Implements the Improved Expectation Maximisation EM* and the traditional EM algorithm for clustering
big data (gaussian mixture models for both multivariate and univariate datasets). This version
implements the faster alternative-EM* that expedites convergence via structure based data segregation.
The implementation supports both random and K-means++ based initialization. Reference: Parichit Sharma,
Hasan Kurban, Mehmet Dalkilic (2022)
Package Overview
Implements the Expectation Maximisation Algorithm for clustering the multivariate and univariate datasets. The package has been tested with numerical datasets (not recommended for categorical/ordinal data). The package comes bundled with a dataset for demostration (ionosphere_data.csv). More help about the package can be seen by typing ?DCEM
in the R console (after installing the package).
Currently, data imputation is not supported and user has to handle the missing data before using the package.
Contact
For any Bug Fixes/Feature Update(s)
[Parichit Sharma: [email protected]edu]
For Reporting Issues
GitHub Repository Link
Installation Instructions
Installing from CRAN
install.packages(dcem)
Installing from the Binary Package
install.packages(dcem_1.0.0.tgz, repos = NULL, type="source")
How to use the package (An Example: working with the default bundled dataset)
The dcem package comes bundeled with the ionosphere_data.csv for demostration. Help about the dataset can be seen by typing ?ionosphere_data
in the R console. Additional details can be seen at the link Ionosphere data
To use this dataset, paste the following code into the R console.
ionosphere_data = read.csv2(
file = paste(trimws(getwd()),"/data/","ionosphere_data.csv",sep = ""),
sep = ",",
header = FALSE,
stringsAsFactors = FALSE
)
dcem_train()
function), the data must be cleaned. This simply means to remove all redundant columns (example can be label colum). This datset contains labels in the last column (35th) and only 0's in the 2nd column so let's remove them,Paste the below code in the R session to clean the dataset.
ionosphere_data = trim_data("35,2", ionosphere_data)
dcem_cluster_mv()
or dcem_cluster_uv()
function for multivariate and univariate data respectively. These
functions assign(s) the probabilistic weights to the sample(s) in the dataset.Paste the below code in the R session to call the dcem_train() function.
dcem_out = dcem_train(data = ionosphere_data, threshold = 0.0001, iteration_count = 50, num_clusters = 2)
dcem_train()
is stored in the dcem_out object. It contains the parameters associated with the clusters (Gaussian(s)). These parameters are namely - posterior probabilities, mean, co-variance (multivariate data) or standard-deviation (univariate data) and priors. Paste the following code in the R session to access any/all the output parameters. [1] Posterior Probabilities: `**dcem_out$prob**`: A matrix of posterior-probabilities for the
points in the dataset.
[2] Mean(s): `**dcem_out$mean**`
For multivariate data: It is a matrix of means for the gaussians. Each row in the
matrix corresponds to a mean for the gaussian.
For univariate data: It is a vector if means. Each element of the vector corresponds
to one gaussian.
[3] Co-variance matrices
For multivariate data: `**dcem_out$cov**`: list of co-variance matrices for the gaussians.
For univariate data: Standard-deviation `**dcem_out$sd**`: vector of standard deviation(s)
for the gaussians.
[4] Priors: `**dcem_out$prior**`: a vector of priors for the gaussians.
How to access the help (after installing the package)
?dcem_train()
?dcem_test()
?DCEM
DCEM 0.0.1
This is the first stable realease of the DCEM package.
Major Features
Support clustering of both multivariate and univariate data for finite gaussian misxture models.