danieldk / seqalign

Sequence alignments

seqalign

Introduction

This crate implements commonly-used sequence alignment methods based on edit operations. There are multiple crates available to compute edit distances. However, to my knowledge there was no crate that supports all of the following seqalign features:

• Works on slices of any type.
• Can return both the edit distance and the edit script/alignment.
• Can be extended with new measures.

Example

use seqalign::Align;
use seqalign::measures::LevenshteinDamerau;

let incorrect = &['t', 'p', 'y', 'o'];
let correct = &['t', 'y', 'p', 'o', 's'];

let measure = LevenshteinDamerau::new(1, 1, 1, 1);
let alignment = measure.align(incorrect, correct);

// Get the edit distance
assert_eq!(2, alignment.distance());

// Get the edit script.
use seqalign::measures::LevenshteinDamerauOp;
use seqalign::op::IndexedOperation;

assert_eq!(vec![
  	IndexedOperation::new(LevenshteinDamerauOp::Match, 0, 0),
  	IndexedOperation::new(LevenshteinDamerauOp::Transpose(1), 1, 1),
  	IndexedOperation::new(LevenshteinDamerauOp::Match, 3, 3),
  	IndexedOperation::new(LevenshteinDamerauOp::Insert(1), 4, 4)
  ], alignment.edit_script());

seqalign's Issues

Update documentation with Levensthein-Damerau properties

The documentation of the Levenshtein-Damerau distance should state that this computes the restricted edit distance and therefore does not satisfy the triangle inequality (is not a metric distance).

Add a method that returns all alignments

We currently provide a single alignment, but there are often edit scripts that lead to the lowest cost. Add a separate method that returns all alignments.

Non-integer costs

The Alignment struct now uses a Vec of usize to store alignments. However, some literature uses floating point costs (e.g. for align strings with perceived distances between phonemes). Add support for measures that use non-integer costs.