clusterabilitytest {clusterability}

Perform a test of clusterability

Description

Performs tests for clusterability of a data set and returns results in a clusterability object. Can do data reduction via PCA or pairwise distances and standardize data prior to performing the test.

Usage

clusterabilitytest(data, test, reduction = "pca",
  distance_metric = "euclidean", distance_standardize = "std",
  pca_center = TRUE, pca_scale = TRUE, is_dist_matrix = FALSE,
  completecase = FALSE, d_simulatepvalue = FALSE, d_reps = 2000,
  s_m = 999, s_adjust = TRUE, s_digits = 6, s_setseed = NULL,
  s_outseed = FALSE)

Arguments

data

the data set to be used in the test. Must contain only numeric data.

test

the test to be performed. Either "dip" or "silverman". See 'Details' section below for how to pick a test.

reduction

any dimension reduction that is to be performed.

• "none" performs no dimension reduction.

• "pca" uses the scores from the first principal component.

• "distance" computes pairwise distances (using distance_metric as the metric).

For multivariate data, dimension reduction is required.

distance_metric

if applicable, the metric to be used in computing pairwise distances.

The "euclidean" (default), "maximum", "manhattan", "canberra", "binary" choices work the same as in dist. The Minkowski metric is available by providing "minkowski(p)".

Additional choices are:

• "sqeuc": squared Euclidean distances.

• "cov": covariance similarity coefficient,

• "corr": correlation similarity coefficient

• "sqcorr": squared correlation similarity coefficient.

CAUTION: Not all of these have been tested, but instead are provided to potentially be useful. If in doubt, use the default "euclidean".

distance_standardize

how the variables should be standardized, if at all.

• "none": no standardization is performed

• "std" (default) each variable standardized to have mean 0 and standard deviation 1

• "mean": each variable standardized to have mean 0 (standard deviation is unchanged)

• "median": each variable standardized to have median 0 (standard deviation is unchanged)

pca_center

if applicable, a logical value indicating whether the variables should be shifted to be zero centered (see prcomp for more details). Default is TRUE.

pca_scale

if applicable, a logical value indicating whether the variables should be scaled to have unit variance before the analysis takes place (see prcomp for details). Default is TRUE.

is_dist_matrix

a logical value indicating whether the data argument is a distance matrix. If TRUE then the lower triangular portion of data will be extracted and be used in the multimodality test.

completecase

a logical value indicating whether a complete case analysis should be performed. For both tests, missing data must be removed before the test can be performed. This can be done manually by the user or by setting completecase = TRUE.

d_simulatepvalue

for Dip Test, a logical value indicating whether p~values should be obtained via Monte Carlo simulation (see dip.test for details).

d_reps

for Dip Test, a positive integer. The number of replicates used in Monte Carlo simulation. Only used if d_simulatepvalue is TRUE.

s_m

for Silverman Test, a positive integer. The number of bootstrap replicates used in the test. Default is 999.

s_adjust

for Silverman Test, a logical value indicating whether p-values are adjusted using work by Hall and York.

s_digits

for Silverman Test, a positive integer indicating the number of digits to round the p value. Default is 6 and is only used when s_adjust = TRUE.

s_setseed

for Silverman Test, an integer used to set the seed of the random number generator. If the default value of NULL is used, then no seed will be set.

s_outseed

for Silverman Test, a logical value indicating whether to return the state of the random number generator as part of the output. This is used in limited cases for troubleshooting, so the default is FALSE.

Value

clusterabilitytest returns a clusterability object containing information on the test performed and results. Can be printed using the print.clusterability function.

References

Hall, P. and York, M., 2001. On the calibration of Silverman's test for multimodality. Statistica Sinica, pp.515-536.

Silverman, B.W., 1981. Using kernel density estimates to investigate multimodality. Journal of the Royal Statistical Society. Series B (Methodological), pp.97-99.

Martin Maechler (2016). diptest: Hartigan's Dip Test Statistic for Unimodality - Corrected. R package version 0.75-7. https://CRAN.R-project.org/package=diptest

Schwaiger F, Holzmann H. Package which implements the silvermantest; 2013. Available from: https://www.mathematik.uni-marburg.de/stochastik/R packages/.

See Also

print.clusterability

Examples

### Quick start ###
# Load data and remove Species
data(iris)
iris_num <- iris[,-5]
plot(iris_num)

# Run test using default options
clust_result <- clusterabilitytest(iris_num, "dip")

# Print results
print(clust_result)


### Longer Example: Specifying Parameters ###
# Load data and plot to visualize
data(normals2)
plot(normals2)

# Using Silverman's test, pairwise distances to reduce dimension,
# 1,000 bootstrap replicates, with an RNG seed of 12345
clust_result2 <- clusterabilitytest(normals2, "silverman", reduction = "distance",
     s_m = 1000, s_setseed = 12345)

# Print result
print(clust_result2)