Prototipo
per algoritmi di clustering incrementale di dati:
MICluE
|
Stefano Lodi, Claudio Sartori
| Tema |
Tema
3: Data Mining |
| Codice |
D3-P1 |
| Data |
8 ottobre 2002 |
| Tipo di prodotto |
Prodotto software |
| Unità
responsabile |
BO |
| Unità
coinvolte |
BO |
| Autori |
Stefano Lodi, Claudio Sartori
|
| Autore da
contattare |
Stefano Lodi
Dipartimento di Elettronica, Informatica e Sistemistica
via Risorgimento 2
40136 Bologna BO
slodi@deis.unibo.it
|
| Presentazione
prototipo |
D3-P1 |
| Documentazione in
linea |
D3.P1.pdf |
Descrizione
MICluE (Metric Incremental Clustering Engine) is a prototype
for static and dynamic clustering in metric spaces. MICluE
accepts data sets embedded in both categorical and numeric spaces,
provided that a distance function is defined on the data space. The
prototype consists of two components: a graphical user interface and a
clustering engine. The graphical user interface is built using
standard visual components for input and parameter settings, and a
custom plotting utility to interactively display 2-dimensional
projections of clusters, if the data set is numeric. The plotting
utility allows zooming and panning in the data space. A distinct
plotting utility to display neighbourhood information for every data
object is also provided for debugging and exploration purposes. The
clustering engine uses the Multi-Vantage Point Tree (MVP-Tree) to
speed up distance and neighbourhood calculations, and provides five
static clustering methods based on neighbourhood information, and both
a static and a fully dynamic clustering method based on non-parametric
kernel density estimation. The static clustering methods based on
neighbourhood information include the shared near neighbours method,
the mutual neighbourhood value method, the kth neighbour method, and two
generalizations of such methods. The density estimate based method
allows for selecting the kernel function and its smoothing parameter,
as well as the approximation used to compute the density estimate.
Ambiente di sviluppo e di esecuzione
Borland Delphi 6 Enterprise
Microsoft Windows 2000 Professional
Back