Orange Forum • View topic - Examples using Parallel coordinates in Orange

Examples using Parallel coordinates in Orange

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

Examples using Parallel coordinates in Orange

Postby colinmac » Tue Jun 12, 2007 23:33

Hiya folks,

After being so impressed by Wigman's Cystal Vision software for viewing my data. And after going through the painful process of get Orange to work (only under Windows alas).

I have a question to the community regarding Oranges capabilities to view data using parallel coordinates (PC). Does anyone know of any documentation or paper which talk about using Orange's PC facilities?

Im also quite new to the data mining game (not counting 2dsingal scatter plots) and would be grateful to anyone for any information, papers or software (links to) regarding the use of PC for visualizing physical data.

many thanks,

Colin.

Postby Gregor Leban » Wed Jun 13, 2007 17:17

Hi,

there is no documentation on parallel coordinates in Orange yet.

Here is a short description of the options:

In the General tab you can select wich attributes you wish to visualize and in what order. You can also select the target class and all the examples with this class will be emphasized. There is "zoom/select" group of buttons on the button, where you determine what happens when you draw a rectangle in the PC graph.

The optimization dialog can be used to find interesting attribute orders. If you choose correlation and press "start optimization" it will try to find such attribute orders, so that the neighboring attributes will have as high correlation as possible. Such attribute orders are interesting because correlation between variables can be itself interesting but also because this improves the visual perception of examples.

In the Settings tab of the widget you can determine the following settings:

- jittering - randomization of values when visualizing the discrete data
- minimum axis distance - (as the name explains)
- show attribute values - show minimum and maximum attribute values on attr. axes
- hide pure examples - when drawing examples from left to right, if all examples that have a specific discrete value have the same class label then stop drawing these examples. This can be useful when you try to create a classification rule for discriminating between classes (otherwise don't use this).
- show splines - instead of lines draw splines - easier to follow examples across the axes
- line tracking - when you move mouse pointer to a point on one axis where a line crosses the axis it will make this line bolder to help you to see its values at different attributes
- show legend - show class legend
- global value scaling - do you wish to scale attributes based on the minimum and maximum value over all attributes?
- statistics - do you wish to see means, deviations, ...
- middle labels - do you want to see the correlation between two neighboring attributes?
- set colors - set color palette used when visualizing examples
auto send selected data - when you select the examples in the graph, do you want to send them to the next widget automatically or only when you press "send selections" tool button on the bottom of the widget.

I don't have any papers on PC for physical data but you can probably find something interesting on Google scolar.

Regards,
Gregor

Postby colinmac » Thu Jun 14, 2007 9:48

Many thanks Gregor.


Return to Questions & Support