Very large data sets

1
Very large data sets
Clustering methods Part 10
Pasi Fränti
5.5.2014

• Speech and Image Processing UnitSchool of
  Computing
• University of Eastern Finland

2
Methods for large data sets

• Birch
• Clarans
• On-line EM
• Scalable EM
• GMG

3
Gradual model generator (GMG) Kärkkäinen
Fränti, 2007 Pattern Recognition
4
Goal of the GMG algorithm
EM
GMG
5
Contours of probability density distributions
EM
GMG
6
Model update

• New data points are mapped immediately when
  input.
• Points too far (from any model) will remain in
  buffer.
• Buffered points are re-tested when new models
  created.

Before update
After update
7
Generating new components

• When buffer full, selected points are used to
  generate new components.
• Most compact k-neighborhood is selected as seed
  for a new component.

Selected points and a new component
Data in buffer
8
Example
9
Example
10
Example
11
Example
12
Example
13
Example
14
Post-processing
Model before processing
15
Post-processing
Model before processing
Updated model
16
Post-processing
Model before processing
Updated model data
17
Literature

1. I. Kärkkäinen and P. Fränti, "Gradual model
   generator for single-pass clustering", Pattern
   Recognition, 40 (3), 784-795, March 2007.
2. P. Bradley, U. Fayyad, C. Reina, Clustering Very
   Large Databases Using EM Mixture Models, Proc. of
   the 15th Int. Conf. on Pattern Recognition, vol.
   2, 2000, pp. 76-80.
3. R. Ng, J. Han, CLARANS A Method for Clustering
   Objects for Spatial Data Mining, IEEE Trans.
   Knowledge Data Engineering 14(5) (2002)
   1003-1016.
4. M. Sato, S. Ishii, On-line EM Algorithm for the
   Normalized Gaussian Network, Neural Computation
   12(2) (2000) 407-432.
5. T. Zhang, R. Ramakrishnan, M. Livny, BIRCH A New
   Data Clustering Algorithm and Its Applications,
   Data Mining and Knowledge Discovery 1(2) (1997)
   141-182.