Asymptotic behaviour of a BIPF algorithm with an improper target

Claudio Asci; Mauro Piccioni

Kybernetika (2009)

  • Volume: 45, Issue: 2, page 169-188
  • ISSN: 0023-5954

Abstract

top
The BIPF algorithm is a Markovian algorithm with the purpose of simulating certain probability distributions supported by contingency tables belonging to hierarchical log-linear models. The updating steps of the algorithm depend only on the required expected marginal tables over the maximal terms of the hierarchical model. Usually these tables are marginals of a positive joint table, in which case it is well known that the algorithm is a blocking Gibbs Sampler. But the algorithm makes sense even when these marginals do not come from a joint table. In this case the target distribution of the algorithm is necessarily improper. In this paper we investigate the simplest non trivial case, i. e. the 2 × 2 × 2 hierarchical interaction. Our result is that the algorithm is asymptotically attracted by a limit cycle in law.

How to cite

top

Asci, Claudio, and Piccioni, Mauro. "Asymptotic behaviour of a BIPF algorithm with an improper target." Kybernetika 45.2 (2009): 169-188. <http://eudml.org/doc/37729>.

@article{Asci2009,
abstract = {The BIPF algorithm is a Markovian algorithm with the purpose of simulating certain probability distributions supported by contingency tables belonging to hierarchical log-linear models. The updating steps of the algorithm depend only on the required expected marginal tables over the maximal terms of the hierarchical model. Usually these tables are marginals of a positive joint table, in which case it is well known that the algorithm is a blocking Gibbs Sampler. But the algorithm makes sense even when these marginals do not come from a joint table. In this case the target distribution of the algorithm is necessarily improper. In this paper we investigate the simplest non trivial case, i. e. the $2\times 2\times 2$ hierarchical interaction. Our result is that the algorithm is asymptotically attracted by a limit cycle in law.},
author = {Asci, Claudio, Piccioni, Mauro},
journal = {Kybernetika},
keywords = {log-linear models; marginal problem; null Markov chains; log-linear models; marginal problem; null Markov chains},
language = {eng},
number = {2},
pages = {169-188},
publisher = {Institute of Information Theory and Automation AS CR},
title = {Asymptotic behaviour of a BIPF algorithm with an improper target},
url = {http://eudml.org/doc/37729},
volume = {45},
year = {2009},
}

TY - JOUR
AU - Asci, Claudio
AU - Piccioni, Mauro
TI - Asymptotic behaviour of a BIPF algorithm with an improper target
JO - Kybernetika
PY - 2009
PB - Institute of Information Theory and Automation AS CR
VL - 45
IS - 2
SP - 169
EP - 188
AB - The BIPF algorithm is a Markovian algorithm with the purpose of simulating certain probability distributions supported by contingency tables belonging to hierarchical log-linear models. The updating steps of the algorithm depend only on the required expected marginal tables over the maximal terms of the hierarchical model. Usually these tables are marginals of a positive joint table, in which case it is well known that the algorithm is a blocking Gibbs Sampler. But the algorithm makes sense even when these marginals do not come from a joint table. In this case the target distribution of the algorithm is necessarily improper. In this paper we investigate the simplest non trivial case, i. e. the $2\times 2\times 2$ hierarchical interaction. Our result is that the algorithm is asymptotically attracted by a limit cycle in law.
LA - eng
KW - log-linear models; marginal problem; null Markov chains; log-linear models; marginal problem; null Markov chains
UR - http://eudml.org/doc/37729
ER -

References

top
  1. Quantum Mechanics and Experience, Harvard College, Cambridge 1992. MR1221080
  2. Beta-hypergeometric distributions and random continued fractions, Statist. Probab. Lett. 78 (2008), 1711–1721. MR2453912
  3. The IPF algorithm when the marginal problem is unsolvable: the simplest case, Kybernetika 39 (2003), 731–737. MR2035647
  4. Functionally compatible local characteristics for the local specification of priors in graphical models, Scand. J. Statist. 34 (2007), 829–840. MR2396941
  5. I-divergence geometry of probability distributions and minimization problems, Ann. Probab. 3 (1975), 146–158. MR0365798
  6. Hyper-Markov laws in the statistical analysis of decomposable graphical models, Ann. Statist. 21 (1993), 1272–1317. MR1241267
  7. The art of data augmentation, With discussions, and a rejoinder by the authors. J. Comput. Graph. Statist. 10 (2001), 1–111. MR1936358
  8. Can the quantum-mechanical description of physical reality be considered complete? Phys, Rev. 47 (1935), 777–780. 
  9. Bayesian Data Analysis, Chapman and Hall, London 1995. MR1385925
  10. Functional compatibility, Markov chains, and Gibbs sampling with improper posteriors, J. Comput. Graph. Statist. 7 (1998), 42–60. MR1628268
  11. Graphical Models, Clarendon Press, Oxford 1996. Zbl1055.62126MR1419991
  12. Chain graph models and their causal interpretations (with discussion), J. Roy. Statist. Soc. Ser. B 64 (2002), 321–348. MR1924296
  13. Markov Chains and Stochastic Stability, Springer-Verlag, London 1993. MR1287609
  14. Independence structure of natural conjugate densities to exponential families and the Gibbs’ sampler, Scand. J. Statist. 27 (2000), 111–127. Zbl0938.62025MR1774047
  15. Special Functions, MacMillan, New York 1960. Zbl0231.33001MR0107725
  16. Analysis of Incomplete Multivariate Data, Chapman and Hall, London 1997. Zbl0997.62510MR1692799
  17. R-theory for Markov chains on a general state space I, II, Ann. Probab. 2 (1974), 840–878. MR0368151
  18. Weighing the Odds, Cambridge University Press, Cambridge 2001. Zbl0984.62001MR1854128

NotesEmbed ?

top

You must be logged in to post comments.