mirror of https://github.com/apache/lucene.git
hunspell: speed up "compress"; minimize the number of the generated entries; don't even consider "forbidden" entries anymore (#13429)
Hunspell: speed up "compress"; minimize the number of the generated entries; don't even consider "forbidden" entries anymore
This commit is contained in:
Peter Gromov
GitHub
parent
4438404457
commit
54d3ff64bc
|
|
@ -309,6 +309,8 @@ Improvements
|
|||
|
||||
* GITHUB#13276: UnifiedHighlighter: new 'passageSortComparator' option to allow sorting other than offset order. (Seunghan Jung)
|
||||
|
||||
* GITHUB#13429: Hunspell: speed up "compress"; minimize the number of the generated entries; don't even consider "forbidden" entries anymore (Peter Gromov)
|
||||
|
||||
Optimizations
|
||||
---------------------
|
||||
|
||||
|
|
|
|||
|
|
@ -384,27 +384,30 @@ public class WordFormGenerator {
|
|||
|
||||
private class WordCompressor {
|
||||
private final Comparator<State> solutionFitness =
|
||||
Comparator.comparingInt((State s) -> s.forbidden)
|
||||
.thenComparingInt(s -> s.underGenerated)
|
||||
Comparator.comparingInt((State s) -> -s.potentialCoverage)
|
||||
.thenComparingInt(s -> s.stemToFlags.size())
|
||||
.thenComparingInt(s -> s.underGenerated)
|
||||
.thenComparingInt(s -> s.overGenerated);
|
||||
private final Set<String> forbidden;
|
||||
private final Runnable checkCanceled;
|
||||
private final Set<String> wordSet;
|
||||
private final Set<String> existingStems;
|
||||
private final Map<String, Set<FlagSet>> stemToPossibleFlags = new HashMap<>();
|
||||
private final Map<String, Integer> stemCounts = new LinkedHashMap<>();
|
||||
private final Map<String, Set<String>> stemsToForms = new LinkedHashMap<>();
|
||||
|
||||
WordCompressor(List<String> words, Set<String> forbidden, Runnable checkCanceled) {
|
||||
this.forbidden = forbidden;
|
||||
this.checkCanceled = checkCanceled;
|
||||
wordSet = new HashSet<>(words);
|
||||
|
||||
Stemmer.StemCandidateProcessor processor =
|
||||
for (String word : words) {
|
||||
checkCanceled.run();
|
||||
stemToPossibleFlags.computeIfAbsent(word, __ -> new LinkedHashSet<>());
|
||||
var processor =
|
||||
new Stemmer.StemCandidateProcessor(WordContext.SIMPLE_WORD) {
|
||||
@Override
|
||||
boolean processStemCandidate(
|
||||
char[] word,
|
||||
char[] chars,
|
||||
int offset,
|
||||
int length,
|
||||
int lastAffix,
|
||||
|
|
@ -412,29 +415,34 @@ public class WordFormGenerator {
|
|||
int innerPrefix,
|
||||
int outerSuffix,
|
||||
int innerSuffix) {
|
||||
String candidate = new String(word, offset, length);
|
||||
stemCounts.merge(candidate, 1, Integer::sum);
|
||||
String candidate = new String(chars, offset, length);
|
||||
CharHashSet flags = new CharHashSet();
|
||||
if (outerPrefix >= 0) flags.add(dictionary.affixData(outerPrefix, AFFIX_FLAG));
|
||||
if (innerPrefix >= 0) flags.add(dictionary.affixData(innerPrefix, AFFIX_FLAG));
|
||||
if (outerSuffix >= 0) flags.add(dictionary.affixData(outerSuffix, AFFIX_FLAG));
|
||||
if (innerSuffix >= 0) flags.add(dictionary.affixData(innerSuffix, AFFIX_FLAG));
|
||||
FlagSet flagSet = new FlagSet(flags, dictionary);
|
||||
StemWithFlags swf = new StemWithFlags(candidate, Set.of(flagSet));
|
||||
if (forbidden.isEmpty()
|
||||
|| allGenerated(swf).stream().noneMatch(forbidden::contains)) {
