estimate_ERPM {ERPM}

Estimate ERPM

Description

Function to estimate a given model for a given observed partition. All options of the algorithm can be specified here.

Usage

estimate_ERPM(
  partition,
  nodes,
  objects,
  effects,
  startingestimates,
  gainfactor = 0.1,
  a.scaling = 0.8,
  r.truncation.p1 = -1,
  r.truncation.p2 = -1,
  burnin = 30,
  thining = 10,
  length.p1 = 100,
  min.iter.p2 = NULL,
  max.iter.p2 = NULL,
  multiplication.iter.p2 = 100,
  num.steps.p2 = 6,
  length.p3 = 1000,
  neighborhood = c(0.7, 0.3, 0),
  fixed.estimates = NULL,
  numgroups.allowed = NULL,
  numgroups.simulated = NULL,
  sizes.allowed = NULL,
  sizes.simulated = NULL,
  double.averaging = FALSE,
  inv.zcov = NULL,
  inv.scaling = NULL,
  parallel = FALSE,
  parallel2 = FALSE,
  cpus = 1,
  verbose = FALSE
)

Arguments

partition

observed partition

nodes

nodeset (data frame)

objects

objects used for statistics calculation (list with a vector "name", and a vector "object")

effects

effects/sufficient statistics (list with a vector "names", and a vector "objects")

startingestimates

first guess for the model parameters

gainfactor

numeric used to decrease the size of steps made in the Newton optimization

a.scaling

numeric used to reduce the influence of non-diagonal elements in the scaling matrix (for stability)

r.truncation.p1

numeric used to limit extreme values in the covariance matrix (for stability)

r.truncation.p2

numeric used to limit extreme values in the covariance matrix (for stability)

burnin

integer for the number of burn-in steps before sampling

thining

integer for the number of thining steps between sampling

length.p1

number of samples in phase 1

min.iter.p2

minimum number of sub-steps in phase 2

max.iter.p2

maximum number of sub-steps in phase 2

multiplication.iter.p2

value for the lengths of sub-steps in phase 2 (multiplied by 2.52^k)

num.steps.p2

number of optimisation steps in phase 2

length.p3

number of samples in phase 3

neighborhood

way of choosing partitions: probability vector (actors swap, merge/division, single actor move)

fixed.estimates

if some parameters are fixed, list with as many elements as effects, these elements equal a fixed value if needed, or NULL if they should be estimated

numgroups.allowed

vector containing the number of groups allowed in the partition (now, it only works with vectors like num_min:num_max)

numgroups.simulated

vector containing the number of groups simulated

sizes.allowed

vector of group sizes allowed in sampling (now, it only works for vectors like size_min:size_max)

sizes.simulated

vector of group sizes allowed in the Markov chain but not necessraily sampled (now, it only works for vectors like size_min:size_max)

double.averaging

option to average the statistics sampled in each sub-step of phase 2

inv.zcov

initial value of the inverted covariance matrix (if a phase 3 was run before) to bypass the phase 1

inv.scaling

initial value of the inverted scaling matrix (if a phase 3 was run before) to bypass the phase 1

parallel

whether the phase 1 and 3 should be parallelized

parallel2

whether there should be several phases 2 run in parallel

cpus

how many cores can be used

verbose

logical: should intermediate results during the estimation be printed or not? Defaults to FALSE.

Value

A list with the outputs of the three different phases of the algorithm

Examples

# define an arbitrary set of n = 6 nodes with attributes, and an arbitrary covariate matrix
n <- 6 
nodes <- data.frame(label = c("A","B","C","D","E","F"),
                    gender = c(1,1,2,1,2,2),
                    age = c(20,22,25,30,30,31)) 
friendship <- matrix(c(0, 1, 1, 1, 0, 0,
                       1, 0, 0, 0, 1, 0,
                       1, 0, 0, 0, 1, 0,
                       1, 0, 0, 0, 0, 0,
                       0, 1, 1, 0, 0, 1,
                       0, 0, 0, 0, 1, 0), 6, 6, TRUE)

# choose the effects to be included (see manual for all effect names)
effects <- list(names = c("num_groups","same","diff","tie"),
                objects = c("partition","gender","age","friendship"))
objects <- list()
objects[[1]] <- list(name = "friendship", object = friendship)

# define observed partition
partition <- c(1,1,2,2,2,3)


# estimate
startingestimates <- c(-2,0,0,0)
estimation <- estimate_ERPM(partition, 
                            nodes, 
                            objects, 
                            effects, 
                            startingestimates = startingestimates, 
                            burnin = 100, 
                            thining = 20,
                            length.p1 = 500, # number of samples in phase 1
                            multiplication.iter.p2 = 20, # iterations in phase 2
                            num.steps.p2 = 4, # number of phase 2 subphases
                            length.p3 = 1000) # number of samples in phase 3

# get results table
estimation