| classical | Scale a data set from a greater to a lower dimension by using the PCA (Principal Component Analysis) |
| -d <data> | name of the input data frame |
| -dout <dataout> | name of the output data frame |
| -dim <dim> | dimension for the output data |
| [-pcout <pcdata>] | saves eigenvalues and eigenvectors of covariance matrix |
| sammon | Scale data from a greater to a lower dimension by using the Sammon's mapping |
| -d <data> | name of the input data frame |
| -dout <dataout> | name of the output data frame |
| -dim <dim> | dimension for output data (default 2) |
| [-eps <stop>] | stopping criterion (default 0.0) |
| [-ef <efile>] | write error in every iteration to a file and window |
| [-imax <imax>] | maximum iteration steps (default 30) |
| [-dinit <inidata>] | initial values for the output data |
These multidimensional scaling commands seek to preserve some notions of the geometric structure of the original data, while reducing dimensionality. The points, which lie close to each other in the input space will appear similarly close in the output space. These two commands use different methods for preserving the intrinsic structure of the data. Classical scaling tries to map data points in an optimal fashion to the output space by using a principal component analysis. The Sammon's mapping maps data points to the output space by minimizing the distance difference between data points in the input and output spaces.
Example (ex5.13): The Sammon's mapping and classical scaling.
...
NDA> prepro -d boston -dout preboston -e -n
NDA> somtr -d preboston -sout somboston -l 5 -cout cluboston
...
# Pick weights
NDA> psl -s somboston -dout grid3 -l 3 -wout w3
# Sammon mapping
NDA> sammon -dinit grid3 -d w3 -dout out3 -imax 30 -eps 0.01
-ef virhe
# Setting transformed values for SOM
NDA> ssl -s somboston -sset out3 -sout trans -l 3
# The same for the classical scaling
NDA> classical -d w3 -dout class_out -dim 2
NDA> ssl -s somboston -sset class_out -sout class_weights -l 3
...
