Ported to Javascript about lapjv-algorithm-c HOT 7 CLOSED

from lapjv-algorithm-c.

Please check the value of epsilon in my version. Without it my tests were quite often going into infinite loops, because of rounding errors. With epsilon I don't have this problem anymore. I'm sorry I don't have a reproductible example at hand, I think it was when there were around 5000 positions to optimize.

from lapjv-algorithm-c.

Here is a reproducible test:
https://gist.github.com/Fil/377f9aa3f8f07c428d54eec3564f0eba

run it with node.js; if you change epsilon to 0 on line 82, you quite often get into an infinite loop because of rounding errors on the tests

from lapjv-algorithm-c.

Hi Phil,
I found a bug in my code, in lap.h file, the row and col type should be int, not double,I have fixed it and now the cpp version can work. Since I'm not familiar with java script, I haven't find error in your js version.

from lapjv-algorithm-c.

To be clear: the JS version works well.

However it is not strictly a port of your C code: the difference is this epsilon that I added, otherwise it would sometimes go into an infinite loop.

My question is if the C version also has this problem when two distances are very close and the comparison fails because of rounding.

from lapjv-algorithm-c.

Hi Phil,
I see you add epsilon as tolerance in js version. Some cases can cause Jonker's code to loop forever. According to Dr. Volgenant, "The failure is a consequence of a tolerance not suited for the data ... You have to reduce the value of the tolerance such that it is smaller than the smallest difference between any pair of cost elements in the cost matrix."
You can also see Nota Bene part in this link.

Yong Yang

from lapjv-algorithm-c.

Thank you I will mention my solution there too. Maybe a good solution will come out of this!

from lapjv-algorithm-c.