Sunday, August 25, 2013

Robustness of a network

The following graph is the final network model state as discussed in my previous blog (http://businessfractal.blogspot.com/2013/08/to-rent-or-buy-dashboard-analytical.html), where I indicated the ability to determine the robustness of a network from a network simulation (http://youtu.be/oBb5RjorMP8).


 
The key statistics of the network can be found as being:

Number of Vertices: 303

Number of Edges: 326

Diameter of network: 13

Mean Degree of the network: 2.151815

Maximum Degree of the network: 32

Mean Geodesic of the network: 5.136059

Density of the network : 0.0071017

Transitivity of the network:0.003640777



Robustness of the network can be tested in two ways, first by attacking the highest connected hubs in a structured manner, and then by removing nodes in a random manner. The simplest measure to test this impact is the average distance (mean geodesic) of the network. The following graphs shows a few simulations by removing 55% of nodes in the network. The blue line shows the failure in the average path if a structured attack is carried out, with the red lines resulting from random attacks. From a structured perspective this network is highly fragile, failing in less than 1% of node removals, where-as the network is quite robust when being attack in a random manner. Whether this is good or bad naturally depends on the application; if we want to design a marketing campaign, then we might rather use a structured approach to implement it, as apposed to a random approach. If it is a supply chain network, the structured attack will be of high concern from a risk management perspective.

 
Finally, analysis of the network model indicates that this model is a scale-free network, as the connection degree do follow a power-law behaviour, which does indicates the reason for the network fragility under structured and random attacks.


 

 

 


Thursday, August 22, 2013

TO RENT OR BUY DASHBOARD ANALYTICAL TOOLS FOR +10000-20000 EURO PER ANNUM ?

I just evaluated a fancy dashboard/analytical toolset in my search for solving a specific problem; very nice to drag and drop stuff and get some fancy pictures out.

I then received the quote for the perpetual/annual licence agreement/pay-me-as-much-as-possible-for-fancy-graphs.

Got a bit green when I saw the price and then to calm myself I went and grab a few lines of brilliant opensource R code from a blog to do a dynamic network model (I am busy trying to calculate the diffusion rates in market segments from complex network structures).

Here is the result: http://youtu.be/oBb5RjorMP8

The next step will be to insert 3 lines of R code to calculate and model a random attack on the network to see the level of network robustness. And add another 3 lines to model a structured attack on the key nodes and understand the overall fragility of the design. Then plot it all with R ggplot which is free.

Advice asked & advice given is free.

Can check all code and see all results, can also mingle a bit of Monte Carlo simulation into it as well; it still stays free. It feels like a good deal to me, especially having the freedom to do as I like.

Don't see how a scientist or engineer can work with analytical tools which generates by "magic"
answers which cannot be verified - too restrictive and dangerous for me. Not to get it wrong, there are some very specialist good ones out there like for example the Disco toolset from www.Fluxicon.com , but then at least I can read the Phd's from the founders and understand what the
wiring is all about... that's cool.

I still think that a fool with a tool is still a fool - even if the tool is Visio or Powerpoint or Excel.

Popular Posts