Subgroup Analyzer

Description

Subgroup Analyzer uses Expectation Maximizatio algorithm trying to find 2,3 or 4 gaussian distributed subgroups in data sets. If e.g. input are height values of many people, button 2 probably will show seperated distributions of women and man graphically and numerically.

To do this EM algorithm starts with most of time bad estimations of first gaussian distributions. For every input value it is calculated then it belongs more to the "bad" or the "worse" estimation. For searching two subgroups first for all data values (e.g 0.7) the values of the distributions (e.g. G1(0.7)=0.5 and G2(0.7)=0.1) are calculated. Then new deviations are calculated with weightened values (e.g. 0.7 ist 5 times stronger weightend in G1 as in G2. With this new, weightened values new parameters for distributions (G1, G2) are calculated. Then for all data values (e.g 0.7) the values of the distributions (e.g. G1(0.8)=0.5 and G2(0.7)=0.02) are calculated again for next loop of algorithm...and so on, until the new calculated parameters are nearly the same as the older ones.

Subgroup Analyzer is very simple to use. Just enter your data as a list of values via:

KEYBOARD: Select board in menu, touch click under button 4 and enter one value a line. Use clipboard to import/export data from or to other apps. Too use keyboard to edit data.

MICROFONE: If keyboard is shown, click on microfone symbol und speak in your values seperated by saying „newline“ or „neue Zeile“ (german board).

BUTTON (single mode): Chose with down left button, press START and input continuous data like typing Morse-code SOS.


BUTTON (double mode): Chose down left button, press START and finish with STOP to input one time measurement value.

OUTPUT:
Buttons 2,3,4 starts searching 2,3 or 4 gaussian distributions in input data. If single loop is selected multiple touches are needed to continue calculation
Plot shows input data as black lines on horizontal axis. Input average is shown as white mark. Red deviation always has largest area, then yellow, green and blue deviation. Change apperance of plot by touching it. If set on in background the sum of differences α, μ and σ between loops are showed to visualize convergation of calculation. A grid can be set off or on via menu.

Result window shows area α, position μ and standard deviation σ of gausssian distributions in corresponding colors. Touching result window changes range in plot window too.