Hi,

  Circos will be benificial for you while explorating data analysis!

Regards

Muddsair

Dear friend

Greetings.

There are several methods available depending on the purpose that you are seeking.

You will get lot of options in MATLAB and you will really enjoy.

You check out the options in 2014 versions, there are plenty of options.

I have used several visualization methods in my papers that I have published.

Best regards

Dr.Indrajit Mandal

You can try Kohonen Maps. In MATLAB you can use the SOMTOOLBOX to show your n-dimentional data in 2D projections provided by U-Matrix. However, it is not a big task to implement U-Matrix in other languages like Python or even in C.

First, you must understand your data... what is in there?  What are the nodes?  the links?  What is the cutoff point for a visible object (node or link)?  Then, extract sub-networks to show underlying patterns, which are easier to see.    

I wrote a blog post about this: http://www.thenetworkthinkers.com/2013/03/big-data.html

The big hair ball graph in the post above includes all book sold by Amazon during a specific period of time.  The links are the "also-bought" info Amazon reveals for each book.  The sub networks are books of a certain type/topic -- we filtered by node attributes.  We can also filter by link attributes to show different sub-networks. All sub-networks show interesting communities/clusters.

Christian... following on what Valdis said.... only maybe more strongly. Don't visualize big networks. It's not just that visualizing graphs at scale (e.g. > 500k nodes) is hard, there's often little to be gained for doing it. There are very small number of specific kinds of questions that might be answered at scale and these typically have to do with either looking for structure / patterns in unknown data. If that's what you're doing...go for it. The reality is though that at this kind of scale you lose a lot of structure and detail as well.  You'll have to work hard to make sure that your rendering algorithm prioritizes the salience of the features you think are important.  The specific visual rendering is secondary to that.  Typically a much better option is, as Valdis said, to pull out sub-networks, do mathematical clustering, attribute based filtering, whatever you need to do to get to the parts of the graph that are important. At that point other kinds of visualizations are more useful.  What are you trying to do?

@Jack: I am just trying to collect alternative methods for network visualization that work better on larger networks than the usual node-link diagram with a force layout. "Don't bother to draw large networks" may be a valid answer, but before coming to that conclusion I'd like to see which methods are out there.

Christian. What kind of networks are you interested in? How large is large? I usually (and arbitrarily) think of 500,000 nodes as my cut off for a "large" network. Are there any particular characteristics to your large network. Are they randomly distributed networks or do they follow scale-free (or similar) power law distributions?  Is this a pure mathematical network representation (e.g. you only have nodes and links) or do you have a useful attribute set. Are you open to applying mathematical or logical operations on the network prior to visualization or do you want to work with the raw network? Any particular cognitive or visual operations you'd like to be able to perform on the network?  

I realize you're asking a general lit search kind of question, but knowing some details would be helpful.  I've got some notes on different kinds of large graph representations. I'll try to dig them out later today or tonight.  

Yes, Jack!  You put it so simply ...

Don't visualize big networks.

I am in violent agreement!

It seems to be a macho thing -- "I can analyze bigger networks than you".  This statement does not reveal computing or algorithmic prowess of the claimant, instead it reveals ignorance about how human networks function.

@Jack: I would say "large" as in "too large for a traditional drawing" starts at hundreds of edges. Interesting networks easily have millions of edges. They are real-world complex networks, which typically have a skewed degree distribution and a low-diameter. Hence they are very prone to hairballyness in force-directed node-link drawings. They may have attributes, but the main focus is on exposing significant relational structures. I'm open to operations like coarsening and filtering, but these should be elements of the visualization method rather than just a preprocessing step that extracts a smaller subnetwork to then draw it traditionally.

@Valdis: I'm interested in domain-agnostic methods, these networks are not necessarily human.

Dear Christian,

I think it's not so straightforward to suggest the ''best'' way to visualize a complex network (that would solve a lot of our problems wouldn't it?).

Some times modeling using subnetworks or smaller representative networks that carry vital information works better in understanding the big picture than trying to visualize everything (and static pictures with thousands nodes can only lead to hairballs).

Also, other solutions like the hive plots might be useful but they have been created for a reason which might not be applied in any kind of complex network visualization adequately and that has to do with the domain you're working on and what questions you want to be able to ask to the visualized network to get answers.

So as others said, a careful investigation/analysis of your data will reveal the dynamics of the data and the underlying invisible (currently) information and this will help you identify which visualization method works better in your case.

Friendly,

Christos Vaitsis

Dear Christian,

Perhaps by looking at the degree of invariance (in the visual patterns) as one collapses higher dimensions into lower ones. For this to be useful one needs to determine the underlying generative functions for both the higher and lower dimension systems. I find Wolfram's observations in Chapter 10 of his A NEW KIND OF SCIENCE- helpful. Not so much in the technical sense (i.e. how to) but in the sense of providing guiding principles for analyzing complex systems. He seems to agree on at least one point with you and me that visualization is the way forward in a data-abundant world.

http://www.wolframscience.com/nksonline/toc.html

I think the discussion can be broken down into two subcomponent:

1) like Jack said, understanding the data may help and use Shian's suggestion  in minimizing the complexity by focusing on the relevant aspects of the information displayed on the graph.

2) The display part of the problem which can take advantage of:

• Color
• Edge and node shape (width, length, etc.)
• Clustering of components
• type of graph display- tree, bipartite,hypergraph, etc.