构建 graph 问题探讨

SsGood / MMGL

Multi-modal Graph learning for Disease Prediction (IEEE Trans. on Medical imaging, TMI2022)

MIT License

92 stars 14 forks source link

构建 graph 问题探讨 #3

Closed chenluda closed 2 years ago

chenluda commented 2 years ago

恭喜！非常精彩的工作！

这里探讨个问题：

代码里似乎是将 train set 和 test set 统一输入构建 graph，train phase 中 test set 不参与 bp。

如果是这样，那么论文中的

“ the unseen patient u is first added to the existing population graph by MARL with AGL. ... It should be noting that the test patients are not kept in the graph in the testing phase. ”

该怎么理解呢？

期待回复，谢谢！

SsGood commented 2 years ago

首先感谢您对我们工作的关注！我们当前开源的版本是transductive learning的版本，即您提到的

代码里似乎是将 train set 和 test set 统一输入构建 graph，train phase 中 test set 不参与 bp。

论文中的描述是inductive learning的版本，该版本的代码我们将在10月份开源，请您持续关注，谢谢！

Thanks for your attention to our work! The current repository is a PyTorch transductive version of the MMGL model as proposed in our paper, the inductive version will be coming soon in October, please stay tuned, thank you!

chenluda commented 2 years ago

收到，感谢您的回复！