Lightweight Fuzzy evaluation protocol with CollectionType extension
iOS 8.0+ Swift 3.0
struct PlayerModel {
let name: String
let position: String
let goals: Int
}Implementation of FuzzySearchable protocol defines against what search patterns will be evaluated.
extension PlayerModel: FuzzySearchable {
var fuzzyStringToMatch: String {
return name
}
}let maradona = PlayerModel(name: "Diego Maradona", position: "F", goals: 16)
maradona.fuzzyMatch("diema") // FuzzySearchResult(weight: 15, parts: [(0,3), (6,2)])Result of evaluation carries two properties:
weight- weight of the matchparts-NSRange's of pattern matching againstfuzzyStringToMatch
When evaluating collection of FuzzySearchables result is an array of tuples (item: Generator.Element, result: FuzzySearchResult) which is filtered and sorted depending on weight.
let players = [
PlayerModel(name: "Diego Maradona", position: "CF", goals: 16),
PlayerModel(name: "David Beckham", position: "CAM", goals: 8),
PlayerModel(name: "Lionel Messi", position: "RW", goals: 15)
]
players.fuzzyMatch("di")
//[
// (
// FuzzySearchTests.PlayerModel(name: "Diego Maradona", position: "CF", goals: 16),
// FuzzySearch.FuzzySearchResult(weight: 4, parts: [(0,2)])
// ), (
// FuzzySearchTests.PlayerModel(name: "David Beckham", position: "CAM", goals: 8),
// FuzzySearch.FuzzySearchResult(weight: 2, parts: [(0,1), (3,1)])
// )
//]Wraps over a FuzzySearchable instance, caching some underlying metadata generated while fuzzy-matching w/ FuzzySearchable.fuzzyMatch.
Use this cached wrapper over FuzzySearchable instances that are expected to be fuzzy-matched multiple times without mutation to fuzzyStringToMatch:
let players = [
PlayerModel(name: "Diego Maradona", position: "CF", goals: 16),
PlayerModel(name: "David Beckham", position: "CAM", goals: 8),
PlayerModel(name: "Lionel Messi", position: "RW", goals: 15),
// Many more players ...
]
let fuzzyCachedPlayers =
players.map { player in
CachedFuzzySearchable(wrapping: player)
}
fuzzyCachedPlayers.fuzzyMatch("di")
// Subsequente calls to 'fuzzyMatch' over the array above cause less overhead when re-matching
fuzzyCachedPlayers.fuzzyMatch("di") // Runs in about half time
players.fuzzyMatch("di") // Will fuzzy-match against original, non-cached PlayerModel values.CachedFuzzySearchable has storage needs of a little over 2x that of the fuzzyStringToSearch property of the wrapped FuzzySearchable value. Discarding a CachedFuzzySearchable value also discards the extra memory that was allocated.
Returning a different value from the wrapped FuzzySearchable's fuzzyStringToSearch property resets the cache automatically during the next fuzzyMatch call and the overhead is reset to that of a fresh CachedFuzzySearchable instance, so implementers shouldn't worry about mantaining a stable fuzzyStringToSearch return.
FuzzySearch is released under the MIT license.