Optimizing Classification by Maximizing Dufrene and Legendre's Indicator Value
optindval is a iterative re-assignment classification algorithm that assigns samples to clusters to maximize the sum of indicator values.
optindval(comm,clustering,maxitr=100,minsiz=5)
comm |
a vegetation or other taxon table with samples as rows and taxa as columns |
clustering |
an index of cluster membership for each sample. May be either a numeric vector of length equal to the number of samples, or an object that inherits from class ‘cluster’ |
maxitr |
the maximum number of iterations to attempt |
minsiz |
the minimum size cluster to consider reassigning a sample out of |
Iterative re-allocation algorithms temporarily re-assign each sample to each of the other possible clusters and calculate a goodness-of-clustering statistic for each re-assignment. The best of all possible re-assignments is then executed and the algorithm iterates until there are no more good re-assignments or the maximum number of iterations is reached. In optindval, the goodness-of-clustering statistic is the sum of Dufrene and Legendre indicator values
a list of class "optindval","clustering" with components:
numitr |
the number of iterations performed |
sums |
a vector of indicator value probability sums |
clustering |
the vector of cluster memberships (as integers) for each sample |
Like many iterative re-assignment algorithms, optindval is likely to be VERY slow from a random start or poor initial condition. optindval is maybe better used to polish existing classifications
David W. Roberts droberts@montana.edu
optpart, opttdev, optsil
data(shoshveg) # returns a data.frame of vegetation data called shoshveg
dis.bc <- dsvdis(shoshveg,'bray') # generate Bray/Curtis dissimilarity
# matrix
opt.5 <- optpart(5,dis.bc) # generate 5-cluster optpart
## Not run: res <- optindval(shoshveg,opt.5) # polish the optpart result
## Not run: classmatch(opt.5,res) # see the plot re-assignmentsPlease choose more modern alternatives, such as Google Chrome or Mozilla Firefox.