Package 'RJcluster'

A Fast Clustering Algorithm for High Dimensional Data Based on the Gram Matrix Decomposition

Description

Clustering algorithm for high dimensional data. Assuming that P feature measurements on N objects are arranged in an N×P matrix X, this package provides clustering based on the left Gram matrix XX^T. To simulate test data, type "help('simulate_HD_data')" and to learn how to use the clustering algorithm, type "help('RJclust')". To cite this package, type 'citation("RJcluster")'.

Details

Package:	RJcluster
Type:	Package
Version:	3.2.4
Date:	07-15-2021
License:	GPL>=2

Author(s)

Shahina Rahman [aut], Valen E. Johnson [aut], Suhasini Subba Rao [aut], Rachael Shudde [aut, cre, trl]

Maintainer: Rachael Shudde <[email protected]>

---

Mutual_Information

Description

Calculates normalized mutual information and adjusted mutual information. The value for both will be a value bewteen 0 and 1 that measures how close the classification between the two clusters is. A value closer to 1 means the labels are more similar across v1 and v2, and a value closer to 0 means the labels are not as similar.

Usage

Mutual_Information(v1, v2)

Arguments

v1	vector containing first classification labels
v2	vector containing second classification labels

Details

See these links for a more formal definition of AMI and NMI.

Value

Returns mutual information:

nmi	NMI value
ami	AMI value

Examples

cluster1 <- sample(1:5, size = 10, replace = TRUE)
cluster2 <- sample(1:2, size = 10, replace = TRUE)
Mutual_Information(cluster1, cluster2)

---

RJclust

Description

This is a high dimensional clustering algorithm for data in matrix form. There are are two different types of penalty methods that can be used, depending on the size of the data and the desired accuracy. The first is the default method: the hokey stick penalty. There is also the BIC penalty. For large $n$, the scale method can be used, which uses the approximation method of RJclust. For the scaleRJ method, a parmater n_bins (usually $\sqrt(p)$) is required that splits the data into different buckets. For all methods, a C_max variable is needed that is an upper limit on the possible number of clusters.

Usage

RJclust(
  data,
  penalty = "hockey_stick",
  scaleRJ = FALSE,
  C_max = 10,
  criterion = "VVI",
  n_bins = NULL,
  seed = 1,
  verbose = FALSE
)

Arguments

data	Data input, must be in matrix form. Currently no support for missing values
penalty	A string of possible vectors. Options include: "bic" an "hockey_stock" (default = "hockey_stick")
scaleRJ	Should the scaled version of RJ be used, suggested for data where n > 1000 (default = FALSE)
C_max	Maximum number of clusters to look for (default is 10)
criterion	Model of covariance structure (default = "VVI")
n_bins	Number of cuts if penalty = "scale" for the scaled RJ algorithm (default = sqrt(p))
seed	Seed (defalt = 1)
verbose	Should progress be printed? (default = FALSE)

Details

All implementations use backend C++ to increase runtime.

model_names controls the type of covariance structure. See Mclust Documenttion for more information. Note criterion "kmeans" is the same as "EEI". It is not suggested to use "kmeans" if it is suspected the classes are imbalanced

Value

Returns RJ algorithm result for "aic", "bic" ("mclust" and "scale" will return an mclust object:

K	number of clusters found
class	Class labels
penalty	Penalty values at each iteraiton
mean	Mean matrix
prob	Probability values
z	Z values from mclust (NULL penalty = "full_covariance")

Examples

X = simulate_HD_data()
X = X$X
clust = RJclust(X, penalty = "hockey_stick", C_max = 10)

---

simulate_HD_data

Description

This is simulaiton data to check performance of RJcluster. Data can be simulated for any n, P, and size of clusters. The data has two types of data: noisy data and signal data. The percent of the data that is noisy is controlled by the sparsity paramater. The noisy data has two parts: half of it is $N(0,1)$ and half is $N(0, noise_variance)$. The signal data is divided in two as well, half of it is $N(\mu[,1], signal_variance)$ and half $N(\mu[,2], signal_variance)$.

Usage

simulate_HD_data(
  size_vector = c(20, 20, 20, 20),
  p = 220,
  mu = matrix(c(1.5, 2.5, 0, 1.5, 0, -1.5, -2.5, -1.5), ncol = 2, byrow = TRUE),
  signal_variance = 1,
  noise_variance = 1,
  sparsity = 0.09,
  seed = 1234
)

Arguments

size_vector	A list of the size of the different clusters. (default = a balanced case of 4 clusters of size 20, c(20, 20, 20, 20))
p	The number of columns in the simulated matrix (default = 220)
mu	The matrix of means, of dimension length(size_vector)x2. The first column of means is for the first half informative features, the second columns of mean is for the second half of the informative features (default is described in RJcluster paper)
signal_variance	Variance of the signal part of the generated data. A value of 1 indicates a high SNR, a value of 2 indicates a low SNR (default = 1)
noise_variance	Variance of the noisy part of the generated data (Default = 1)
sparsity	What percent of the data should be informative? A value between 0 and 1, a higher value means more data is informative (default = 0.09)
seed	Random seed. Change if generating multiple simulation datasets (default = 1234)

Details

The data in the paper is generated with number of clusters = 4, a balanced case of c(20, 20, 20, 20) and an unbalanced case of c(20, 20, 200, 200), with p = 220 in both cases. The default is a balanced, high signal case with $\mu$ as the matrix in the RJcluster paper.

Value

Returns simulation data for X and Y values

X	Matrix of dimension sum(size_vector)xp
Y	Vector of class labels of length $\sum(size_vector)$, with unique values of 1:length(size_vector)

Examples

data = simulate_HD_data()
X = data$X
Y = data$X
print(head(X))

---