The function optimizes a set partitions based on the value of a criterion function (see crit.fun for details on the criterion function) for a given network and blockmodel for Generalized blockmodeling (<U+017D>iberna, 2006) based on other parameters (see below).
The optimization is done through local optimization, where the neighborhood of a partition includes all partitions that can be obtained by moving one unit from one cluster to another or by exchanging two units (from different clusters).
A list of paritions can be specified (opt.these.par) or the number of clusters and a number of partitions to generate (opt.random.par).
opt.random.par(M, k, n = NULL, rep, approach, ..., return.all =
FALSE, return.err = TRUE, maxiter = 50, trace.iter =
FALSE, switch.names = NULL, save.initial.param = TRUE,
skip.par = NULL, save.checked.par = TRUE,
merge.save.skip.par = any(!is.null(skip.par),
save.checked.par), skip.allready.checked.par = TRUE,
check.skip = "iter", print.iter = FALSE, max.iden =
10, seed = NULL, parGenFun = genRandomPar, mingr = 1,
maxgr = Inf, addParam = list(genPajekPar = TRUE,
probGenMech = NULL), maxTriesToFindNewPar = rep * 10)opt.these.par(M, partitions, approach, ..., return.all = FALSE,
return.err = TRUE, skip.allready.checked.par = TRUE,
maxiter = 50, trace.iter = FALSE, switch.names = NULL,
save.initial.param = TRUE, skip.par = NULL,
save.checked.par = !is.null(skip.par),
merge.save.skip.par = all(!is.null(skip.par),
save.checked.par), check.skip = "never", print.iter =
FALSE)
A matrix representing the (usually valued) network. For now, only one-relational networks are supported. The network can have one or more modes (diferent kinds of units with no ties among themselvs. If the network is not two-mode, the matrix must be square.
The number of clustrs used in generation of partitions.
The vector of the number of units in each mode (only necessary if mode is larger than 2.
The number of repetitions/different starting partitions to check.
A list of partitions. Each unique value represents one cluster. If the nework is one-mode, than this should be a vector, else a list of vectors, one for each mode.
One of the approaches described in <U+017D>iberna (2007). Possible values are: "bin" - binary blockmodeling, "val" - valued blockmodeling, "imp" - implicit blockmodeling, "ss" - sum of squares homogenity blockmodeling, and "ad" - absolute deviations homogenity blockmodeling.
Argumets passed to other functions, see crit.fun and arguments to function gen.crit.fun (as this function is not intented to be called directly, it also has no help files). Some might be obligatory, e.g. argument m when using Valued blockmodeling approach.Therefore these arguments are described below:
use.for.opt: Should FORTRAN function be used for optimization if possible. If FORTRAN function is used, the speed is dramatically increast, however some the output is slightly different and the plotting function might not work. FORTRAN subrutines are available for only very special cases, currently only for using "ss" aproach and only complete blocks. If you are using such setting and some special features (these are not implemented in FORTRAN subrutines - e.g. using function parOK to allow only certain kinds of partitions), it's safer to set it to FASLE, as the fuction may miss that these features are not implemented in FORTRAN subrutines and use them nevertheless, leading to wrong results.
use.for: (default = TRUE) Should FORTRAN subrutines be used where available (available for only very special cases, currently only for using "ss" aproach and only complete blocks. If you are using such setting and some special features (these are not implemented in FORTRAN subrutines), it's safer to set it to FASLE, as the fuction may miss that these features are not implemented in FORTRAN subrutines and use them nevertheless, leading to wrong results.
check.switch: If TRUE (the default), the neighborhood of the selected partition also includes the partitions that can be obtained by exchanging (switching) two units from diferent clusters).
check.all: If TRUE (the default), all partitions in the neighborhood of the selected partition are first evaluated and the current partition than changes to the one with the lowest value of the criterion function (if lower than that of the current partition). If FALSE, the first partition with the criterion lower the current partition becomes the new current partition (and the iteration terminates).
If FALSE, solution for only the best (one or more) partition/s is/are returned.
Should the error for each optimized partition be returned
Maximum number of iterations
Should the result of each iteration (and not only of the best one) be saved
Should partitions that differ only in diferent names of positions be treated as different. It should be set to TRUE only if a asymetric blockmodel via BLOCKS is specified.
Should the inital parameters (approach,...) be saved
The partitions that are not allowed or were already checked and should therfire be skiped.
Should the checked partitions be saved. For example, so that they can be used in the next call as skip.par
Should the checked partitions be merged with skiped ones?
