Acknowledgments Our work and code benefit from two existing works, which we are very grateful. DAG NOTEARhttps://github.com/xunzheng/notears Neural relational inference for interacting systemshttps://github.com/ethanfetaya/nri Contributors2
你可以去这篇论文的github项目里看一下,在modules.py里,以里边的MLPDecoder为例,在其forward()部分,有定义# adj_A_new1 = (I-A^T)^(-1)adj_A_new1 = preprocess_adj_new1(origin_A)而preprocess_adj_new1里边是通过torch.inverse()来求逆的(该函数定义在utils.py里) 2022-03-07 回复1 ...
源码链接:https://github.com/fishmoon1234/DAG-GNN TL;DR 论文中提出一种新的DAG编码架构 DAG-GNN,其实模型的本质就是一个图变分自编码器,模型的优点是既能处理连续型变量又能处理离散型变量;在人工数据集和真实数据集中验证了模型结果可以达到全局最优 ; Model / Algorithm 论文中的整体模型架构如下: Linear ...
code https://github.com/vgsatorras/few-shot-gnn.git paper https://arxiv.org/pdf/1711.04043.pdf 论文的原理可以直接看论文或者博客,这里主要记录论文中一些含糊的细节 setting 分为 few-shot setting,semi-supervised se... Few-Shot Learning with Graph Neural Networks(图神经网络小样本学习) ...