upAndDownSearch: Perform Up-and-Down Search Optimization

In StochBlock: Stochastic Blockmodeling of One-Mode and Linked Networks

Perform Up-and-Down Search Optimization

Description

The upAndDownSearch function performs an up-and-down search (increasing and lowering the number of clusters) on a given dataset, starting from an initial partition and fitness value. The function assumes that the higher fitness value is better. Experimental! The stochBlockForUDS is just a wrapper to stochBlock function to be used within upAndDownSearch function.

Usage

upAndDownSearch(
  data,
  initPart,
  initFit,
  optimFun,
  nRep = 100,
  minPmove = 0.2,
  pFitSumTreshMoveBest = 5,
  ...
)

stochBlockForUDS(data, initPart, ...)

Arguments

data	A data object (e.g., data frame or matrix) on which the search is performed.
initPart	Initial partition or configuration to start the search from. It can be a vector for simple datasets and a list of vectors for temporal or linked networks
initFit	Initial fitness value associated with the initial partition.
optimFun	A function used for optimization. A function must accept a data and a initial partition and return a list with an element part holding the final partition and an element fit fitness value.
nRep	Integer. The number of repetitions for the search algorithm. Defaults to 100.
minPmove	Numeric. The minimum probability of moving to a new partition when the fitness does not improve. Defaults to 0.2.
pFitSumTreshMoveBest	Numeric. The threshold for the sum of 1 - probabilities of moving to a new partition before reverting to the best partition. Defaults to 5.
...	Additional paramters to optimFun.

Value

A list with the following components:

data

The input data.

finalPart

The final partition obtained.

finalFit

The final fitness value after the search.

searchHistory

A list containing the history of partitions and fitness values during the search.

callUsed

The call used to invoke the function, capturing the parameters passed.

initial.param

A list of initial parameters used in the function call withiut the data.

Examples

# Create a synthetic network matrix
set.seed(2022)
library(blockmodeling)
k<-2 # number of blocks to generate
blockSizes<-rep(20,k)
IM<-matrix(c(0.8,.4,0.2,0.8), nrow=2)
clu<-rep(1:k, times=blockSizes)
n<-length(clu)
M<-matrix(rbinom(n*n,1,IM[clu,clu]),ncol=n, nrow=n)
initClu<-rep(1, times=n)
initFit<-ICLStochBlock(M, initClu) # Initial fitness value
# Using up-and-down search to optimise the partition
res<-upAndDownSearch(data=M,initPart=initClu, initFit=initFit, optimFun=stochBlockForUDS, nRep=10) 
plotMat(res$data, clu=res$bestPart) # Have a look at the optimised parition
print(res$bestFit) # Print the final fitness value

# Create a synthetic linked-network matrix
set.seed(2022)
library(blockmodeling)
IM<-matrix(c(0.9,.5,0.1,0.8), nrow=2)
clu<-rep(1:2, each=20) # Partition to generate
n<-length(clu)
nClu<-length(unique(clu)) # Number of clusters to generate
M1<-matrix(rbinom(n^2,1,IM[clu,clu]),ncol=n, nrow=n) # First network
M2<-matrix(rbinom(n^2,1,IM[clu,clu]),ncol=n, nrow=n) # Second network
M12<-diag(n) # Linking network
nn<-c(n,n)
k<-c(2,2)
Ml<-matrix(0, nrow=sum(nn),ncol=sum(nn)) 
Ml[1:n,1:n]<-M1
Ml[n+1:n,n+1:n]<-M2
Ml[n+1:n, 1:n]<-M12 
plotMat(Ml) # Linked network
clu1<-rep(1, n)
clu2<-rep(2, n)
initClu<-list(clu1, clu2)
initFit<-ICLStochBlock(Ml, initClu) # Initial fitness value
# Using up-and-down search to optimise the partition
res<-upAndDownSearch(data=Ml,initPart=initClu, initFit=initFit, optimFun=stochBlockForUDS, nRep=10)
plotMat(res$data, clu=res$bestPart) # Have a look at the optimised parition
print(res$bestFit) # Print the final fitness value

StochBlock documentation built on July 25, 2025, 3 a.m.