transient sclass DeepBitSetWordIndex { S regexp = "\\w+"; new Map singleTextIndices; new ElementInstanceMatrix mainIndex; void add(A a, S text) { singleTextIndices.put(a, new SingleTextWordIndex(regexp, text)); mainIndex.add(a, mapToSet upper(regexpExtractAll(regexp, text)); } void doneAdding { mainIndex.doneAdding(); } LPair wordsAndOffsets(S text) { ret map(regexpFindRanges(regexp, text), r -> pair(upper(substring(text, r)), r.start)); } // assumes word boundaries left and right of query Cl preSearch(S query, O... _) { optPar bool debug; LPair l = wordsAndOffsets(query); Cl candidates = mainIndex.instancesContainingAllElements(pairsA(l)); if (debug) { L lengths = map(candidates, a -> singleTextIndices.get(a).length); print(nCandidates(candidates) + ", total length: " + n2(intSum(lengths)) + ", lengths: " + lengths); } ret filter(candidates, a -> nempty(singleTextIndices.get(a).indicesOfWordCombination(l))); } int numWords() { ret mainIndex.numElements(); } }