h.wickham at gmail.com
Sun Jul 1 17:58:54 CEST 2007
Typically, there is somewhat of a divide between statistics and
visualisation software. Statistics software, particularly R, provides
implementation of cutting edge research methods, but limited graphics.
Visualisation software will provide sophisticated visual interfaces,
but few statistical algorithms. The clusterfly package presents some
early experimentation aimed at overcoming this deficiency by linking R
and GGobi. Cluster analysis was chosen as it is an exploratory method
that needs sophisticated visualisation and statistical algorithms.
Clusterfly provides some tools that work with all clustering
algorithms, and some that are tailored for particular ones. Generic
tools allow you to animate between clusterings (see ?cfly_animate) and
produce common static graphics (?cfly_dist, ?cfly_pcp). Specific
algorithms are available for:
* Self organising maps (aka Kohonen neural networks), ?ggobi.som.
Displays the self organising map/net in the original space of the
* Hierarchical clustering, ?hierfly. Connects data points with lines
like a dendrogram, but in the high-dimensional space of the original
* Model based clustering, ?mefly. Adds ellipsoids from the
multivariate normal distributions the clusters are based on
You will need GGobi (http://www.ggobi.org) and rggobi
(http://www.ggobi.org/rggobi) installed to be able to use clusterfly.
More information about the R-packages