Purpose
This page gives access to various versions of our interactive GUI for viewing data and understanding the new data representation called Geometric Wavelets (GWT for Geometric Wavelet Transform), developed by Mauro Maggioni's research group at Duke University.
The data analysis is done in Matlab, and the results are saved for loading into the GUI. So, on this page, you'll find a couple different binary versions of the application, along with data (.mat) files to be downloaded. Be aware that the data format has changed over time, so you need to use appropriate data files for a given version of the GUI (those listed in the same section as the GUI version).
Resources
Soon there will be a formal publication about the Geometric Wavelets technique, but in the meantime you can take a look at the two-page paper for our poster at the IEEE VAST 2010 conference. We also have a short movie demonstrating the GUI interaction.
- VAST 2010 two-page paper (PDF)
- Interaction demo movie (QuickTime)
- VAST 2010 Poster (PDF)
- VAST 2010 Poster Handout (2-up PDF)
- VAST 2010 Poster Fast-Forward Slide (PPT)
Code
The GUI is programmed in Python using PyQt4 and the VTK Python wrappings. The source code is available at my github repository. More specifically, the GUI code can be found in the MultiScaleSVD repository. The custom VTK classes code can be found in my vtkVTG repository. Building the custom VTK classes requires a build of the VTK 5.10 built with Qt4 support, Python wrappings and the Python SIP wrappings. Details on downloading the VTK source code can be found on Kitware's VTK web site.
Most Recent 64-bit Scientific Linux GUI Version
This is a first try at a Scientific Linux version. Beyond the OS X version below, the icicle (tree) view also gets colored by category labels. After untarring the package, you need to go into the directory called main and run the executable called main (sorry about the horrible temporary names). This should pop up a GUI asking you to choose one of the .mat files you need to download in the Data section (directly) below.
Most Recent Mac GUI Version
This version adds Wordle (e.g. www.wordle.net) generation for representing text-space centers and directions. NOTE: The wordle generation is fast, but there is still a noticeable delay when many must be created for leaf nodes, and they are currently not cached. There is no Fedora version for this generation of the GUI yet.
Data
This is a completely new data format for this generation of the GUI. It also includes new data for text representations. Science News in particular now has titles of articles included which are used as tooltips in the scatter plots.
- MNIST digits: 1000 ea of 1,2
- MNIST digits: 1000 ea of 3,5,6,9
- Science News "magic" set (1147 docs in 8 categories)
- 20 Newsgroups: atheism/religion split
- 20 Newsgroups: graphics/X split
- 20 Newsgroups: MS Windows/PC hardware split
- 20 Newsgroups: baseball/cryptography split
- YaleB faces (data, bases downsampled x10 for the GUI, 258 Mb)
- Frey faces
- Olivetti faces
- Natural image patches
- S curve manifold
Diffusion Embedding Viewer
I've been working on a view for the main GUI to explore the diffusion embedding for a data set. This separate app shows where I am with that effort, and I thought it was worth sharing.
The slider moves you through the dimensions of the Eigenvectors, and the combo box lets you choose between a couple options for what path the projection takes between dimensions. (Diagonal and Alternating Axes are the most useful, Simple is there as a test.) You can open up any of the recent GUI data sets (color by label versions), but I'll include a substitute one below for YaleB since for this app we only need the EigenVecs, EigenVals and Labels, and so there's just an unnecessary delay if you load the whole YaleB set just for these few matrices. The axis labels are confusing and not extremely informative, but I haven't figured out yet how to represent which dimensions are being shown on each axis. There is only an OS X version for now.
Data
As stated above, you can use any of the recent color-by-array data sets in this GUI, but if you are looking at the YaleB set, you might want this much smaller version instead, just to decrease load time.
Older GWT GUI Versions Archive
The GWT GUI versions listed below are for mostly available as an archive of past versions for the developers. This is for the off-chance when you are at a conference demoing the software and find you're having trouble with a recent version, you can at least grab an older one. Not likely to happen, but you know, just in case...
Third Gen Category Colors GWT GUI Version
This version includes coloring scatter and parallel coordinates plots by category (submenu under the View menu). Only the OS X version has had a bug fix applied (v.color_0725), so the Fedora version may be not be super-stable, but it's generally useable. The Fedora version may also have problems with colored points in the scatter plot (or it may be a problem with my Fedora VM).
- OS X 10.6 app bundle (64-bit, v.color_0725, zipped)
- Fedora 13 binary directory (32-bit, v.color_0104, tar.gz)
Data
These are new sets which include the category label. NOTE: Some of these names are the same as they were when I posted an initial version of the color by array app, but they've been changed slightly so please download again. (They were also generated with the 1119 version of the GWT code.) These are all old-style .mat files generated with Matlab R2009b.
Stable Second-Gen GWT GUI Version
This older version of the GUI can handle multi-scale trees which have variable dimensionality. In the 10/22/2010 version I have also added options for the parallel coordinates display viewing different combinations of scales, and indicating which dimensions are being plotted on the XY scatterplot.
Data
These newer data sets are smaller than the old ones since we're now saving a more minimal version of the data. The data structures are different than the originals, too, so you won't be able to load the old data into the new GUI, and vice versa. These are all old-style .mat files generated with Matlab R2009b.
Oldest GWT GUI Version
This old version of the application worked only for fixed-dimensionality analysis. There are versions for OS X and Fedora 13 (32-bit). For the latter, run the "main" executable in the expanded mssvd_gui directory.
- OS X 10.6 (Snow Leopard) app bundle (64-bit, zipped)
- Fedora 13 directory with binary executable (32-bit, tar.gz)
Data
As was stated above, the data sets for the old version will not work with the newer version of the GUI. These files have fixed-dimensionality on the analysis, and different data structures. These are all old-style .mat files generated with Matlab R2009b.