dkondor / maxmatch_mv Goto Github PK

Hi there,

I'm currently working on approximate maximum cardinality matching algorithms and stumbled upon your implementation, when searching for a fast implementation of the Micali-Vazirani algorithm in order to benchmark my own results.

Although your implementation is quite fast, there seems to be a bug in your code: When run several times using the same input, the result differs between two values. Apparently sometimes your program does not search for the augmenting paths after passing the greedy initialization phase. I counter-checked the results of your program using the boost implementation of Edmond's blossom algorithm (https://github.com/ripaul/edmonds_blossom). There is also a test input for which you should be able to observe the bug. I tested your program on two machines using the test, on one the strange behavior only appears, when compiling without the -O3 flag. Also, when compiling with -Wall flag, a lot of warnings show up.

My results for the provided test input are size 4653 (which is wrong) and 6654 (which seems to be right, when counter-checking with the boost implementation and this one)

A fast and reliable MCM-algorithm would be quite helpful for my work, so I'd appreciate any help regarding this issue. Thanks in advance!

Best,
Richi