mdsjs

Projections for JavaScript

A library for dimensionality reducing projections for JavaScript. So far, PCA and MDS are supported.

Refer to index.html for a small example of how to use the library. Pull requests are highly appreciated.

PCA

You simply need to input the high dimensional vectors to compute the first two principal components which then can be used to project your data. The implementation is just straight forward power iteration.

MDS

For MDS only a distance matrix for elements is required to position them on a 2D plane. This can be useful for visualizing high-dimensional data, drawing arbitrary graphs, or projecting points using a custom distance function.

The algorithm used for calculating the MDS is published at

@incollection{
  year = {2007},
  isbn = {978-3-540-70903-9},
  booktitle = {Graph Drawing},
  volume = {4372},
  series = {Lecture Notes in Computer Science},
  editor = {Kaufmann, Michael and Wagner, Dorothea},
  doi = {10.1007/978-3-540-70904-6_6},
  title = {Eigensolver Methods for Progressive Multidimensional Scaling of Large Data},
  url = {http://dx.doi.org/10.1007/978-3-540-70904-6_6},
  publisher = {Springer Berlin Heidelberg},
  author = {Brandes, Ulrik and Pich, Christian},
  pages = {42-53},
  language = {English}
}

Currently, most of the optimizations in the paper are not implemented yet, though. A Java implementation of the paper can be found at mdsj.