perform multidimensional scaling on a difference matrix

This is a stand-alone version of the functions cmdscale, isoMDS, and sammon available in R.

mds [-K | -S] [-i filename] [-m float] [-n int] [-o filename] [-t float] dimensions difference-table-file

-K

use Kruskal's method

-S

use Sammon mapping

-i filename

initialise from file (with -K or -S only)

-m float

initial magic value (with -S only, default: 0.2)

-n int

maximum number of iterations (with -K or -S only, default: 50 with -K; 100 with -S)

-o filename

write results to file (default: <stdout>)

-t float

tolerance (with -S only, default: 0.0001)

The program will abort if a file _CANCEL_.L04 exists in the current directory, or if it is created while the program is running. This is useful for stopping long running calculations from a GUI, such as pyL04.

The input file is a difference matrix file.

The output file, as well as an optional initialisation file, is a vector file.

This command performs Classical Multidimensional Scaling into 2 dimensions on the difference matrix stored in the file difmatrix and stores the result in the file results:

    mds 2 difmatrix > results

This command performs Kruskal's Non-metric Multidimensional Scaling into 3 dimensions on the difference matrix stored in the file difmatrix and stores the result in the file results3:

    mds -K 3 difmatrix > results3

By default, the program does Classical Multidimensional Scaling.

With the option -K the program does Kruskal's Non-metric Multidimensional Scaling.

With the option -S the program does Sammon's Non-Linear Mapping.

Kruskal's Method and Sammon Mapping require a pre-initialisation. By default, this is done by Classical Multidimensional Scaling. However, you can use the result of a previous run of the program as an initialisation. For example:

    mds -o tmp -S 3 difmatrix
    mds -o result -i tmp -K 3 difmatrix

In this case, initialisation is first done by Classical Multidimensional Scaling, then a Sammon Mapping is performed, and then this Sammon Mapping is used as the initialisation for doing Kruskal's Method.