Visualizing Data

[Boyd Duffee]

I'm starting to get into visualizing data and while Perl has lots of tools for drawing, it's the layouting that I'm concerned about.  There are quite a few modules that will layout a graph (nodes connected by arcs) for you, but one of my data sets has no "connections", just a measure of similarity between each object.  The data set is over 700 points and the similarities have been normalized to [0,1].  I'd like to lay them out to show how they cluster together after I've run a K-means algorithm over the set to label the clusters.  One way I've considered is by creating a fully connected graph from the data set with the arcs weighted by the similarity measure.  Any other suggestions?

Feel free to hijack this thread to discuss any aspect of visualization.  Currently I'm skimming Visualizing Data by Ben Fry, but it focusses mostly on working with the Processing java library.  I've also run across several javascript libraries, of which D3 seemed to have the most modern swagger about it, bragging of it's adoption of standards concepts from SVG and CSS.  I see other books out there on using R and python to visualize.  Other options are Raphaël, Protovis and Graphviz.  Thoughts?

[Joel Berger]

I could be understanding you incorrectly, but it sounds like what you are suggesting is comparable to the cpan-explorer map made to show which cpan authors are most closely linked. On it the thickness of the line connecting two authors is related to the number times that modules released by first author use modules released by the second. Other statistics are displayed in other visual ways too. 

http://cpan-explorer.org/category/authors/

Joel

[Joel Berger]

N.B. there is another fun map of CPAN stuff (unrelated to your question) but I thought that after sharing the cpan-explorer, I should also share MapOfCPAN!

http://mapofcpan.org

[Boyd Duffee]

On Friday, October 12, 2012 9:30:42 PM UTC+1, Joel Berger wrote:

I could be understanding you incorrectly, but it sounds like what you are suggesting is comparable to the cpan-explorer map made to show which cpan authors are most closely linked.

Comparable, but not quite.  The Author map actually is a graph whereas I've got a set of abstract vectors which I can compute the cosine similarity between each two vectors.  My problem is representing them.  The Map of CPAN is an interesing way of displaying data.  Is that like Tessellation?