Orange Forum • View topic - Need help with visualisation

Need help with visualisation

A place to ask questions about methods in Orange and how they are used and other general support.

Need help with visualisation

Postby shimrod » Thu Jan 25, 2007 0:00

I need to do a short presentation on Orange with emphasis on its visualisation methods, but as a complete newbie, I could use some help.

Specifically, what are the differences between and the intended applications of Radviz and Polyviz?

Also, am I correct in interpreting the Sieve Multigram as visualisation of correlation between discrete attributes? With blue lines indicating positive correlation (red: negative) and thick, strong colored lines indicating strong correlation (thin, pale: weak)? What does it mean when both lines indicating opposite correlations are equally strong (for example eggs 0 - hair 0 and eggs 0 - hair 1 in the zoo set)?

Postby Gregor Leban » Sat Jan 27, 2007 12:25

Hi,

if you would like a survey of visualization methods describing their advantages and shortcomings you can look at this paper: http://home.comcast.net/~patrick.hoffman/viz/MIV-datamining.pdf

Regarding the polyviz-radviz question. Both methods are very similar. The difference is only that polyviz shows attributes as edges of a polygon while radviz shows attributes as points on the unit circle. In polyviz we can therefore also see the distribution of examples along each visualized attribute.

Sieve multigram is my addition and you are interpreting it correctly. If both lines are equally strong this indicates the same strength of correlation (but with different sign, if they are of different color). Sieve multigram is just a simple extension to sieve diagram, that can only show correlations between two attributes at the same time. If you visualize your pair of attributes (eggs and hair) in sieve diagram, I'm sure that you will understand the relation between the two attributes.

Btw, if you are doing a presentation on Orange and its visualization methods, you might also check out VizRank, that can automatically identify interesting projections with good class separation and is implemented in Orange. Another method which is also implemented in Orange and can find interesting visualizations with good class separation is FreeViz. You can find papers on both methods using google.

Hope this helps. If you need more help, just let me know.

Gregor


Return to Questions & Support