shawnh2 / gnn-cca Goto Github PK

Hi @shawnh2,

Firstly, I want to thank you for the implementation. I have a question regarding the metrics provided in Table 1. When I compared the results given in Table 1 with the results provided in the original paper (TABLE IV), I noticed that the results were not similar. I assume that you used the default settings provided in the paper, I wonder what causes this difference.

For instance, in Table 1, the results for the basketball test sequence are:

In TABLE IV, the basketball sequence corresponds to the S1 set, and the results given in the paper are

I would be glad if you can answer my question. Thanks.

After the experiment, the most time-consuming part is MPN due to a loop inside.
It's more efficient if using graph.update_all and write udf for both message_func and reduce_func rather than looping all subgraphs.
There is about 60x faster than original method.

Hi, I am very interested in this topic.
But i can't find requirements.txt in your github.
Could you provide requirements.txt?

Thank you so much!