- without kmeans init takes usually longer but with same result (not shown)
- However, optimization function for PSO is usually unknown because cluster structures are unknown
- For example, optimzing quantization error does not fit the Atom data set
- Number of clusters has to be derived otherwise
- Quantization error is on Euclidean distances defined
data(Atom)
Data = Atom$Data
out = BehaviorBasedSystems::PSOclustering(
Data,
2,
ParticleNo = 20,
Iterations = 300,
Hybrid_kmeans = T,
BreakIfNoChange = F,
Silent = T
)
## [1] "Make clusters..."
## KMNS(*, k=2): iter= 1, indx=3
## QTRAN(): istep=800, icoun=51
## QTRAN(): istep=1600, icoun=419
## QTRAN(): istep=2400, icoun=545
table(out$Cls, Atom$Cls)
##
## 1 2
## 1 214 196
## 2 186 204
DataVisualizations::Plot3D(Data, out$Cls)
You must enable Javascript to view this page properly.