|
||||
registerStem(candidate);
|
||||
stemToPossibleFlags
|
||||
.computeIfAbsent(candidate, __ -> new LinkedHashSet<>())
|
||||
.add(new FlagSet(flags, dictionary));
|
||||
.add(flagSet);
|
||||
}
|
||||
return true;
|
||||
}
|
||||
};
|
||||
|
||||
for (String word : words) {
|
||||
checkCanceled.run();
|
||||
stemCounts.merge(word, 1, Integer::sum);
|
||||
stemToPossibleFlags.computeIfAbsent(word, __ -> new LinkedHashSet<>());
|
||||
void registerStem(String stem) {
|
||||
stemsToForms.computeIfAbsent(stem, __ -> new LinkedHashSet<>()).add(word);
|
||||
}
|
||||
};
|
||||
processor.registerStem(word);
|
||||
stemmer.removeAffixes(word.toCharArray(), 0, word.length(), true, -1, -1, -1, processor);
|
||||
}
|
||||
|
||||
existingStems =
|
||||
stemCounts.keySet().stream()
|
||||
stemsToForms.keySet().stream()
|
||||
.filter(stem -> dictionary.lookupEntries(stem) != null)
|
||||
.collect(Collectors.toSet());
|
||||
}
|
||||
|
|
@ -442,30 +450,49 @@ public class WordFormGenerator {
|
|||
EntrySuggestion compress() {
|
||||
Comparator<String> stemSorter =
|
||||
Comparator.comparing((String s) -> existingStems.contains(s))
|
||||
.thenComparing(stemCounts::get)
|
||||
.thenComparing(s -> stemsToForms.get(s).size())
|
||||
.reversed();
|
||||
List<String> sortedStems = stemCounts.keySet().stream().sorted(stemSorter).toList();
|
||||
List<String> sortedStems = stemsToForms.keySet().stream().sorted(stemSorter).toList();
|
||||
PriorityQueue<State> queue = new PriorityQueue<>(solutionFitness);
|
||||
Set<Map<String, Set<FlagSet>>> visited = new HashSet<>();
|
||||
queue.offer(new State(Map.of(), wordSet.size(), 0, 0));
|
||||
State result = null;
|
||||
while (!queue.isEmpty()) {
|
||||
State state = queue.poll();
|
||||
if (state.underGenerated == 0) {
|
||||
if (result == null || solutionFitness.compare(state, result) < 0) result = state;
|
||||
if (state.forbidden == 0) break;
|
||||
continue;
|
||||
result = state;
|
||||
break;
|
||||
}
|
||||
|
||||
for (String stem : sortedStems) {
|
||||
if (!state.stemToFlags.containsKey(stem)) {
|
||||
queue.offer(addStem(state, stem));
|
||||
var withStem = addStem(state, stem);
|
||||
if (visited.add(withStem)) {
|
||||
var next = newState(withStem);
|
||||
if (next != null
|
||||
&& (state.underGenerated > next.underGenerated
|
||||
|| next.potentialCoverage > state.potentialCoverage)) {
|
||||
queue.offer(next);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (state.potentialCoverage < wordSet.size()) {
|
||||
// don't add flags until the suggested entries can potentially cover all requested forms
|
||||
continue;
|
||||
}
|
||||
|
||||
for (Map.Entry<String, Set<FlagSet>> entry : state.stemToFlags.entrySet()) {
|
||||
for (FlagSet flags : stemToPossibleFlags.get(entry.getKey())) {
|
||||
if (!entry.getValue().contains(flags)) {
|
||||
queue.offer(addFlags(state, entry.getKey(), flags));
|
||||
var withFlags = addFlags(state, entry.getKey(), flags);
|
||||
if (visited.add(withFlags)) {
|
||||
var next = newState(withFlags);
|
||||
if (next != null && state.underGenerated > next.underGenerated) {
|
||||
queue.offer(next);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -482,7 +509,7 @@ public class WordFormGenerator {
|
|||
|
||||
List<String> extraGenerated = new ArrayList<>();
