The GPA is based on Commandeur (1991), and it is intended to include PINDIS (Commandeur calls it MATCHALS) in the near future.
Efficient MDS algorithms are implemented through the combination of clustering, sampling and layout techniques. To date, the best model runs in O(N^5/4). The code for the FSMvis tool is freely available. The tool generates 2D scatterplot layouts of multidimensional data. Implemented in the available version are a novel hybrid O(N root(N)) model, Chalmers' 1996 O(N*N) algorithm and the standard canonical O(N*N*N) spring model.
There is now a web page for the Locally Linear Embedding Algorithm (LLE) recently proposed by Roweis and Saul.
The web page (below) contains a detailed pseudocode description of the algorithm, some actual code (in MATLAB), example pictures and animations, and references to publications and related work. It also contains a draft of an introductory LLE paper which has a more detailed description of the algorithm and derivations of the equations.
A "wrapper" Matlab function to call Forrest Young's ALSCAL program from Matlab in a transparent way. Written by Sylvain Choisel. It may not be totally foolproof, but could be useful to some.
Free downloads of the software are available.
A free download of the software is available.
PERMAP is a free, Windows-based (Win 95, 98, ME, NT, 2000, XP), real-time interactive program for making perceptual maps (also called product maps, strategic maps, sociograms, sociometric maps, psychometric maps, stimulus-response maps, relationship maps, concept maps, etc.). Its fundamental purpose is to uncover any "hidden structure" that might be residing in a complex data set. PERMAP takes object-to-object proximity values (similarities, dissimilarities, correlations, distances, interactions, psychological distances, dependencies, confusabilities, preferences, joint or conditional probabilities, etc.), or object attribute values, and uses multidimensional scaling (MDS) to make a map that shows the relationships between the objects. Succinctly, it makes classical metric and nonmetric MDS analyses in one, two, three, or four dimensions, for one-mode two-way or two-mode two-way data, with up to 1000 objects and with missing values allowed. In addition, it can make several new types of MDS analyses involving error bounds or boundary conditions, and it can show the affect of degrading the similarity information.
Probabilistic Scaling (PROSCAL) is a powerful and flexible analytical tool for representing and understanding complex objects. The power of PROSCAL comes from its sophisticated modelling of how subjects perceive, prefer and choose objects. PROSCAL models objects, such as consumer packaged goods, and subjects (market segments) as distributions in a multidimensional space. The distributions measure the differences that exist in the certainty with which consumers view different products and attributes. Probabilistic modelling allows PROSCAL to capture the complexity of perceptions and test hypotheses that provide understanding of how consumers view products and make product decisions. By realistically modelling the richness of consumers' cognitive processes, high quality estimates are obtained.
The relational perspective map was developed by James X. Li [James Xinzhi Li: Visualization of High Dimensional Data with Relational Perspective Map. Information Visualization 2004, Vol. 3, No. 1. 49-59] . It is a general purpose method to visualize distance information of data points in high dimensional spaces.
The software, VisuMap 1.5 Standard , is dedicated to exploring high dimensional data. It offers the most comprehensive implementation of Relational Perspective Map. It is free for no-commercial use.
Last amended 12-December-2004