Longest common subsequence: an algorithmic component analysis
Résumé
We study the performance of various algorithmic components for the longest common sequence problem (LCS). In all experiments, a simple and original anytime tree search algorithm, iterative beam search is used. A new dominance scheme for LCS, inspired by dynamic programming, is compared with two known dominance schemes: local and beam dominance. We show how to compute the probabilistic and expectation guides with high precision, using logarithms. We show that the contribution of the components to the algorithm substantially depends on the number of sequences and if the sequences are dependent or not. Out of this component analysis, we build a competitive tree search algorithm that finds new-best-known solutions on various instances of public datasets of LCS. We provide access to our computational code to facilitate further improvements.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...