|
||||
for (String extra : allGenerated(state.stemToFlags).distinct().sorted().toList()) {
|
||||
if (wordSet.contains(extra)) continue;
|
||||
if (wordSet.contains(extra) || existingStems.contains(extra)) continue;
|
||||
|
||||
if (forbidden.contains(extra) && dictionary.forbiddenword != FLAG_UNSET) {
|
||||
addEntry(toEdit, toAdd, extra, CharHashSet.from(dictionary.forbiddenword));
|
||||
|
|
@ -500,39 +527,55 @@ public class WordFormGenerator {
|
|||
(existingStems.contains(stem) ? toEdit : toAdd).add(DictEntry.create(stem, flagString));
|
||||
}
|
||||
|
||||
private State addStem(State state, String stem) {
|
||||
LinkedHashMap<String, Set<FlagSet>> stemToFlags = new LinkedHashMap<>(state.stemToFlags);
|
||||
private Map<String, Set<FlagSet>> addStem(State state, String stem) {
|
||||
Map<String, Set<FlagSet>> stemToFlags = new LinkedHashMap<>(state.stemToFlags);
|
||||
stemToFlags.put(stem, Set.of());
|
||||
return newState(stemToFlags);
|
||||
return stemToFlags;
|
||||
}
|
||||
|
||||
private State addFlags(State state, String stem, FlagSet flags) {
|
||||
LinkedHashMap<String, Set<FlagSet>> stemToFlags = new LinkedHashMap<>(state.stemToFlags);
|
||||
private Map<String, Set<FlagSet>> addFlags(State state, String stem, FlagSet flags) {
|
||||
Map<String, Set<FlagSet>> stemToFlags = new LinkedHashMap<>(state.stemToFlags);
|
||||
Set<FlagSet> flagSets = new LinkedHashSet<>(stemToFlags.get(stem));
|
||||
flagSets.add(flags);
|
||||
stemToFlags.put(stem, flagSets);
|
||||
return newState(stemToFlags);
|
||||
return stemToFlags;
|
||||
}
|
||||
|
||||
private State newState(Map<String, Set<FlagSet>> stemToFlags) {
|
||||
Set<String> allGenerated = allGenerated(stemToFlags).collect(Collectors.toSet());
|
||||
int overGenerated = 0;
|
||||
for (String s : allGenerated) {
|
||||
if (forbidden.contains(s)) return null;
|
||||
if (!wordSet.contains(s)) overGenerated++;
|
||||
}
|
||||
|
||||
int potentialCoverage =
|
||||
(int)
|
||||
stemToFlags.keySet().stream()
|
||||
.flatMap(s -> stemsToForms.get(s).stream())
|
||||
.distinct()
|
||||
.count();
|
||||
return new State(
|
||||
stemToFlags,
|
||||
(int) wordSet.stream().filter(s -> !allGenerated.contains(s)).count(),
|
||||
(int) allGenerated.stream().filter(s -> !wordSet.contains(s)).count(),
|
||||
(int) allGenerated.stream().filter(s -> forbidden.contains(s)).count());
|
||||
overGenerated,
|
||||
potentialCoverage);
|
||||
}
|
||||
|
||||
private final Map<StemWithFlags, List<String>> expansionCache = new HashMap<>();
|
||||
|
||||
private record StemWithFlags(String stem, Set<FlagSet> flags) {}
|
||||
|
||||
private Stream<String> allGenerated(Map<String, Set<FlagSet>> stemToFlags) {
|
||||
private List<String> allGenerated(StemWithFlags swc) {
|
||||
Function<StemWithFlags, List<String>> expandToWords =
|
||||
e -> expand(e.stem, FlagSet.flatten(e.flags)).stream().map(w -> w.getWord()).toList();
|
||||
return expansionCache.computeIfAbsent(swc, expandToWords);
|
||||
}
|
||||
|
||||
private Stream<String> allGenerated(Map<String, Set<FlagSet>> stemToFlags) {
|
||||
return stemToFlags.entrySet().stream()
|
||||
.map(e -> new StemWithFlags(e.getKey(), e.getValue()))
|
||||
.flatMap(swc -> expansionCache.computeIfAbsent(swc, expandToWords).stream());
|
||||
.flatMap(
|
||||
