AMONG Implied Constraints for Two Families of Time-Series Constraints

Ekaterina Arafailova (1, 2, 3) , Nicolas Beldiceanu (1, 2, 3) , Helmut Simonis (4, 5)

1 LS2N - Laboratoire des Sciences du Numérique de Nantes
2 LS2N - équipe TASC - Théorie, Algorithmes et Systèmes en Contraintes
3 IMT Atlantique - DAPI - Département Automatique, Productique et Informatique
4 UCC - University College Cork
5 Insight Centre for Data Analytics

We consider, for an integer time series, two families of constraints restricting the max, and the sum, respectively, of the surfaces of the elements of the sub-series corresponding to occurrences of some pattern. In recent work these families were identified as the most difficult to solve compared to all other time-series constraints. For all patterns of the time-series constraints catalogue, we provide a unique per family parameterised among implied constraint that can be imposed on any prefix/suffix of a time-series. Experiments show that it reduces both the number of backtracks/time spent by up to 4/3 orders of magnitude.

Domaines: Informatique [cs]

Ekaterina Arafailova, Nicolas Beldiceanu, Helmut Simonis. AMONG Implied Constraints for Two Families of Time-Series Constraints. CP 2017 : 23rd International Conference on Principles and Practice of Constraint Programming, Aug 2017, Melbourne, Australia. ⟨10.1007/978-3-319-66158-2_3⟩. ⟨hal-01651585⟩