Robert Hailey wrote:
> I don't know, it just seems like a degree-6 graph should not have  
> node(s) with 21 connections.

Agreed, it should be very rare (though not impossible). The degree 
distribution should be Poisson, approximately normal for large graphs. 
The histogram at the end of the test code output looks about right to me 
- what kind of degree distribution are you getting in RealNodeTest?

> Something's afowl here, the array-scanning was just my best guess when  
> I ran out of time to track it down. Running your sim at the stats of  
> the present network coloring test (50 nodes, degree=6), I get a graph  
> with an average degree of about 4 and a peak connection count of 8  
> (not 21).

That sounds about right - the actual mean will usually be lower than the 
specified mean, especially in small networks, because duplicate 
connections are discarded.

> Without Michael's forceNeighborConnections mod, it would also make  
> quite a number of leaf nodes. I don't recall if it makes zero- 
> connection nodes.

Again, it's possible but it should be very rare.

(Slightly updated version of the test code attached - you can now choose 
regular or random locations. Still no correlation between index and degree.)

Cheers,
Michael
-------------- next part --------------
A non-text attachment was scrubbed...
Name: KleinbergDegreeTest.java
Type: text/x-java
Size: 2375 bytes
Desc: not available
URL: 
<https://emu.freenetproject.org/pipermail/devl/attachments/20080304/b42f38a7/attachment.java>

Reply via email to