AdaptGauss2D: Adapt a two-dimensional Gaussian Mixture Model

In Mthrun/AdaptGauss2D: Gaussian Mixture Modeling in 2D

Adapt a two-dimensional Gaussian Mixture Model

Description

GUI for manually fitting a multivariate (2 dimensional) gaussian mixture to a dataset.

Usage

AdaptGauss2D(Data, Means = NULL, CovarianceMatrices = NULL, Weights = NULL,
Cls = NULL, Debug = F, dbt = F, WorkingDirectory = getwd())

Arguments

Data	[1:n, 1:2] Numeric matrix with n observations and 2 feature dimensions.
Means	List of l two-dimensional numeric vectors carrying mean vector.
CovarianceMatrices	List of l 2x2 numeric matrices carrying covariance matrices.
Weights	[1:l] Numeric vector with l weights for l Gaussian Mixture Model components.
Cls	[1:n] Integer vector with n entries for classes assigned to each data point in Data.
Debug	Boolean (Default=FALSE). T: Show developer information and warnings in terminal. F: Show nothing.
dbt	Boolean (Default=FALSE). T: Provide more complex functionality to the dbt. F: Hide unnecessary details to keep it simple for a broad audience especially from the non computer science community.
WorkingDirectory	Character String (Default=getwd()). Indicates the working directory to which to save the setting obtained with AdaptGauss2D to.

Details

The AdaptGauss2D app is a GUI made for an intuitive approach for two-dimensional gaussian mixture modeling (GMM). There is also an approach for the one-dimensional case, called AdaptGauss. The output of the app can be used as input. Be aware, that the principal component axis and the angle are not required, since this information is within the covariance matrices and is only provided in the output as a quick access of the results. The essence of AdaptGauss2D lies in the coarse manual fitting as a first step and a fine automatized fitting with an EM algorithm as the second step. That way the user can communicate a rough expectation of the model which is then optimized. In other words, the user provides a rough desired initialization and the automaton of AdaptGauss2D finds a local optimum for that. AdaptGauss2D yields an initialized solution for a GMM as proposal and start for the user.

Value

List with multiple elements of different data structures:

Means	List of l numerical 2D vectors. Each vector representing the expectation values for one Gaussian component.
CovarianceMatrices	List of l numerical 2x2 matrices.
Weights	[1:l] Numerical vector of size l containing the weight for each Gaussian component within the mixture.
PrincipalComponentAxis	List of l numerical 2D vectors. Each matrix representing the prinicpal component axis of the covariance matrix of a Gaussian component.
Angle	List of l integers. Each integer represents the angle of the first principal component ax of the covariance matrix.
Cls	[1:n] Numerical vector of size n containing the classes of each observation.

Author(s)

Quirin Stier

Examples

## Not run: 
data(EngyTime)
# Call shiny application
Data=EngyTime$Data
Res = AdaptGauss2D(Data, Means = NULL, CovarianceMatrices = NULL, Weights = NULL,
Cls = NULL, Debug = F)

## End(Not run)

Mthrun/AdaptGauss2D documentation built on July 19, 2022, 3:11 a.m.