partition message
I'm using METIS to do some basic partitioning work. Sometimes I got the following message
***Cannot bisect a graph with 0 vertices!
***You are trying to partition a graph into too many parts!
The partitions are not approximately equal (e.g., nparts==8 with only 3 parts generated) but the results look okay. Could you comment on this message from METIS?
Submitted by wxzhou on Tue, 2009-05-05 21:24
»
- Login to post comments
RE: In most cases the underlying
In most cases the underlying reason for these type of messages is a graph whose vertex weights are very skewed. Is this the case over here?
RE: There are several isolated
There are several isolated regions with different weights in the graph. Could this be the reason? Is this message harmless?
RE: That can be one of the
That can be one of the reasons. Also, are any vertices whose weight is greater than the final weight of each partition?
RE: Thanks for your reply. No,
Thanks for your reply. No, if an empty partition is not counted.
Those isolated regions are not evenly distributed and the weights of vertices vary from 0 to 10. After METIS call, there are a few empty partitions and sometimes it just puts all vertices into one partition.
So this message is more like a warning and may not indicate a real problem?
RE: Metis does not like
Metis does not like zero-weight vertices...