If TRUE,the partitions that were already checked when runing opt.par form different statrting points will be skiped.
When should the check be preformed:
"all" - before every call to 'crit.fun' (Time demanding)
"iter" - at the end of eack iteratiton
"opt.par" - before every call to opt.par, when starting the optimization of a new partition.
"never" - never
Should the progress of each iteration be printed?
The maximum number of results that should be saved (in case there are more than max.iden results with minimal error, only the first max.iden will be saved).
Optional. The seed for random generation of partitions.
The fucntion (object) that will generate rendom partitions. The deault fuction is genRandomPar. The function has to accept the following parameters: k (number of partitions by modes, n (number of units by modes), seed (seed value for random generation of partition), addParam (a list of additional parametres)
Minimal alowed group size
Maximal alowed group size
A list of additional parameters for function specified above. In the usage section they are specified for the dthe default function genRandomPar:
The maximum number of partition try when trying to find a new partition to optimize that was not yet checked before - the default value is rep*1000
The matrix of the network analyzed
If return.all = TRUE - A list of results the same as best - one best for each partition optimized
A list of results from crit.fun.tmp with the same elements as the result of crit.fun, only without M
If return.err = TRUE - The vector of errors or inconsistencies of the emplirical network with the ideal network for a given blockmodel (model,approach,...) and parititions
The vector of number of iterations used - one value for each starting partition that was optimized. It can show that maxiter is to low if a lot of these values have the value of maxiter
If selected - A list of checked parititions. If merge.save.skip.par is TRUE, this list also includs the partitions in skip.par.
The call used to call the function.
If selected - The inital parameters used.
This function can be extremly slow. The time complexity is incrising with the number od units and the number of clusters. It is advaisable to firtst time the function on a smaller network.
<U+017D>IBERNA, Ale<U+0161> (2007): Generalized Blockmodeling of Valued Networks. Social Networks, Jan. 2007, vol. 29, no. 1, 105-126. http://dx.doi.org/10.1016/j.socnet.2006.04.002.
<U+017D>IBERNA, Ale<U+0161>. Direct and indirect approaches to blockmodeling of valued networks in terms of regular equivalence. J. math. sociol., 2008, vol. 32, no. 1, 57-84. http://www.informaworld.com/smpp/content?content=10.1080/00222500701790207.
DOREIAN, Patrick, BATAGELJ, Vladimir, FERLIGOJ, Anu<U+0161>ka (2005): Generalized blockmodeling, (Structural analysis in the social sciences, 25). Cambridge [etc.]: Cambridge University Press, 2005. XV, 384 p., ISBN 0-521-84085-6.
BATAGELJ, Vladimir, MRVAR, Andrej (2006): Pajek 1.11, http://mrvar.fdv.uni-lj.si/pajek/ (accessed January 6, 2006).
# NOT RUN {
n<-8 #if larger, the number of partitions increases dramaticaly,
#as does if we increase the number of clusters
net<-matrix(NA,ncol=n,nrow=n)
clu<-rep(1:2,times=c(3,5))
tclu<-table(clu)
net[clu==1,clu==1]<-rnorm(n=tclu[1]*tclu[1],mean=0,sd=1)
net[clu==1,clu==2]<-rnorm(n=tclu[1]*tclu[2],mean=4,sd=1)
net[clu==2,clu==1]<-rnorm(n=tclu[2]*tclu[1],mean=0,sd=1)
net[clu==2,clu==2]<-rnorm(n=tclu[2]*tclu[2],mean=0,sd=1)
#we select a random parition and then optimise it
all.par<-nkpartitions(n=n, k=length(tclu))
#forming the partitions
all.par<-lapply(apply(all.par,1,list),function(x)x[[1]])
# to make a list out of the matrix
#optimizing one partition
res<-opt.par(M=net,
clu=all.par[[sample(1:length(all.par),size=1)]],
approach="ss",blocks="com")
plot(res) #Hopefully we get the original partition
#optimizing 10 random chosen partitions with opt.these.par
res<-opt.these.par(M=net,
partitions=all.par[sample(1:length(all.par),size=10)],
approach="ss", blocks="com")
plot(res) #Hopefully we get the original partition
#optimizing 10 random chosen partitions with opt.random.par
res<-opt.random.par(M=net,k=2,rep=10,approach="ss",blocks="com")
plot(res) #Hopefully we get the original partition
# }
Run the code above in your browser using DataLab