It turned out to make weight * -1 const can easily transfer the problem to MAX-K-CUT issues, but now find a little problem
The problem is that the UPBound I used by the algorithm is the weight and as a benchmark. When you do a problem conversion, you will change this value. If it is too big, it will reduce the algorithm performance. This is not thinking about it. of. Although the result of the final group is unaffected, it feels that the core prove is a step, it is always uncomfortable. . . .
So, then, more about the paper is good. . . I found some papers, I found Metis, seemingly powerful, at least it directly supports Balance, Zhou 1 to compare with my algorithm. . . Happy. . .