Venn Diagrams & Multivariate Boolean Analysis

[TheFaithfulStone]

Okay, so I'm helping the wife with a project and I need a little help.

 

We have a Data Set consisting of approximately 500 items which may (or may not) have some number of traits.

 

I don't want to give away too much of the specifics, because she'd be pissed off at me if I disclosed too much of the research specifics, as part of the data is... not classified, perse ... but private.

 

So, I'm using people as an example - even though that's not really the data set.

 

We have

 

Bob: Who has blue eyes, brown hair, a large nose, and light skin, freckled skin.

Sue: Blue eyes, blonde hair, small nose and light skin, unfreckled skin.

Greg: Green eyes, red hair, large nose and light, unfreckled skin.

Dave: Brown eyes, brown hair, large nose and dark, unfreckled skin.

 

Now, the best way to show the relationship between all of these people since there is no dependent variable is to use a Venn / Euler diagram. Although, in this abbreviated data set you wouldn't find it for instance, that's how you find out that people with blond hair rarely have freckles, and that dark skinned folks have brown eyes and brown hair, since the overlap is small.

 

And this works great with smaller data sets of less than about 100 or so. However, when you get up into those larger areas it becomes increasing difficult to represent all of the relationships with a Venn/Euler diagram. Of course, if we had 200 items, we can easily drop all sets with a membership of less than two individuals as being less significant than more numerous ones, but we still wind up with having difficulty representing all significant overlaps. By way of example, try drawing the nose size or freckles on that diagram using only circles!

 

Currently, we are using a program called "VennMaster" which is for doing this kind of analysis on genetic data, (I belive it called "synteny" in this sense) and it's working pretty well. However, the well-formedness rules for Venn diagrams specify that each proportional shape must be perfectly round, which is why it's difficult to represent the data accurately. VennMaster doesn't do Venns/Eulers with relaxed well-formedness rules. Does anybody know of a program that does?

 

Now, we realize that this is ideally suited for Bayesian networking / decision tree analysis work, but the statistical methods employed in the field that she works in make it ... problematic to get too in depth with those kinds of statistics and machine learning problems. She's going to address it, and talk briefly about using those kinds of methods in the future, but even looking at her data set in this way is a big step. The Venn Diagrams are a huge improvement, and it's already going to be an uphill battle to get them adopted as a valid way of looking at things.

 

In the example above for instance, the CW on this would be to say "Bob's Dutch, Sue's Swedish, and Greg is Irish, and Dave is Zulu" and leave it at that. It doesn't tell you much about the similarities of Bob, Sue, or Greg, or how those difference might have arisen, or what the relationship between Dutch, Swedish, Irish, and Zulu is. It doesn't tell you that all of those people live in the same neighborhood.

 

So anyway, the questions are:

 

1) Anybody know of a Euler / Venn program with relaxed well formedness rules that could help draw this out?

 

2) Anybody know of a good BASIC introduction to Set Theory / Bayesian networking stuff.

 

3) Any tips on touching on decision tree anaylsis and influence diagrams here? They seem like they could be well applied to this problem. (For instance, skin color is more strongly correlated with eye and hair color.)

 

TFS

 

[aside: It seems that the idea of a directed acyclic graph and Venn Diagrams are largely the same, but that they show the relationship in a different way - with the dag showing a predictive model, while the Venn is primarily descriptive? see here.]

[InfiniteNow]

For diagram stuff, check here:

 

Diagrams (Stanford Encyclopedia of Philosophy)

 

For software, check here:

 

Easy Venn Diagram Software - Quickly and Easily Create Venn Diagrams, Marketing Charts, and More. Download our Venn Diagramming Software FREE.

 

For reasoning, check here:

 

http://www.it.bton.ac.uk/Research/vmg/papers/EulerDiagramReasoning.pdf

 

 

For someone who actually understands this stuff, and is not just good at googling, ask another member.

 

 

:shrug:

[InfiniteNow]

Hmmm... have you considered adapting the 2D data above into a 3-dimensional Venn/Euler?

 

 

Example:

 

RasMol V2.6-beta-2 Reference Manual

 

 

[TheFaithfulStone]

I don't know how that Venn-Diagram was drawn, but that's basically exactly what I need! Unfortunately, the "RasMol" program is a molecular modeler, and as far as I can tell, doesn't include the functionality to draw diagrams like that. Too bad - cause that's almost EXACTLY what I want.

 

SmartDraw (which I have) draws Venns, but doesn't construct them from data-sets, so it's not really going to help.

 

Thanks for the help though, now I just gotta find this Kurt Giles cat....

 

TFS

[InfiniteNow]

I don't know how that Venn-Diagram was drawn, but that's basically exactly what I need! Unfortunately, the "RasMol" program is a molecular modeler, and as far as I can tell, doesn't include the functionality to draw diagrams like that. Too bad - cause that's almost EXACTLY what I want.

 

SmartDraw (which I have) draws Venns, but doesn't construct them from data-sets, so it's not really going to help.

I think the following can be downloaded for free. It's from Nevron, and is a demo, so may not have full functionality, but is worth a shot (I think the "Chart" app is what you're after):

 

Downloads / Downloads

 

 

 

Full application which helped me find this:

 

Nevron Chart for .NET (Windows Forms and ASP.NET) - Enterprise - Summary

Add 3D charts to your applications and Web pages. Nevron Chart for .NET is a powerful, flexible and extensible charting component for Windows Forms and ASP.NET. It fully supports Visual Studio 2005 and .NET Framework 2.0. It aims to provide your applications with data visualization capabilities which will significantly improve the looks and presentation of your product. Nevron Chart for .NET combines the power of raster and vector graphics and employs state of art 2D and 3D rendering technologies. The Enterprise Edition contains everything included in the Professional Edition plus Polar, Radar, Shape, Mesh Surface, Venn, XYZ graphs and more. Includes both Visual Studio 2003 and Visual Studio 2005 compatible components.

 

Full Enterprise version, likely way more than you've ever considered spending, at the below:

Nevron Chart for .NET (Windows Forms and ASP.NET) - Enterprise

[InfiniteNow]

The last post I made is probably more appropriate to your need, however...

 

 

Holy crap balls! I've just stumbled into the world of molecular chemical modelling and animation. This is freakin awesome!!!

 

I don't know if this does or does not cover your "Jonny has blue eyes and brown hair and mocha skin, but Sally has brown eyes, blond hair, and albino" example above, but if you can use this, I'd love to see the output! :shrug:

 

Download/Install Protein Explorer

 

 

Check out the demo. I just played around with a 3D molecule for like 20 minutes. It's really sweet.

 

Protein Explorer FrontDoor

[TheFaithfulStone]

Hmm... my browser won't run it.

 

TFS