Dear Andre

Sorry - missed this mail first.

I'll try to send the data, but the file is big and I am in a place without good connections. May be it is easier to use the data of Edwin.

Round jaccard = choice which similarity is used from the raw jaccard query.

Ok, I understand the bug: in the old version all three are equal as they all were treated as unvalued.

In the new version this still the case for the roundjaccard and the unvalued. But not equal to the old unvalued one. May be the only problem in the new version is that you now do not correctly calclate the unvalued.

Best
Peter

Peter van den Besselaar
----------------------------
Professor
Organization Sciences, VU University Amsterdam
www.vandenbesselaar.net
-------------------------------------------------------
Rathenau Instituut / Royal Netherlands Academy of Arts and Sciences professor of organization and dynamics of science
-------------------------------------------------------
(with my BlackBerry)


--------------------------------------------------------------------------------

From: André Somers <a.somers@rathenau.nl>
Date: Fri, 16 Jul 2010 10:13:33 +0200
To: Peter van den Besselaar<p.vandenbesselaar@rathenau.nl>; James Dawson<j.dawson@rathenau.nl>; Thomas Gurney<t.gurney@rathenau.nl>; 'edwin.horlings@gmail.com'<edwin.horlings@gmail.com>
Cc: Edwin Horlings<e.horlings@rathenau.nl>
Subject: RE: New version of NetworkTools


Peter,

Thanks for your thorough test! That is very helpful.
Could you please send me the dataset you worked on. I don't know what "rounded jaccard" is exactly.

As noted in the email, the new version fixes a bug where the value of the link was (sometimes?) not taken into account. That is: the network was treated as an unvalued network. That is obviously incorrect. It is strange however that the 'unvalued' version is not equal in both. I think this needs a bit more analysis.

André


--------------------------------------------------------------------------------
From: Peter van den Besselaar
Sent: 16 July 2010 09:57
To: André Somers; James Dawson; Thomas Gurney; 'edwin.horlings@gmail.com'
Cc: Edwin Horlings
Subject: RE: New version of NetworkTools


Dear all,


I tried the new version and it comes up with completely different community structures than the old did. I did therefore some testing.

1. Whereas the old version created 18 communities on the highest level, the new gives 50.

2. If the new one is correct, the analyses done with the old are wrong. James, what about chemistry and robotics? Edwin, was the esocialscience analysis done with the old version?

3. the old version gives the same outcomes using the jaccard, roundjaccard and unvalued
4. the new version gives the same for roundjaccard and unvalued, but a different one for valued.
Both seem therefore problematic!!

please comments!

best regards

peter



Previous version (round jaccard)
output stored in table [Articles communities 4]
number of nodes: 14537
number of connections: 555667
number of community levels: 4
community count overview:
level 1 : 14537
level 2 : 132
level 3 : 24
level 4 : 18



Previous version (jaccard)
output stored in table [Articles communities 5]
number of nodes: 14537
number of connections: 555667
number of community levels: 4
community count overview:
level 1 : 14537
level 2 : 132
level 3 : 24
level 4 : 18


Pevious version (unvalued)
output stored in table [Articles communities 9]
number of nodes: 14537
number of connections: 555667
number of community levels: 4
community count overview:
level 1 : 14537
level 2 : 132
level 3 : 24
level 4 : 18

new version (jaccard)
output stored in n [Articles communities 7]
number of nodes: 14537
number of connections: 555667
mber of community levels: 5
community count overview:
level 1 : 14537
level 2 : 1340
level 3 : 117
level 4 : 52
level 5 : 50

• new version (round jaccard)*

output stored in table [Articles communities 8]
number of nodes: 14537
number of connections: 555667
number of community levels: 5
community count overview:
level 1 : 14537
level 2 : 1360
level 3 : 130
level 4 : 71
level 5 : 69

new version (unvalued)
output stored in table [Articles communities 10]
number of nodes: 14537
number of connections: 555667
number of community levels: 5
community count overview:
level 1 : 14537
level 2 : 1360
level 3 : 130
level 4 : 71
level 5 : 69