Is there a problem whith the ufactor for kway partitionning?
Hi everybody,
I'm trying to use the new version of hmetis (hmetis-2.0pre1) to partition an hypergraph.
But it seems like the ufactor have no influence. I think i don't use it properly!!
I did a test in a simple image (white square on dark background) and i created the corresponding hypergraph (40000 nodes and 40000 edges).
When i used ufactor=20.0, i got this:
Partition Sizes & External Degrees:
( 10073.000)[ 250.000] ( 9781.000)[ 98.000] ( 10073.000)[ 266.000]
( 10073.000)[ 418.000]
The partition sizes are almost the same, whereas, according to the definition of ufactor, the size of the largest partition can be 1.20 * (40000/4) = 12000.
When i used ufactor=150.0, i got this:
Partition Sizes & External Degrees:
( 9717.000)[ 98.000] ( 10027.000)[ 249.000] ( 10129.000)[ 262.000]
( 10127.000)[ 413.000]
wheareas the size of the largest partition can be 2.50 * (40000/4) = 25000.
I dont understand why, because the results i get are not relevant.
Here is the command line i did:
./hmetis2.0pre1 -ptype kway -ctype gfc1 -ufactor 150.0 -nruns 25 -nvcycles 5 -cmaxnet 50 /data/lienou/clc/progs2/hypergraphes/testhgraph.graph 4
Can anybody help me please?
- Login to post comments
RE: Can you open an issue on
Can you open an issue on flyspray for that as it appears to be a bug.
RE: Flyspray??
Hello,
Do you mean Flyspray, the bug tracking system???
I don't know exactly how it works, but i will see.
I wonder if anybody else had the same problem...
RE: Yes. The one for Metis is
Yes. The one for Metis is here: http://glaros.dtc.umn.edu/flyspray/index.php?tasks=all&project=2
RE: ok
ok, i will open an issue on flyspray.
Thank you