From temporal data to dynamic causal models
Abstract
We present a brief review of dynamic causal model inference from data. A vector autoregressive models is of our prime interest. The architecture, representation and schemes of measurement of temporal data and time series data are outlined. We argue that require- ment to data characteristics should come from the nature of dynamic process at hand and goals of model inference. To describe and evaluate temporal data one may use terms of longitude, measurement frequency etc. Data measurement frequency is crucial factor in order to an inferred model be adequate. Data longitude and observation session duration may be expressed via several temporal horizons, such as closest horizon, 2-step horizon, influence attainability horizon, oscillatory horizon, and evolutionary horizon. To justify a dynamic causal model inference from data, analyst needs to assume the dynamic process is stationary or at least obeys structural regularity. The main specificity of task of dynamic causal model inference is known temporal order of variables and certain structural regularity. If maximal lag of influence is unknown, inference of dynamic causal model faces additional problems. We examine the Granger’s causality concept and outline its deficiency in real circumstances. It is argued that Granger causality is incorrect as practical tool of causal discovery. In contrast, certain rules of edge orientation (included in known constraint-based algorithms of model inference) can reveal unconfounded causal relationship.
Prombles in programming 2022; 3-4: 183-195
Keywords
Full Text:
PDF (Українська)References
Handbook of Big Data. P. Bühlmann, P. Drineas, M. Kane, M. van der Laan (Eds.). Boca Raton, FL.: CRC Press. Taylor & Francis Group, 2016. 452p.
Balabanov O.S. (2020) Tasks and methods of Big Data analysis (a survey). (Revised). (preprint at ResearchGate) DOI: 10.13140/ RG.2.2.18586.39367 [in Ukrainian]
Pearl J. Causality: models, reasoning, and inference. Cambridge: Cambridge Univ. Press, 2000. 526 p.
Spirtes P., Glymour C. and Scheines R. Causation, prediction and search. New York: MIT Press, 2001. 543 p. CrossRef
Spirtes P., Zhang K. (2016) Causal discovery and inference: concepts and recent methodological advances. Applied Informatics. V.3: 3. 28 p.CrossRef
Zhang J. (2008) On the completeness of orientation rules for causal discovery in the presence of latent confounders and selection bias. Artificial Intelligence. V. 172. P. 1873-1896. CrossRef
Balabanov O.S. (2022). Logic of causal inference from data under presence of latent confounders. Cybernetics and Systems Analysis. V. 58. No. 2, P. 171-185. CrossRef
Aalen O.O., Borgan O., Gjessing H.K. Survival and Event History Analysis. A Process Point of View. Springer, New York, 2008, 539 p. CrossRef
Lütkepohl H. New introduction to multiple time series analysis. Springer-Verlag, 2005, 764 p. CrossRef
Shumway R.H., Stoffer D.S. Time series analysis and its applications with R examples. Springer, 2011, 596 p. CrossRef
Kulkarni V.G. Introduction to modeling and analysis of stochastic systems (2nd Ed.) Springer 2011. 313 p. CrossRef
Schweder T. (1970) Composable Markov processes. J. Appl. Probab. V. 7. P. 400-410. CrossRef
Reconstructing regime-dependent causal relationship from observational time series / E. Saggioro, J. de Wiljes, M. Kretschmer, J. Runge - Chaos. (2020). V. 30 (n.11). 113115. -22p. ISSN 1089-7682 CrossRef
Gong M., Zhang K., Schölkopf B., Tao D. and Geiger P. (2015) Discovering temporal causal relations from subsampled data. Proc. of the 32nd Intern. Conf. on Machine Learning, P. 1898-1906.
Plis S., Danks D., Freeman C., and Calhoun V. Rate-agnostic (causal) structure learning. In: Advances in Neural Information Processing Systems. 2015. P. 3303-3311.
Granger C.V.J. (1980) Testing for causality. A personal viewpoint. Journal of Economic Dynamics and Control. V. 2. issue 1, P. 329-352. CrossRef
Granger C.V.J. (1969) Investigating causal relations by econometric models and cross-spectral methods. Econometrica. V. 37. P. 424-459.CrossRef
Swanson N.R. and Granger C.W.J. (1997) Impulse response functions based on a causal approach to residual orthogonalization in vector autoregressions. J. of the American Statistical Association. V. 92. N 437. P. 357-367. CrossRef
Malinsky D. and Spirtes P. (2019) Learning the structure of a nonstationary vector autoregression. The 22nd Intern. Conf. on Artificial Intelligence and Statistics. - Proc. of Machine Learning Research, PMLR. V. 89. P. 2986-2994.
Entner D. and Hoyer P.O. (2010) On causal discovery from time series data using FCI. Proc. of the 5th European Workshop on Probabilistic graphical models. Helsinki, Finland. P. 121-128.
Runge J. (2018) Causal network reconstruction from timeseries: From theoretical assumptions to practical estimation. Chaos. V. 28, paper 075310. 20 p. CrossRef
Malinsky D. and Spirtes P. (2018) Causal structure learning from multivariate time series in settings with unmeasured confounding. Proc. of 2018 ACM SIGKDD Workshop on Causal Discovery, London, UK. - PMLR, V. 92. P. 23-47.
DOI: https://doi.org/10.15407/pp2022.03-04.183
Refbacks
- There are currently no refbacks.