KronEM problem about the AS dataset experiment in Section 4.2 of the paper "The Network Completion Problem: Inferring Missing Nodes and Edges in Networks"

[yonz to]

Hello everyone, 

 I just review the KronEM code on SNAP in C++(snap\examples\kronem\) and have some questions about the experiment.


1. I find that the data used in kronem.cpp is "as20graph.txt". But it contains 6474 nodes and 26467 edges of G, which is different from the experiment data described in Section 4.2 of the paper. So if it is the file used for the trial and how do you get the part of 4,096 nodes, 29,979 edges of the Autonomous System (AS) network of the Internet connectivity?

2. For the KronEM code in C++(snap\examples\kronem\), I guess the authors mainly use "ERR"(=fabs(EZero-ProbMtx.GetMtxSum())) to check the KronEM algorithm correctness. But I didn't find the ERR in the paper and don't know how it works. How do you prove the final FITTED PARAMS is the TRUE parameter matrix of the input graph?  

3. Continuing the previous point, the task of KronEM is to infer the missing part Z of the network. I wonder if you use the method to prove the Correctness of KronEM in the KronEM C++ code(snap\examples\kronem\), i.e., do you have the REAL COMPLETE/WHOLE NETWORK for the "as20graph.txt" so that I can check if the inferred Z part is close to the real Z part?
Besides, I find the result of the cpp file(snap\examples\kronem\kronem.cpp) is only a convergent 2*2 FITTED PARAMS which still leaves a final step for the recovery of the inferred Z part. How do you recover the missing part(Z part) from the final 2D FITTED PARAMS?

4. Could you please provide more details about the calculation of AUC as described in Section 4.2 of the paper? 

Thank you!

Best regards,
Yongzai