Dynamic Programming Based Gaussian Mixture Modelling Tool for 1D and 2D Data.
dpGMM 
R package dpGMM for dynamic programming based Gaussian mixture modelling clustering for 1D and 2D data.
Description
Package functionality:
- Variety of criteria for selection component number (AIC, AICc, BIC, ICL-BIC and LR),
- Posibility of merging the components with a small standard deviation within each other,
- Control of minimum allow variance of component to avoid picks,
- Quick stop if LR test does not show improvement (off/on) accompanied by possibility to control significance level,
- Analysis of single measurements/vector as well as binned data,
- Distribution plot with selected components in ggplot,
- QQ-plot of fitted distribution and standard normal distribution,
- Gaussian Mixture Modeling for 2D data.
Installation
You can install the package from GitHub with:
# install.packages("devtools")
devtools::install_github("ZAEDPolSl/dpGMM")
Run of simple data
At first after isnatlation load library into R and load examplary data.
library(dpGMM)
data(example)
Next, let's load GMM control parameters for 1D and change maximum number of iteration to 1000 (just to speed up).
custom.settings <- GMM_1D_opts
custom.settings$max_iter <- 1000
To run GMM for 1D vector data the following code will work.
mix_test <- runGMM(example$Dist, opts = custom.settings)
References
[1] Polanski, Andrzej, et al." "Initializing the EM algorithm for univariate Gaussian, multi-component, heteroscedastic mixture models by dynamic programming partitions." International Journal of Computational Methods 15.03 (2018): 1850012.
[2] Zyla, J., Szymala, K., Polanski, A., Polanska, J., Marczyk,. M: "dpGMM: R package for dynamic programming-based Gaussian mixture modeling of 1D and 2D data" in preparation.