Title: | Optimized Automated Gaussian Mixture Assessment |
---|---|
Description: | Necessary functions for optimized automated evaluation of the number and parameters of Gaussian mixtures in one-dimensional data. Various methods are available for parameter estimation and for determining the number of modes in the mixture. A detailed description of the methods ca ben found in Lotsch, J., Malkusch, S. and A. Ultsch. (2022) <doi:10.1016/j.imu.2022.101113>. |
Authors: | Jorn Lotsch [aut,cre] , Sebastian Malkusch [aut] , Martin Maechler [ctb], Peter Rousseeuw [ctb], Anja Struyf [ctb], Mia Hubert [ctb], Kurt Hornik [ctb] |
Maintainer: | Jorn Lotsch <[email protected]> |
License: | GPL-3 |
Version: | 0.4 |
Built: | 2024-11-27 06:47:10 UTC |
Source: | CRAN |
Data set containing times of detector hits after chromatographic separation of five different lysophosphatidic acids (Classes CLs = LPA 16:0, 18:0, 18:3, 20:0, and 20:4).
data("Chromatogram")
data("Chromatogram")
Size 1166 x 3 , stored in Chromatogram$[Cls, Time, Lipids]
data(Chromatogram) str(Chromatogram)
data(Chromatogram) str(Chromatogram)
The function plots the components of a Gaussian mixture and superimposes them on a histogram of the data.
GMMplotGG(Data, Means, SDs, Weights, BayesBoundaries, SingleGausses = TRUE, Hist = FALSE, Bounds = TRUE, SumModes = TRUE, PDE = TRUE)
GMMplotGG(Data, Means, SDs, Weights, BayesBoundaries, SingleGausses = TRUE, Hist = FALSE, Bounds = TRUE, SumModes = TRUE, PDE = TRUE)
Data |
the data as a vector. |
Means |
a list of mean values for a Gaussian mixture. |
SDs |
a list of standard deviations for a Gaussian mixture. |
Weights |
a list of weights for a Gaussian mixture. |
BayesBoundaries |
a list of Bayesian boundaries for a Gaussian mixture. |
SingleGausses |
whether to plot the single Gaussian components as separate lines. |
Hist |
whether to plot a histgram of the original data. |
Bounds |
whether to plot the Bayesian boundaries for a Gaussian mixture as vertical lines. |
SumModes |
whether to plot the summed-up mixes. |
PDE |
whether to use the Pareto density estimation instead of the standard R density function. |
Returns a ggplot2 object.
p1 |
the plot of Gaussian mixtures. |
Jorn Lotsch and Sebastian Malkusch
Lotsch, J., Malkusch S. (2021): opGMMassessment – an R Package for automated Guassian mixture modeling.
## example 1 data(iris) Means0 <- tapply(X = as.vector(iris[,3]), INDEX = as.integer(iris$Species), FUN = mean) SDs0 <- tapply(X = as.vector(iris[,3]), INDEX = as.integer(iris$Species), FUN = sd) Weights0 <- c(1/3, 1/3, 1/3) GMM.Sepal.Length <- GMMplotGG(Data = as.vector(iris[3]), Means = Means0, SDs = SDs0, Weights = Weights0, Hist = TRUE)
## example 1 data(iris) Means0 <- tapply(X = as.vector(iris[,3]), INDEX = as.integer(iris$Species), FUN = mean) SDs0 <- tapply(X = as.vector(iris[,3]), INDEX = as.integer(iris$Species), FUN = sd) Weights0 <- c(1/3, 1/3, 1/3) GMM.Sepal.Length <- GMMplotGG(Data = as.vector(iris[3]), Means = Means0, SDs = SDs0, Weights = Weights0, Hist = TRUE)
Data set containing 1000 instances distributed according to a Gaussian mixture with m = [-10, 0, 10], s = [1, 2, 3], w = [0.07, 0.05, 0.88].
data("Mixture3")
data("Mixture3")
Size 1000 x 1
data(Mixture3) str(Mixture3)
data(Mixture3) str(Mixture3)
The package provides the necessary functions for optimized automated evaluation of the number and parameters of Gaussian mixtures in one-dimensional data. It provides various methods for parameter estimation and for determining the number of modes in the mixture.
opGMMassessment(Data, FitAlg = "MCMC", Criterion = "LR", MaxModes = 8, MaxCores = getOption("mc.cores", 2L), PlotIt = FALSE, KS = TRUE, Seed)
opGMMassessment(Data, FitAlg = "MCMC", Criterion = "LR", MaxModes = 8, MaxCores = getOption("mc.cores", 2L), PlotIt = FALSE, KS = TRUE, Seed)
Data |
the data as a vector. |
FitAlg |
which fit algorithm to use: "ClusterRGMM" = GMM from ClusterR, "densityMclust" from mclust, "DO" from DistributionOptimization (slow), "MCMC" = NMixMCMC from mixAK, or "normalmixEM" from mixtools. |
Criterion |
which criterion should be used to establish the number of modes from the best GMM fit: "AIC", "BIC", "FM", "GAP", "LR" (likelihood ratio test), "NbClust" (from NbClust), "SI" (Silverman). |
MaxModes |
the maximum number of modes to be tried. |
MaxCores |
the maximum number of processor cores used under Unix. |
PlotIt |
whether to plot the fit directly (plot will be stored nevertheless). |
KS |
perform a Kolmogorow-Smirnow test of the fit versus original distribution. |
Seed |
optional seed parameter set internally. |
Returns a list of Gaussian modes.
Cls |
the classes to which the cases are assigned according to the Gaussian mode membership. |
Means |
means of the Gaussian modes. |
SDs |
standard deviations of the Gaussian modes. |
Weights |
weights of the Gaussian modes. |
Boundaries |
Bayesian boundaries between the Gaussian modes. |
Plot |
Plot of the obtained mixture. |
KS |
Results of the Kolmogorov-Smirnov test. |
Jorn Lotsch and Sebastian Malkusch
Lotsch J, Malkusch S, Ultsch A. Comparative assessment of automated algorithms for the separation of one-dimensional Gaussian mixtures. Informatics in Medicine Unlocked, Volume 34, 2022, https://doi.org/10.1016/j.imu.2022.101113. (https://www.sciencedirect.com/science/article/pii/S2352914822002507)
## example 1 data(iris) opGMMassessment(Data = iris$Petal.Length, FitAlg = "normalmixEM", Criterion = "BIC", PlotIt = TRUE, MaxModes = 5, MaxCores = 1, Seed = 42)
## example 1 data(iris) opGMMassessment(Data = iris$Petal.Length, FitAlg = "normalmixEM", Criterion = "BIC", PlotIt = TRUE, MaxModes = 5, MaxCores = 1, Seed = 42)