[Swarm-Modelling] re re diabetes modelling

[Pieter Steenekamp]

Hi,

 

A combined reply to Paul Johnston who wrote:

> I'm not doing that kind of thing right now, but i'd be very interested
> to hear more on how you implement it.
 and Gulyas Laszlo who wrote:

> The most famous algorithm to create scale-free networks (rather: networks
> with power law degree distribution) is the Preferential Attachment model
> of Albert and Barabasi. Try Google, you'll be successful.

 

My choice of wanting to use a scale-free network has been influenced by the book "Linked" by Albert-Laszlo Barabasi, who is Professor of physics at the University of Notre Dame. (Are the Laszlo in Gulyas Laszlo and Albert-Laszlo Barabasi linked?). I realise that there are different algorithms to generate a scale free model, my current requirement is just to generate a network with static links, following the power law distribution, so I think a simpler algorithm than the Preferential Attachment model will do.

 

I haven't programmed it in SWARM as yet, but having played with the following in Excel seemed to give a reasonably good (for my application at least) power law distribution:

If you have a list of people, and now you want to generate a number "numLinks", for each person in your list, that is the number of other people in the list influencing her, then use the following:

numLinks = a * exp [ b/(c + x)]

with x = d + e*f

with f = randomly distributed number between 0 and 1

a,b,c,d & e are adjustable coefficients giving different power law distributions.

A set of values that I have tried is:

a = 10

b = 20

c = 4

d = 5

e = 195

 

Pieter Steenekamp