|
University of Jyväskylä
Department of Mathematical Information Technology
NDA - Neural Data Analysis environment
|
SOM Based Data Analysis
The best support of the NDA is addressed to the SOM based data analysis. Thus, the NDA has been aimed
-
To provide a strong support to the SOM based data analysis, visualization and decision making.
-
To facilitate human interpretations in data analysis through interactive and visual tools.
-
To utilize the SOM based data analysis in different applications.
The NDA uses the Tree-Structured Self-Organizing Map (TS-SOM) as a basic data reduction method. It provides some useful properties for interactive data analysis:
- It decreases time needed to train the neural network
- It produces good results with fixed default parameters
- One training produces several SOM results with different resolutions
A Sample Analysis Process
For instance, the user could select the following operations to perform the basic SOM based analysis. We have also listed the NDA commands needed to perform the phases of the analysis.
NDA> somtr -d predata -sout som -l 5
NDA> somcl -c som -d predata -cout cld
Select any data for visualization and compute statistics for neurons.
NDA> select visdata -d trdata
NDA> select visdata -d boston.price boston.lstat boston.indus
NDA> clstat -c cld -d visdata -dout clu_stat -avg -hits -var -min -max
Create a graphical structure for the SOM and connect data clustering and cluster statistics
NDA> mkgrp graph -s som
NDA> setgdat graph -d clu_stat
NDA> setgcld graph -c cld
Show the graphics and use commands or an interactive visualization tool to set the visual presentation:
Select groups of the neurons and give names for them. The selection can be supported by different views to the trained SOM, groups and data.
Use results, for instance:
- Pick data records from specified groups for more detailed analysis
- Code groups for later analysis by using binary or fuzzy variables
Back to SECI homepage
This page was last modified on May 31st, 2000 by Erkki Häkkinen