LineUp should support a "count" mode for homogeneous attributes

[ngehlenborg]

Sometimes I want to look at homogeneous attributes (e.g. counts for each item but spread out across multiple dimensions, e.g. years, days of the week, experimental conditions, etc.) and in those cases the scaling applied in combined columns is very confusing (because the default mapping will map the max of each column to 1, no matter what the max values are in other columns).

If LineUp would support a count mode and appropriate scaling/mapping techniques (e.g. scale to total per row, scale to max per column, scale to max per row, etc.) that can be easily applied to all columns of a combined column, it would be much more intuitive to work with such data in LineUp. The default behavior that assumes a different data type or range for each column easily leads to confusion.

Here is an example data set: https://gist.github.com/anonymous/836fe14223dc29d7d63837ceb3c49e25
http://www.caleydo.org/lineup.js/demo/#gist:836fe14223dc29d7d63837ceb3c49e25

[sgratzl]

The default behavior that assumes a different data type or range for each column easily leads to confusion.

In LineUp, columns are independent of each other having its own definition, e.g. mapping in case of a numerical columns. LineUp itself just takes the column definitions and the data and visualizes them. If you have multiple columns that should have the same domain, you have to define it or improve the demo import process.

[ngehlenborg]

What I am proposing is an option to apply a particular column definition to all (or a subset of) columns and to integrate support for a couple of common tasks: e.g. scale to total per row, scale to max per column, scale to max per row, etc.. Right now it is possible but extremely complicated and error prone.

In essence, this would require column definition functionality independent of a particular column (but still data type-specific), that offers the option to assign the definition to multiple columns of the same type.

[alexsb]

Bertifier does this very elegantly.

[ngehlenborg]

Can you clarify or explain how it is done there?

[alexsb]

I can write two paragraphs, or you can watch this video:

https://www.youtube.com/watch?v=tJxAF_a_yBQ#t=61

Crossset interaction technique.

[ngehlenborg]

Ah. Great feature but incredibly hard to figure this out without having seen the video. I got annoyed after 3 minutes that I had to click each row separately ...