search

YuxiangRen / Heterogeneous-Deep-Graph-Infomax

HDGI code
59 stars 14 forks source link

The Question about the cluster experiment of HDCN #16

Open YcZ76 opened 3 years ago

YcZ76 commented 3 years ago

Hello, Mr Ren, after reading your paper, as you described in the paper, the gcns performance is better than the gat, while in the code, i only discover the cluster code about the gat, in the DGI-HGCN, i cannot find the cluster experiment of this gcns framework, could it really improve the cluster performance? (owing to missing this part code, leading these two framework is not symmetric, no offense, i`m curious because im a rookie ,thanks!!!)

YuxiangRen commented 3 years ago

Because the codes for clustering experiments on DGI-HGCN and DGI-HGAT are exactly the same, so I did not put a redundant code in DGI-HGCN. You can use the code in executeCla.py in DGI-HGAT to test DGI-HGCN version. The process is totally the same, that is, extracting embedding and executing the clustering. I hope this response can help you.

YcZ76 commented 3 years ago

Because the codes for clustering experiments on DGI-HGCN and DGI-HGAT are exactly the same, so I did not put a redundant code in DGI-HGCN. You can use the code in executeCla.py in DGI-HGAT to test DGI-HGCN version. The process is totally the same, that is, extracting embedding and executing the clustering. I hope this response can help you.

Thanks for replying to me so quickly. Could i ask more several simple questions (that might seem to be naive,1: could you give me more information about settings,like the version of python and pytorch? thx. 2: Using gae to substitute gcn to encode the node representation, could this way is available? ) This is my first time to comment ，receiving your response helps me a lot ,thank you very much!!!

YuxiangRen commented 3 years ago

Because the codes for clustering experiments on DGI-HGCN and DGI-HGAT are exactly the same, so I did not put a redundant code in DGI-HGCN. You can use the code in executeCla.py in DGI-HGAT to test DGI-HGCN version. The process is totally the same, that is, extracting embedding and executing the clustering. I hope this response can help you.

Thanks for replying to me so quickly. Could i ask more several simple questions (that might seem to be naive,1: could you give me more information about settings,like the version of python and pytorch? thx. 2: Using gae to substitute gcn to encode the node representation, could this way is available? ) This is my first time to comment ，receiving your response helps me a lot ,thank you very much!!!

In fact, HDGI is flexible for choosing graph encoder. You absolutely can try GAE as the graph encoder, but I didn't try it before and have no idea about the performance. Besides, I used python 3.6.6 and pytorch 1.1.0 when conducting the experiments.

YcZ76 commented 3 years ago

Because the codes for clustering experiments on DGI-HGCN and DGI-HGAT are exactly the same, so I did not put a redundant code in DGI-HGCN. You can use the code in executeCla.py in DGI-HGAT to test DGI-HGCN version. The process is totally the same, that is, extracting embedding and executing the clustering. I hope this response can help you.

Thanks for replying to me so quickly. Could i ask more several simple questions (that might seem to be naive,1: could you give me more information about settings,like the version of python and pytorch? thx. 2: Using gae to substitute gcn to encode the node representation, could this way is available? ) This is my first time to comment ，receiving your response helps me a lot ,thank you very much!!!

In fact, HDGI is flexible for choosing graph encoder. You absolutely can try GAE as the graph encoder, but I didn't try it before and have no idea about the performance. Besides, I used python 3.6.6 and pytorch 1.1.0 when conducting the experiments.

Thank you a lot, because i dont have a strong code capability, i cannot understand your choice about graph encoder, that is, i even do not know youve chose graph encoder in HDGI,(just GCN i think before) maybe i misunderstand the definition of graph encoder, i`m so sorry to wasting your time here, thanks for your kindness and patience!! wish you have a good day!

YcZ76 commented 3 years ago

Hello,Mr Ren,i am sorry to bother you again, i want to use IMDB to take place of DBLP in HGAT, after change the process.py, def_load_data ,it cannot work, RunTimeError enfore fail at CPUALlocator.cpp , should i change the other code? thanks.

YuxiangRen commented 3 years ago

CPUALlocator.cpp

The error may occur when there's no enough CPU memory. I don't think it is caused by codes in other places.