Abstract
Markov chains with variable length are useful parsimonious stochastic models able to generate most stationary sequence of discrete symbols. The idea is to identify the suffixes of the past, called contexts, that are relevant to predict the future symbol. Sometimes a single state is a context, and looking at the past and finding this specific state makes the further past irrelevant. States with such property are called renewal states and they can be used to split the chain into independent and identically distributed blocks. In order to identify renewal states for chains with variable length, we propose the use of Intrinsic Bayes Factor to evaluate the hypothesis that some particular state is a renewal state. In this case, the difficulty lies in integrating the marginal posterior distribution for the random context trees for general prior distribution on the space of context trees, with Dirichlet prior for the transition probabilities, and Monte Carlo methods are applied. To show the strength of our method, we analyzed artificial datasets generated from different models and one example coming from the field of Linguistics.
This is a preview of subscription content, access via your institution.
References
Aitkin, M.: Posterior bayes factors. J. R. Stat. Soc. Ser. B (Methodol.) 53(1), 111–128 (1991)
Aitkin, M.: Posterior Bayes factor analysis for an exponential regression model. Stat. Comput. 3(1), 17–22 (1993)
Aitkin, M., Finch, S., Mendell, N., et al.: A new test for the presence of a normal mixture distribution based on the posterior Bayes factor. Stat. Comput. 6(2), 121–125 (1996)
Balding, D., Ferrari, P.A., Fraiman, R., et al.: Limit theorems for sequences of random trees. Test 18(2), 302–315 (2009)
Berger, J.O., Pericchi, L.R.: The intrinsic Bayes factor for model selection and prediction. J. Am. Stat. Assoc. 91(433), 109–122 (1996)
Berger, J.O., Bernardo, J.M., Sun, D.: Overall objective priors. Bayesian Anal. 10(1), 189–221 (2015)
Bühlmann, P., Wyner, A.J.: Variable length Markov chains. Ann. Stat. 27(2), 480–513 (1999)
Busch, J.R., Ferrari, P.A., Flesia, A.G., et al.: Testing statistical hypothesis on random trees and applications to the protein classification problem. Ann. Appl. Stat. 3(2), 542–563 (2009)
Cabras, S., Castellanos, M.E., Perra, S.: A new minimal training sample scheme for intrinsic Bayes factors in censored data. Comput. Stat. Data Anal. 81, 52–63 (2015)
Charitidou, E., Fouskakis, D., Ntzoufras, I.: Objective Bayesian transformation and variable selection using default Bayes factors. Stat. Comput. 28(3), 579–594 (2018)
Chib, S., Greenberg, E.: Understanding the Metropolis–Hastings algorithm. Am. Stat. 49(4), 327–335 (1995)
Csiszár, I., Talata, Z.: Context tree estimation for not necessarily finite memory processes, via BIC and MDL. IEEE Trans. Inf. Theory 52(3), 1007–1016 (2006)
Dimitrakakis, C.: Bayesian variable order Markov models. In: Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, JMLR Workshop and Conference Proceedings, pp. 161–168 (2010)
Galves, A, Löcherbach, E.: Stochastic chains with memory of variable length. In: Grünwald et al. (eds.) Festschrift for Jorma Rissanen, TICSP Series, vol. 38, pp. 117–133 (2008)
Galves, C., Moraes, M.A.T., Ribeiro, I.: Syntax and morphology in the placement of clitics in European and Brazilian Portuguese. J. Port. Linguist. 4(2) (2005)
Galves, A., Galves, C., Garcia, J.E., et al.: Context tree selection and linguistic rhythm retrieval from written texts. Ann. Appl. Stat. 6(1), 186–209 (2012)
Hastings, W.K.: Monte carlo sampling methods using Markov chains and their applications. Biometrika (1970)
Kass, R.E., Raftery, A.E.: Bayes factors. J. Am. Stat. Assoc. 90(430), 773–795 (1995)
Kennel, M.B., Mees, A.I.: Context-tree modeling of observed symbolic dynamics. Phys. Rev. E 66(5), 056209 (2002)
Kontoyiannis, I., Mertzanis, L., Panotopoulou, A. et al.: Bayesian context trees: modelling and exact inference for discrete time series. J. R. Stat. Soc. Ser. B (Methodol.) 1–37 (2022)
Madigan, D., York, J., Allard, D.: Bayesian graphical models for discrete data. Int. Stat. Rev./Rev. Int. de Stat. 215–232 (1995)
O’Hagan, A.: Fractional Bayes factors for model comparison. J. R. Stat. Soc. Ser. B (Methodol.) 57(1), 99–118 (1995)
Rissanen, J.: A universal data compression system. IEEE Trans. Inf. Theory 29(5), 656–664 (1983)
Ron, D., Singer, Y., Tishby, N.: The power of amnesia: learning probabilistic automata with variable memory length. Mach. Learn. 25(2), 117–149 (1996)
Smith, W.L.: Regenerative stochastic processes. Proc. R. Soc. Lond. A 232(1188), 6–31 (1955)
Steuer, R., Ebeling, W., Russell, D., et al.: Entropy and local uncertainty of data from sensory neurons. Phys. Rev. E 64(6), 061911 (2001)
Villa, C., Walker, S.G.: An objective Bayes factor with improper priors. Comput. Stat. Data Anal. 107404 (2021)
Willems, F.M., Shtarkov, Y.M., Tjalkens, T.J.: The context-tree weighting method: basic properties. IEEE Trans. Inf. Theory 41(3), 653–664 (1995)
Xiong, J., Jääskinen, V., Corander, J.: Recursive learning for sparse Markov models. Bayesian Anal. 11(1), 247–263 (2016)
Acknowledgements
This work was funded by Fundação de Amparo à Pesquisa do Estado de São Paulo - FAPESP Grants 2017/25469-2 and 2017/10555-0, CNPq Grant 304148/2020-2 and by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001. We thank Helio Migon and Alexandra Schmidt for fruitful discussions.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Appendices
Simulation results
We present the distribution of the PBF values used to compute the GIBF results in the simulation studies from Sect. 4 in Figs. 8, 9, 10, 11 and 12. Each point represents one PBF computed in \(\log _{10}\) scale, with black points being considered in both 10%-trimmed and untrimmed results, while light-gray points are not considered for in the trimmed GIBF value.
Distribution of computed PBF values for algorithm runs on simulations using Model 3
Distribution of computed PBF values for algorithm runs on simulations using Model 4
Distribution of computed PBF values for algorithm runs on simulations using Model 5
Figure 11 is a good example of why the arithmetic mean of PBFs (AIBF) may be very unstable, as even in the trimming version a few values may be as high as \(10^5\), which completely dominates the arithmetic mean (we recall that at most 100 values are used in the average), not taking into account the fact that most values are lower than \(10^{-5}\) (evidence against the renewal hypothesis).
Application results
We present the distribution of the PBF values used to compute the GIBF results in the simulations from Sect. 5 in Fig. 13. The interpretation of the black and light-gray points are the same as in “Appendix B”, but the trimming considered is 5% as specified in the application description.
Distribution of computed PBF values for algorithm runs for the application dataset
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Freguglia, V., Garcia, N.L. Detecting renewal states in chains of variable length via intrinsic Bayes factors. Stat Comput 33, 21 (2023). https://doi.org/10.1007/s11222-022-10191-2
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11222-022-10191-2
Keywords
- Variable length Markov Chains
- Renewal states
- Bayes factor
- Intractable normalizing constant