Skip to contents

R package rEMM - Extensible Markov Model for Modelling Temporal Relationships Between Clusters

Package on CRAN CRAN RStudio mirror downloads Licenser-universe status

Maintainer: Michael Hahsler

Implements TRACDS (Temporal Relationships between Clusters for Data Streams), a generalization of the Extensible Markov Model (EMM), to model transition probabilities in sequence data (Hahsler and Dunham 2011). TRACDS adds a temporal or order model to data stream clustering by superimposing a dynamically adapting Markov chain. It also provides an implementation of EMM (TRACDS on top of tNN data stream clustering) (Hahsler and Dunham 2010).

The package also provides interface classes DSC_tNN and DSC_EMM for data stream mining with stream (Hahsler et al. 2017).

Installation

Stable CRAN version: Install from within R with

Current development version: Install from r-universe.

install.packages("rEMM",
    repos = c("https://mhahsler.r-universe.dev",
              "https://cloud.r-project.org/"))

Usage

We use an artificial dataset with four clusters. Points are generated using the fixed sequence <1,2,1,3,4> through the four clusters. The lines below indicate the sequence.

library(rEMM)

data("EMMsim")

plot(EMMsim_train, pch = NA)
lines(EMMsim_train, col = "gray")
points(EMMsim_train, pch = EMMsim_sequence_train)

EMM recovers the components and sequence information. We use EMM and then recluster the learned structure, assuming that we know there are four components. The graph below represents a Markov model of the sequence.

emm <- EMM(threshold = 0.1, measure = "euclidean")
build(emm, EMMsim_train)
emmc <- recluster_hclust(emm, k = 4, method = "average")
plot(emmc, method = "MDS")

We can now score new sequences. Here, we use a test sequence created in the same way as the training data and calculate the product of the transition probabilities in the model. A high score indicates a good match.

score(emmc, EMMsim_test)
## [1] 0.71

Vignettes

Acknowledgements

Development of this package was supported in part by NSF IIS-0948893 and R21HG005912 from the National Human Genome Research Institute.

References

Hahsler, Michael, Matthew Bolaños, and John Forrest. 2017. “Introduction to Stream: An Extensible Framework for Data Stream Clustering Research with r.” Journal of Statistical Software 76 (14): 1–50. https://doi.org/10.18637/jss.v076.i14.
Hahsler, Michael, and Margaret H. Dunham. 2010. “rEMM: Extensible Markov Model for Data Stream Clustering in R.” Journal of Statistical Software 35 (5): 1–31. https://doi.org/10.18637/jss.v035.i05.
Hahsler, Michael, and Margaret H. Dunham. 2011. “Temporal Structure Learning for Clustering Massive Data Streams in Real-Time.” Proceedings of the 2011 SIAM International Conference on Data Mining, 664–75. https://doi.org/10.1137/1.9781611972818.57.