classCleaner: classCleaner: A package for cleaning outliers when data is...

In melissakey/classCleaner: Tools to Identify Outliers in grouped (class) Data

Description

Test whether each instance in a class actually belongs.

Usage

classCleaner(
  D,
  assignment,
  classes = "all",
  alpha0 = 0.05,
  q = 0.5,
  labels = NULL,
  exclude_classes = NULL
)

Arguments

D	A distance matrix containing the pairwise dissimilarity scores between instances
assignment	The assigned group of each instance
classes	The subset of classes on which filtering is performed, or "all" if all classes should be analyzed.
alpha0	Desired global type I (v1) or type II (v2) error rate.
q	(v2 only) - the proportion of distances expected to be "close enough" to keep an instance. Defaults to 0.5.
labels	a vector of labels for each instance. Must be the same length as D. If NULL, the algorithm will check for rownames and column names in D. If none are found, the instances will be labeled with numbers 1:nrow(D).
exclude_classes	names of "mega" classes which should not be included in determining whether or not classCleaner2 is appropriate. By default, these classes will not be included in the analysis.

Details

For each instance in an analyzed class, this function will estimate the probability that it was correctly placed in that class.

Examples

set.seed(23)

X <- simulate_clustered_data(
  n = 200,
  Nk = rep(50, 100),
  s = rep(1, 100),
  rho = .2,
  tau = 1,
  method = "by-class"
)
# true assignment
a <- rep(1:100, each = 50)

# corrupted assignment
b <- sample(100, 50 * 100, replace = TRUE)

# corrupt 10% of samples
a.corrupt <- ifelse(runif(50 * 1000) < 0.1, b, a)

D <- 1 - cor(X)
result <- identify_outliers(a.corrupt, D, 1000, colnames(D))

melissakey/classCleaner documentation built on Feb. 11, 2022, 3:33 a.m.