flowClust &
flowMerge.A wrapper for flowClust, clustering
a time-series object tset provided by processTimeseries,
where specifically the DFT of a time-series and requested data
transformation were calculated. This is intended to work in the same way
as clusterTimeseries but was so far only tested for
clustering of the final segment time-series, as previously applied
to microarray data from yeast by Machne & Murray (2012)
<doi:10.1371/journal.pone.0037906> and from cyanobacteria by Lehmann
et al. (2013) <doi:10.1186/1471-2105-14-133>.
It could in principle also be used for segmentation, but that has not
been extensively tested. flowClust
implements a model-based clustering approach and is much slower then
kmeans used in clusterTimeseries.
Please see option ncpu on how to use parallel mode, which
does not work on some installations. However, model-based clustering has
the advantage of an intrinsic measure (BIC) to decide on the optimal
cluster numbers. Additionally, the clusters can be "merged" to fewer
clusters at constant BIC using
flowMerge.
flowclusterTimeseries(tset, ncpu = 1, K = 10, selected,
merge = FALSE, B = 500, tol = 1e-05, lambda = 1, nu = 4,
nu.est = 0, trans = 1, ...)processed time-series as provided by
processTimeseries
number of cores available for parallel mode of
flowClust. NOTE: parallel mode of
flowClust is often non-functional.
Alternatively, you can set options(mc.cores=ncpu) directly.
the requested cluster numbers (vector of integers)
a pre-selected cluster number which is then
used as a start clustering for flowMerge
(if option merge==TRUE)
logical indicating whether cluster merging with
flowMerge should be attempted
maximal number of EM iterations
tolerance for EM convergence
initial Box-Cox trafo
degrees of freedom used for the t distribution, Inf for pure Gaussian
0: no, 1: non-specific, 2: cluster-specific estimation of nu
0: no, 1: non-specific, 2: cluster-specific estim. of lambda
further parameters for
flowClust
Machne & Murray (2012) <doi:10.1371/journal.pone.0037906>