On Locating Paths in Compressed Cardinal Trees
share
A compressed index is a data structure representing a text within compressed space and supporting fast indexing queries: given a pattern, count/return all positions where the pattern occurs. In recent years, powerful compressed indexes have emerged. These are based on Entropy, the Lempel-Ziv factorization, the run-length Burrows-Wheeler Transform (BWT), context-free grammars and, more recently, string attractors. Trees add a whole new dimension to the problem: one needs not only to compress the labels, but also the tree's topology. On this side, less is known. Jacobson showed how to represent the topology of a tree with n nodes in 2n+o(n) bits of space (succinct) while also supporting constant-time navigation queries. Ferragina et al. presented the first entropy-compressed labeled tree representation (the XBWT) able to count, but not locate, paths labeled with a given pattern. Grammars and the Lempel-Ziv factorization have been extended to trees, but those representations do not support indexing queries. In this paper, we extend to cardinal trees (i.e. tries) the most powerful string compression and indexing tools known to date. We start by proposing suitable generalizations of run-length BWT, high-order entropy, and string attractors to cardinal trees. We show that the number r≤ n of XBWT-runs upper-bounds the size of the smallest tree attractor and lower-bounds the trie's high-order worst-case entropy H^wc_k. The main result of this paper is the first tree index able to locate in pre-order nodes reached by a path labeled with a given pattern. Our index locates path occurrences in constant time each and takes 2n + o(n) + O(rlog n) ≤ 2n + o(n) + O( H^wc_klog n) bits of space: the reporting time is optimal and the locate machinery fits within compressed space on top of the tree's topology.
READ FULL TEXT