entry -> allGenerated(new StemWithFlags(entry.getKey(), entry.getValue())).stream());
|
||||
}
|
||||
|
||||
private List<AffixedWord> expand(String stem, CharHashSet flagSet) {
|
||||
|
|
@ -561,5 +604,7 @@ public class WordFormGenerator {
|
|||
Map<String, Set<FlagSet>> stemToFlags,
|
||||
int underGenerated,
|
||||
int overGenerated,
|
||||
int forbidden) {}
|
||||
|
||||
// The maximum number of requested forms possibly generated by adding only flags to this state
|
||||
int potentialCoverage) {}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -213,7 +213,7 @@ public class TestHunspell extends LuceneTestCase {
|
|||
Hunspell h = loadNoTimeout("base");
|
||||
String[] createQuery = {"create", "created", "creates", "creating", "creation"};
|
||||
checkCompression(h, "toEdit=[create/DGNS], toAdd=[], extra=[]", createQuery);
|
||||
checkCompression(h, "toEdit=[created], toAdd=[creates], extra=[]", "creates", "created");
|
||||
checkCompression(h, "toEdit=[create/DS], toAdd=[], extra=[]", "creates", "created");
|
||||
checkCompression(h, "toEdit=[], toAdd=[creation/S], extra=[]", "creation", "creations");
|
||||
checkCompression(h, "toEdit=[], toAdd=[abc, def], extra=[]", "abc", "def");
|
||||
checkCompression(h, "toEdit=[], toAdd=[form/S], extra=[]", "form", "forms");
|
||||
|
|
@ -227,6 +227,20 @@ public class TestHunspell extends LuceneTestCase {
|
|||
Hunspell h = loadNoTimeout("compress");
|
||||
checkCompression(
|
||||
h, "toEdit=[], toAdd=[form/GS], extra=[]", "formings", "forming", "form", "forms");
|
||||
|
||||
checkCompression(h, "toEdit=[], toAdd=[f/def], extra=[]", "f", "fd", "fe", "ff");
|
||||
|
||||
WordFormGenerator gen = new WordFormGenerator(h.dictionary);
|
||||
EntrySuggestion fAbc =
|
||||
gen.compress(List.of("f", "fa", "fb", "fc"), Set.of("fyy", "fxx"), () -> {});
|
||||
assertEquals("toEdit=[], toAdd=[f/abc], extra=[]", fAbc.internalsToString());
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testCompressingIsFastOnLargeUnrelatedWordSets() throws Exception {
|
||||
Hunspell h = loadNoTimeout("compress");
|
||||
String[] letters = {"a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l"};
|
||||
checkCompression(h, "toEdit=[], toAdd=[a, b, c, d, e, f, g, h, i, j, k, l], extra=[]", letters);
|
||||
}
|
||||
|
||||
@Test
|
||||
|
|
|
|||
|
|
@ -1,5 +1,3 @@
|
|||
FORBIDDENWORD *
|
||||
|
||||
SFX G Y 1
|
||||
SFX G 0 ing/S .
|
||||
|
||||
|
|
@ -12,3 +10,47 @@ SFX S 0 s .
|
|||
SFX X Y 2
|
||||
SFX X 0 s .
|
||||
SFX X 0 x .
|
||||
|
||||
# Flags for f,fa,fb,fc
|
||||
|
||||
SFX A Y 3
|
||||
SFX A 0 a .
|
||||
SFX A 0 b .
|
||||
SFX A 0 yy .
|
||||
|
||||
SFX B Y 3
|
||||
SFX B 0 c .
|
||||
SFX B 0 b .
|
||||
SFX B 0 xx .
|
||||
|
||||
SFX a Y 1
|
||||
SFX a 0 a .
|
||||
|
||||
SFX b Y 1
|
||||
SFX b 0 b .
|
||||
|
||||
SFX c Y 1
|
||||
SFX c 0 c .
|
||||
|
||||
# Flags for f,fd,fe,ff with red herring -+* flags that bias the greedy heuristics to prefer the "fd" stem initially
|
||||
|
||||
SFX d Y 1
|
||||
SFX d 0 d .
|
||||
|
||||
SFX e Y 1
|
||||
SFX e 0 e .
|
||||
|
||||
SFX f Y 1
|
||||
SFX f 0 f .
|
||||
|
||||
SFX - Y 2
|
||||
SFX - d 0 d
|
||||
SFX - d e d
|
||||
|
||||
SFX + Y 2
|
||||
SFX + d 0 d
|
||||
SFX + d e d
|
||||
|
||||
SFX * Y 2
|
||||
SFX * d 0 d
|
||||
SFX * d e d
|
||||
|
|
|
|||
Loading…
Reference in New Issue