parameter_tuning {ssMRCD}

Parameter Tuning

Description

This function provides insight into the effects of different parameter settings.

Usage

parameter_tuning(
  data,
  coords,
  N_assignments,
  lambda = c(0, 0.25, 0.5, 0.75, 0.9),
  weights = NULL,
  k = NULL,
  dist = NULL,
  cont = 0.05,
  repetitions = 5
)

Arguments

data

matrix with observations.

coords

matrix of coordinates of these observations.

N_assignments

numeric vector, the neighborhood structure that should be used for ssMRCD.

lambda

scalar, the smoothing parameter.

weights

weighting matrix used in ssMRCD.

k

vector of possible k-values to evaluate.

dist

vector of possible dist-values to evaluate.

cont

level of contamination, between 0 and 1.

repetitions

number of repetitions wanted to have a good picture of the best parameter combination.

Value

Returns a matrix of average false-negative rate (FNR) values and the total number of outliers found by the method as aproxy for the false-positive rate. Be aware that the FNR does not take into account that there are also natural outliers included in the data set that might or might not be found. Also a plot is returned representing these average. The best parameter selection depends on the goal of the analysis.

Examples

# get data set
data("weatherAUT2021")

# make neighborhood assignments
cut_lon = c(9:16, 18)
cut_lat = c(46, 47, 47.5, 48, 49)
N = ssMRCD::N_structure_gridbased(weatherAUT2021$lon, weatherAUT2021$lat, cut_lon, cut_lat)
table(N)
N[N == 2] = 1
N[N == 3] = 4
N[N == 5] = 4
N[N == 6] = 7
N[N == 11] = 15
N = as.numeric(as.factor(N))

# tune parameters
set.seed(123)
parameter_tuning(data = weatherAUT2021[, 1:6 ],
                 coords = weatherAUT2021[, c("lon", "lat")],
                 N_assignments = N,
                 lambda = c(0.5, 0.75),
                 k = c(10),
                 repetitions = 1)