""":param lm_X: feature of landmarks [..., 30]: 14 attribute + 16 measurement:param lm_Y: location of landmarks [..., 2]: longitude + latitude:param tg_X: feature of targets [..., 30]:param tg_Y: location of targets [..., 2]:param lm_delay: delay from landmark to the common router [..., 1]:param tg_delay: delay from target to the common router [..., 1]:return:"""N1 = lm_Y.size(0)N2 = tg_Y.size(0)ones = torch.ones(N1 + N2 + 1).cuda()lm_feature = torch.cat((lm_X, lm_Y), dim=1)tg_feature_0 = torch.cat((tg_X, torch.zeros(N2, 2).cuda()), dim=1)router_0 = torch.mean(lm_feature, dim=0, keepdim=True)all_feature_0 = torch.cat((lm_feature, tg_feature_0, router_0), dim=0)'''star-GNNproperties:1. single directed graph: feature of <landmarks> will never be updated.2. the target IP will receive from surrounding landmarks from two ways: (1) attribute similarity-based one-hop propagation;(2) delay measurement-based two-hop propagation via the common router;'''# GNN-step 1adj_matrix_0 = torch.diag(ones)delay_score = torch.exp(-self.gamma_1 * (self.alpha * lm_delay + self.beta))rou2tar_score_0 = torch.exp(-self.gamma_2 * (self.alpha * tg_delay + self.beta)).reshape(N2)# feature_, attribute_score = self.att_attribute(tg_X, lm_X, lm_feature)attribute_score = torch.exp(attribute_score)adj_matrix_0[N1:N1 + N2, :N1] = attribute_scoreadj_matrix_0[-1, :N1] = delay_scoreadj_matrix_0[N1:N1 + N2:, -1] = rou2tar_score_0degree_0 = torch.sum(adj_matrix_0, dim=1)degree_reverse_0 = 1.0 / degree_0degree_matrix_reverse_0 = torch.diag(degree_reverse_0)degree_mul_adj_0 = degree_matrix_reverse_0 @ adj_matrix_0step_1_all_feature = self.w_1(degree_mul_adj_0 @ all_feature_0)tg_feature_1 = step_1_all_feature[N1:N1 + N2, :]router_1 = step_1_all_feature[-1, :].reshape(1, -1)# GNN-step 2adj_matrix_1 = torch.diag(ones)rou2tar_score_1 = torch.exp(-self.gamma_3 * (self.alpha * tg_delay + self.beta)).reshape(N2)adj_matrix_1[N1:N1 + N2:, -1] = rou2tar_score_1all_feature_1 = torch.cat(还有谁？)
