funHDDCwavelet: Perform the funHDDCwavelet algorithm on a matrix of time...

In Erwangf/funHDDC-wavelet: Time series clustering with wavelet and gaussian mixture model

Description

This package provide an implementation of the funHDDCwavelet algorithm. This algorithm allow the clustering of time series by representing them in a parcimonious and multi-resolution way.

Usage

funHDDCwavelet(X, K, minD = 1, maxD = 1,
  max.level = round(log2(ncol(X))), dimTest = "scree",
  wavelet.family = "DaubExPhase", wavelet.filter.number = 4,
  init = "kmeans", minPerClass = 10, threshold = 0.01,
  minIter = 10, maxIter = 50, poolSize = 5, verbose = FALSE,
  viz = F)

Arguments

X	A matrix of time series (each line correspond to a time serie, each column to an time point)
K	The number of classes to find
minD	The minimum dimension of the subspace where a level can be represented (default : 1)
maxD	The maximum dimension of the subspace where a level can be represented (default : 1)
max.level	The maximum wavelet transform level used in the algorithm (default : all levels, depending of the time serie length)
dimTest	The intrinsec dimension estimation method. Available : "scree", "kss", "mean", "XX%" (ex "85%") (default: "scree")
wavelet.family	The wavelet family used for the discrete wavelet transform (see wavethresh package). Default : "DaubExPhase"
wavelet.filter.number	The filter number used in for the discrete wavelet transform (see wavethresh package). Default : 4
init	Type of initialization method. Available : "kmeans", "random" (default : "kmeans")
minPerClass	The minimal size of the initial classes
threshold	The threshold used to determine that the log-likelihood converged. Default : 0.01
minIter	The minimal number of iterations of the EM algorithm
maxIter	The maximal number of iterations of the EM algorithm
poolSize	The number of run of the algorithm (the algorithm is executed poolSize times, and the best model is selected via BIC) Default : 5
verbose	if TRUE, prompt some information of the algorithm status over time
viz	if TRUE, plot the first 2 axes of the data after principal component analysis, with current cluster repartitions and colors

Examples

## Not run: 
dataset = generateDataset()
X = dataset$X
real_cluster = dataset$col

# Haar wavelet
result = funHDDCwavelet(X,K=3,minD=1,maxD=1,wavelet.family="DaubExPhase",
                        wavelet.filter.number=1,viz=TRUE,minIter=10)
clusters = apply(result$tm,1,which.max)
adjustedRandIndex(clusters,real_cluster)
plot.curve.dataset(X,col=clusters,ratio=0.1)

## End(Not run)

Erwangf/funHDDC-wavelet documentation built on June 7, 2019, 12:51 a.m.