Project code of clustering stochastic processes using covariance-based dissimilarity


Prerequisite:  
To run the fBm simulation code in 'main_simulation_study.m', the user needs to set up FracLab for Matlab. To download FracLab, please refer to:  
https://project.inria.fr/fraclab/download/  


Summary:
The Matlab codes contained in this repository are based on the research (see Citation) of clustering weakly stationary stochastic processes. This technique is applicable to the unsupervised learning with respect to time series that are weakly stationary.


Author: Ran Zhao (All rights reserved.)

Content: 
dist_ts_log.m: Calculation the dissimilarity between two stochastic processes by their covariance (or correlation) matrices, with logarithm transformation applied.

main_simulation_study_fBM.m: The main function produces the main conclusions as shown in the simulation studies of cited paper (see citation). The major steps include 1) simulating weakly stationary stochastic processes, 2) constructing offline dataset and performing clustering analysis, and 3) constructing online dataset and performing clustering analysis.

main_simulation_study_nonGaus.m: Similar to 'main_simulation_study_fBM.m' but uses non-Gaussian process as underlying observations. 

misclassify_rate.m: Calculate the misclassification rate of clustering results. The misclassification rate is defined as the number of mis-clustered observations divided by the total number of observations. 

scale_mean.m: Scaled arbitrate matrix's rows to specified mean value.

sim_wssp_paths.m: The function simulates one example of weakly stationary stochastic process. 

sim_wssp_paths_nonGaus.m: The function simulates one example of non-Gaussian weakly stationary stochastic process. 

unsup_wssp_offline_algo.m: This unsupervised clustering script uses offline methodology (Algorithm 1 in cited paper) to perform clustering analysis. 

unsup_wssp_online_algo.m: This unsupervised clustering script uses online methodology (Algorithm 2 in cited paper) to perform clustering analysis.


Citation:
Q. Peng, N. Rao and R. Zhao. Covariance-based Dissimilarity Measures Applied to Clustering Wide-sense Stationary Ergodic Processes. ArXiv, 2018.