Affects Version/s: 4.0-ALPHA, 4.0-BETA, 4.0, 4.1, 4.2, 4.2.1
Fix Version/s: None
I was encouraged by Mike McCandless to open an issue concerning this after I contacted him privately about it. Thanks Mike!
I'd like to submit my Java implementation of the Levenshtein Automaton as a homogenous replacement for the current heterogenous, multi-component implementation in Lucene.
Benefits of upgrading include
- Reduced code complexity
- Better performance from components that were previously implemented in Python
- Support for on-the-fly dictionary-automaton manipulation (if you wish to use my dictionary-automaton implementation)
The code for all the components is well structured, easy to follow, and extensively commented. It has also been fully tested for correct functionality and performance.
The levenshtein automaton implementation (along with the required MDAG reference) can be found in my LevenshteinAutomaton Java library here: https://github.com/klawson88/LevenshteinAutomaton.
The minimalistic directed acyclic graph (MDAG) which the automaton code uses to store and step through word sets can be found here: https://github.com/klawson88/MDAG
Transpositions aren't currently implemented. I hope the comment filled, editing-friendly code combined with the fact that the section in the Mihov paper detailing transpositions is only 2 pages makes adding the functionality trivial. Update introduces transposition inclusion in edit distance calculations!
*As a result of support for on-the-fly manipulation, the MDAG (dictionary-automaton) creation process incurs a slight speed penalty. In order to have the best of both worlds, i'd recommend the addition of a constructor which only takes sorted input. The complete, easy to follow pseudo-code for the simple procedure can be found in the first article I linked under the references section in the MDAG repository)