@Override
public DocScoreList mostSimilar(String phrase, int maxResults, TIntSet validIds)
throws IOException {
final TIntDoubleHashMap scores = getConceptVector(phrase, validIds);
Integer luceneIds[] = ArrayUtils.toObject(scores.keys());
Arrays.sort(
luceneIds,
new Comparator<Integer>() {
@Override
public int compare(Integer id1, Integer id2) {
return -1 * new Double(scores.get(id1)).compareTo(scores.get(id2));
}
});
DocScoreList result = new DocScoreList(Math.min(luceneIds.length, maxResults));
for (int i = 0; i < result.numDocs(); i++) {
result.set(i, esaHelper.luceneIdToWpId(luceneIds[i]), scores.get(luceneIds[i]));
}
return normalize(result);
}
public void seek(TermEnum terms) throws IOException {
original.seek(terms);
docFreq = terms.docFreq();
pointer = -1;
if (docFreq > postingMaps.length) { // grow postingsMap
PostingMap[] newMap = new PostingMap[docFreq];
System.arraycopy(postingMaps, 0, newMap, 0, postingMaps.length);
for (int i = postingMaps.length; i < docFreq; i++) {
newMap[i] = new PostingMap();
}
postingMaps = newMap;
}
out.reset();
int i = 0;
while (original.next()) {
PostingMap map = postingMaps[i++];
map.newDoc = oldToNew[original.doc()]; // remap the newDoc id
map.offset = out.getFilePointer(); // save pointer to buffer
final int tf = original.freq(); // buffer tf & positions
out.writeVInt(tf);
int prevPosition = 0;
for (int j = tf; j > 0; j--) { // delta encode positions
int p = original.nextPosition();
out.writeVInt(p - prevPosition);
prevPosition = p;
}
}
out.flush();
docFreq = i; // allow for deletions
Arrays.sort(postingMaps, 0, docFreq); // resort by mapped doc ids
// HeapSorter.sort(postingMaps,docFreq); // TODO MC - due to the lack of space
// NOTE: this might be substantially faster if RAMInputStream were public
// and supported a reset() operation.
in = tempDir.openInput(TEMP_FILE);
}
// private static int[] oldToNew(IndexReader reader, Searcher searcher) throws IOException {
private static DocScore[] newToOld(IndexReader reader, Searcher searcher) throws IOException {
int readerMax = reader.maxDoc();
DocScore[] newToOld = new DocScore[readerMax];
// use site, an indexed, un-tokenized field to get boost
// byte[] boosts = reader.norms("site"); TODO MC
/* TODO MC */
Document docMeta;
Pattern includes = Pattern.compile("\\|");
String value = NutchConfiguration.create().get(INCLUDE_EXTENSIONS_KEY, "");
String includeExtensions[] = includes.split(value);
Hashtable<String, Boolean> validExtensions = new Hashtable<String, Boolean>();
for (int i = 0; i < includeExtensions.length; i++) {
validExtensions.put(includeExtensions[i], true);
System.out.println("extension boosted " + includeExtensions[i]);
}
/* TODO MC */
for (int oldDoc = 0; oldDoc < readerMax; oldDoc++) {
float score;
if (reader.isDeleted(oldDoc)) {
// score = 0.0f;
score = -1f; // TODO MC
} else {
// score = Similarity.decodeNorm(boosts[oldDoc]); TODO MC
/* TODO MC */
docMeta = searcher.doc(oldDoc);
if (validExtensions.get(docMeta.get("subType"))
== null) { // searched extensions will have higher scores
score = -0.5f;
} else {
score = Integer.parseInt(docMeta.get("inlinks"));
/*
if (score==0) {
score=0.001f; // TODO MC - to not erase
}
*/
}
/* TODO MC */
// System.out.println("Score for old document "+oldDoc+" is "+score+" and type
// "+docMeta.get("subType")); // TODO MC debug remove
}
DocScore docScore = new DocScore();
docScore.doc = oldDoc;
docScore.score = score;
newToOld[oldDoc] = docScore;
}
System.out.println("Sorting " + newToOld.length + " documents.");
Arrays.sort(newToOld);
// HeapSorter.sort(newToOld); // TODO MC - due to the lack of space
/* TODO MC
int[] oldToNew = new int[readerMax];
for (int newDoc = 0; newDoc < readerMax; newDoc++) {
