Visualization of Large Multivariate Data Sets using Parallel Coordinates
-
Upload
kvaderlipa -
Category
Technology
-
view
326 -
download
0
description
Transcript of Visualization of Large Multivariate Data Sets using Parallel Coordinates
![Page 1: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/1.jpg)
Visualization of Large Multivariate Data Sets using Parallel Coordinates
Ing. Ľuboš TakáčPhD student
Faculty of Management Science and InformaticsUniversity of Žilina
![Page 2: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/2.jpg)
Presentation overview• Visualization
• Parallel Coordinates (PC)
• Large Multivariate Data Sets (LMDS)
• Problem of visualization LMDS
• Solutions
• Developed Software Tool
• Further research and application
![Page 3: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/3.jpg)
Visualization• One of the best approach for presenting data from
PC to human
• Advantages• Global view of data (all in one picture)• Significant features highlighted (and vice versa)• Fast understanding of data by human
• Purpose• Understand raw data• To see some significant characteristics or anomalies which
can be further examined to gain some additional information about raw data.
![Page 4: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/4.jpg)
Parallel Coordinates (parallel axes)• Invented by Maurice d’Ocagne, 1885• Popularized by Alfred Inselberg, 1959
• Easy construction (like ordinary graph)• Data are represented by polylines
• Suitable for visualizing multivariate data (more than 3 dimension)
• You can see dependecies between variables• Distribution per variable
![Page 5: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/5.jpg)
Parallel Coordinates - Construction
Main difference between ordinary graph of function and parallel coordinates is in position of axes.
![Page 6: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/6.jpg)
Parallel Coordinates – Examples
Examples of some 2D functions visualized using parallel coordinates by developed software tool.
![Page 7: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/7.jpg)
Large Multivariate Data Sets• Collection of data usually presented in tabular
form
Multivariate data set of movies with 7 dimensions.
![Page 8: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/8.jpg)
Problem of visualization LMDS• Records overlapping• by simply painting records should be overlapped, you
loose some information
• Overlapping the same records• by simply painting you do not see the difference between
overlapping two or hundreds same records
• Too many records to visualize => one big blur• imagine resolution 1024x768, ten thousand of records
uniformly distributed over axes (height 768 px means about 13 records per pixel)
![Page 9: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/9.jpg)
Problem of visualization LMDS
Problem of overlapping painted records.
![Page 10: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/10.jpg)
Possible Solutions
• Preprocessing data before visualization
• Paint data sophisticated by Alpha Compositing
![Page 11: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/11.jpg)
Alpha compositing• Computer graphics painting method which use
alpha channel to define each color (alpha channel – transparency of color)
• If you paint object with non opaque color, the resulting color depends on background too
http://en.wikipedia.org/wiki/Alpha_compositing
![Page 12: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/12.jpg)
Implemented solution
Visualizing the same randomly generated multivariate data sets by opaque color (upper image) and using alpha compositing technique (right image).
![Page 13: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/13.jpg)
Developed software tool
• Based on mentioned principles• Interactive analyzing of LMDS• Interactive set operation (selection, difference,
intersection …)• High quality, antialiased image• Data import from text file• Record count is limited to hundreds thousand at
rs. 1920x1080
![Page 14: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/14.jpg)
Developed software tool
Demonstration of developed software tool. Visualized data sets come from IMDb (Internet Movie Database).
![Page 15: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/15.jpg)
Further research and application• Tool can help decision makers and data analyst to
gain some added information to do better decisions.
• Medical data• Scholar data
![Page 16: Visualization of Large Multivariate Data Sets using Parallel Coordinates](https://reader036.fdocuments.in/reader036/viewer/2022062419/55850647d8b42ad71b8b525b/html5/thumbnails/16.jpg)
Thank you for your [email protected]