Anton Tsitsulin

Hello, you can request the code by email. We want to make sure people use the correct VERSE version in their experiments.

Hi, icpc is Intel's c++ compiler. https://software.intel.com/content/www/us/en/develop/tools/compilers/c-compilers.html You can also use gcc, which will work just fine.

No, you are not correct. wrapper.py should work after you run `make` in the src folder. You need to input the correct cpath parameter to the wrapper.

Note that I still sort the nodes, so the order should be deterministic.

What do you mean by results? By default, the embeddings are initalized randomly each time, so it's normal that you get different embedding each time you run the model.

If you converted the graphs with the supplied converter, then the node IDs are sorted in the ascending order.

Are you sure you add cores rather than threads? What is the size of the graph you are trying to process? Most likely it's a memory issue.

If the size of the graph is really large, there might still be memory issues, especially in NUMA architectures. Can you (roughly) specify the size of the graph, and memory...

From this limited information my most probably cause is memory access time. There is single DRAM contoller fetching graph (and embedding) parts from the memory, causing a bottleneck. This is...

If you want to see where the problem comes from, I would recommend running the process under GNU perf (tutorial here http://www.brendangregg.com/perf.html). I would expect that the running time is...