DocScore docScore = newToOld[newDoc];
//oldToNew[docScore.oldDoc] = docScore.score > 0.0f ? newDoc : -1; // TODO MC
oldToNew[docScore.oldDoc] = newDoc; // TODO MC
}
*/
/* TODO MC *
for (int newDoc = 0; newDoc < readerMax; newDoc++) {
DocScore docScore = newToOld[newDoc];
System.out.println("Score for new document "+newDoc+" is "+docScore.score); // TODO MC debug remove
}
* TODO MC */
// return oldToNew; TODO MC
return newToOld; // TODO MC
}
/**
* @param scrollSort Whether to ignore the from and sort all hits in each shard result. Only used
* for scroll search
* @param resultsArr Shard result holder
*/
public ScoreDoc[] sortDocs(
boolean scrollSort, AtomicArray<? extends QuerySearchResultProvider> resultsArr)
throws IOException {
List<? extends AtomicArray.Entry<? extends QuerySearchResultProvider>> results =
resultsArr.asList();
if (results.isEmpty()) {
return EMPTY_DOCS;
}
if (optimizeSingleShard) {
boolean canOptimize = false;
QuerySearchResult result = null;
int shardIndex = -1;
if (results.size() == 1) {
canOptimize = true;
result = results.get(0).value.queryResult();
shardIndex = results.get(0).index;
} else {
// lets see if we only got hits from a single shard, if so, we can optimize...
for (AtomicArray.Entry<? extends QuerySearchResultProvider> entry : results) {
if (entry.value.queryResult().topDocs().scoreDocs.length > 0) {
if (result != null) { // we already have one, can't really optimize
canOptimize = false;
break;
}
canOptimize = true;
result = entry.value.queryResult();
shardIndex = entry.index;
}
}
}
if (canOptimize) {
int offset = result.from();
if (scrollSort) {
offset = 0;
}
ScoreDoc[] scoreDocs = result.topDocs().scoreDocs;
if (scoreDocs.length == 0 || scoreDocs.length < offset) {
return EMPTY_DOCS;
}
int resultDocsSize = result.size();
if ((scoreDocs.length - offset) < resultDocsSize) {
resultDocsSize = scoreDocs.length - offset;
}
ScoreDoc[] docs = new ScoreDoc[resultDocsSize];
for (int i = 0; i < resultDocsSize; i++) {
ScoreDoc scoreDoc = scoreDocs[offset + i];
scoreDoc.shardIndex = shardIndex;
docs[i] = scoreDoc;
}
return docs;
}
}
@SuppressWarnings("unchecked")
AtomicArray.Entry<? extends QuerySearchResultProvider>[] sortedResults =
results.toArray(new AtomicArray.Entry[results.size()]);
Arrays.sort(sortedResults, QUERY_RESULT_ORDERING);
QuerySearchResultProvider firstResult = sortedResults[0].value;
final Sort sort;
if (firstResult.queryResult().topDocs() instanceof TopFieldDocs) {
TopFieldDocs firstTopDocs = (TopFieldDocs) firstResult.queryResult().topDocs();
sort = new Sort(firstTopDocs.fields);
} else {
sort = null;
}
int topN = firstResult.queryResult().size();
// Need to use the length of the resultsArr array, since the slots will be based on the position
// in the resultsArr array
TopDocs[] shardTopDocs = new TopDocs[resultsArr.length()];
if (firstResult.includeFetch()) {
// if we did both query and fetch on the same go, we have fetched all the docs from each
// shards already, use them...
// this is also important since we shortcut and fetch only docs from "from" and up to "size"
topN *= sortedResults.length;
}
for (AtomicArray.Entry<? extends QuerySearchResultProvider> sortedResult : sortedResults) {
TopDocs topDocs = sortedResult.value.queryResult().topDocs();
// the 'index' field is the position in the resultsArr atomic array
shardTopDocs[sortedResult.index] = topDocs;
}
int from = firstResult.queryResult().from();
if (scrollSort) {
from = 0;
}
// TopDocs#merge can't deal with null shard TopDocs
for (int i = 0; i < shardTopDocs.length; i++) {
if (shardTopDocs[i] == null) {
shardTopDocs[i] = Lucene.EMPTY_TOP_DOCS;
}
}
TopDocs mergedTopDocs = TopDocs.merge(sort, from, topN, shardTopDocs);
return mergedTopDocs.scoreDocs;
}