what to modify in order to be able to run on Mac OS?

[Yujun-Yan]

Hi, I am trying to do the experiment on my computer which has a Mac system and I have downloaded the corresponding matlab runtime for Mac.
However it seems that coarsening cannot be properly executed. I guess it has something wrong with the LD_LIBRARY_PATH in the run_coarsening.sh file.
My question is: if I would like to run it with a Mac system, how should I modify the files?

[zhangzhiru]

Thanks for your interest. Unfortunately, we currently don't have support for macOS. One option is work around this issues is to install a Linux VM. In the future, we will likely release a version that does not require the matlab package.

[dreambear1234]

Thanks for your interest. Unfortunately, we currently don't have support for macOS. One option is work around this issues is to install a Linux VM. In the future, we will likely release a version that does not require the matlab package.

Thanks for your great job! In graph_coarsening, simple version differ from lamg version, and accuracy is low for simple version as well as it has no gauss-seidel iteration, so that it is a random process for simple version. so lamg version is very important !
unfortunately, for me, your matlab code is hard to read, and cannot debug easily, too many details in your code which not appeared in paper, could you supply python or c++ version. if the work is not over, draft version is also welcome.
Thank you very much!

[Chenhui1016]

Hi,

Simple coarsening adopts a similar idea of LAMG. Gauss-Seidel (GS) iteration is just one way to smooth node embedding. For simple coarsening, we use low-pass graph filter instead of GS to smooth the embedding. We are still working on improving simple coarsening to make its accuracy close to LAMG.

[dreambear1234]

Hi,

Simple coarsening adopts a similar idea of LAMG. Gauss-Seidel (GS) iteration is just one way to smooth node embedding. For simple coarsening, we use low-pass graph filter instead of GS to smooth the embedding. We are still working on improving simple coarsening to make its accuracy close to LAMG.

Hi,
https://github.com/cornell-zhang/GraphZoom/blob/master/graphzoom/utils.py#L185
From your code, In spec_coarsen function, tv_feat is a random matrix before low-pass graph filter, and use it to calculate the affinity. if process of the smooth is skipped, it means that we choose node to be clustered randomly? isn't it weired ?

[Chenhui1016]

As shown in our paper (specifically, Equation (1)), we first generate random vector for each node and then apply smoothing function to smooth it. The smoothed vector mainly consists of low-frequency component and thus preserves the key spectral properties for each node. Finally, we use such smoothed vector to measure node affinity. Hope this makes sense to you.