Questions about evaluation details

[xmlyqing00]

Hi,

Thank you for providing the training and evaluation codes. In the evaluation part, you solve the quadratic constraint (Eqn 7) to get the optimal assignment. In your code, the QC-optimization starts from line 128 in QCDGM/model.py. The quadratic constraints are AA_src and BB_tgt

After that, I think the assignment is stored in the variable s (or X? I'm confused here, I think it should be X but you return s instead) and it is returned to eval.py as s_pred.

You do a Hungarian algorithm here (eval.py line 81) to discrete the values. That makes sense. But I can't follow the following several lines. You do the qc_opt again with constraints A_src and A_tgt, which represents edge connections. Why do you need to do it as "post-processing"? In your training codes, I don't see such post-processing there.

Thanks for your time.