Flip the order of matching in ProcessSemgrexRequest so that for each pattern, it matches all of the sentences at once. Allows for operations on the complete batch of matches, such as the new uniq operator

Author: AngledLuffa

src/edu/stanford/nlp/semgraph/semgrex/ProcessSemgrexRequest.java

@@ -32,10 +32,9 @@ public class ProcessSemgrexRequest extends ProcessProtobufRequest {
   /**
    * Builds a single inner SemgrexResult structure from the pair of a SemgrexPattern and a SemanticGraph
    */
-  public static CoreNLPProtos.SemgrexResponse.SemgrexResult matchSentence(SemgrexPattern pattern, SemanticGraph graph, int patternIdx, int graphIdx) {
+  public static CoreNLPProtos.SemgrexResponse.SemgrexResult matchSentence(SemgrexPattern pattern, SemanticGraph graph, List<SemgrexMatch> matches, int patternIdx, int graphIdx) {
     CoreNLPProtos.SemgrexResponse.SemgrexResult.Builder semgrexResultBuilder = CoreNLPProtos.SemgrexResponse.SemgrexResult.newBuilder();
-    SemgrexMatcher matcher = pattern.matcher(graph);
-    while (matcher.find()) {
+    for (SemgrexMatch matcher : matches) {
       CoreNLPProtos.SemgrexResponse.Match.Builder matchBuilder = CoreNLPProtos.SemgrexResponse.Match.newBuilder();
       matchBuilder.setMatchIndex(matcher.getMatch().index());
       matchBuilder.setSemgrexIndex(patternIdx);
@@ -105,14 +104,27 @@ public static CoreNLPProtos.SemgrexResponse processRequest(CoreNLPProtos.Semgrex
     }
 
     List<SemgrexPattern> patterns = request.getSemgrexList().stream().map(SemgrexPattern::compile).collect(Collectors.toList());
-    int graphIdx = 0;
+    List<Pair<CoreMap, List<Pair<SemgrexPattern, List<SemgrexMatch>>>>> allMatches = new ArrayList<>();
     for (CoreMap sentence : sentences) {
+      allMatches.add(new Pair<>(sentence, new ArrayList<>()));
+    }
+    for (SemgrexPattern pattern : patterns) {
+      List<Pair<CoreMap, List<SemgrexMatch>>> patternMatches = pattern.matchSentences(sentences, true);
+      for (int i = 0; i < sentences.size(); ++i) {
+        Pair<CoreMap, List<SemgrexMatch>> sentenceMatches = patternMatches.get(i);
+        allMatches.get(i).second().add(new Pair<>(pattern, sentenceMatches.second()));
+      }
+    }
+
+    int graphIdx = 0;
+    for (Pair<CoreMap, List<Pair<SemgrexPattern, List<SemgrexMatch>>>> sentenceMatches : allMatches) {
       CoreNLPProtos.SemgrexResponse.GraphResult.Builder graphResultBuilder = CoreNLPProtos.SemgrexResponse.GraphResult.newBuilder();
 
       int patternIdx = 0;
-      SemanticGraph graph = sentence.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class);
-      for (SemgrexPattern pattern : patterns) {
-        graphResultBuilder.addResult(matchSentence(pattern, graph, patternIdx, graphIdx));
+      SemanticGraph graph = sentenceMatches.first().get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class);
+      for (Pair<SemgrexPattern, List<SemgrexMatch>> patternMatches : sentenceMatches.second()) {
+        SemgrexPattern pattern = patternMatches.first();
+        graphResultBuilder.addResult(matchSentence(pattern, graph, patternMatches.second(), patternIdx, graphIdx));
         ++patternIdx;
       }
 

src/edu/stanford/nlp/semgraph/semgrex/SemgrexPattern.java

@@ -351,13 +351,16 @@ public List<Pair<CoreMap, List<SemgrexMatch>>> postprocessMatches(List<Pair<Core
    *<br>
    * Non-matching sentences are currently not returned (may change in the future to return an empty list).
    */
-  public List<Pair<CoreMap, List<SemgrexMatch>>> matchSentences(List<CoreMap> sentences) {
+  public List<Pair<CoreMap, List<SemgrexMatch>>> matchSentences(List<CoreMap> sentences, boolean keepEmptyMatches) {
     List<Pair<CoreMap, List<SemgrexMatch>>> matches = new ArrayList<>();
     for (CoreMap sentence : sentences) {
       SemanticGraph graph = sentence.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class);
       SemanticGraph enhanced = sentence.get(SemanticGraphCoreAnnotations.EnhancedDependenciesAnnotation.class);
       SemgrexMatcher matcher = matcher(graph);
-      if ( ! matcher.find()) {
+      if (!matcher.find()) {
+        if (keepEmptyMatches) {
+          matches.add(new Pair<>(sentence, new ArrayList<>()));
+        }
         continue;
       }
       matches.add(new Pair<>(sentence, new ArrayList<>()));
@@ -602,7 +605,7 @@ public static void main(String[] args) throws IOException {
       }
     }
 
-    List<Pair<CoreMap, List<SemgrexMatch>>> matches = semgrex.matchSentences(sentences);
+    List<Pair<CoreMap, List<SemgrexMatch>>> matches = semgrex.matchSentences(sentences, false);
 
     for (Pair<CoreMap, List<SemgrexMatch>> sentenceMatches : matches) {
       CoreMap sentence = sentenceMatches.first();

test/src/edu/stanford/nlp/semgraph/semgrex/SemgrexTest.java

@@ -1492,7 +1492,7 @@ public List<CoreMap> buildSmallBatch() {
   public void testBatchSearch() {
     List<CoreMap> sentences = buildSmallBatch();
     SemgrexPattern semgrex = SemgrexPattern.compile("{word:foo}=x > {}=y");
-    List<Pair<CoreMap, List<SemgrexMatch>>> matches = semgrex.matchSentences(sentences);
+    List<Pair<CoreMap, List<SemgrexMatch>>> matches = semgrex.matchSentences(sentences, false);
     String[] expectedMatches = {
       BATCH_PARSES[0],
       BATCH_PARSES[1],
@@ -1535,21 +1535,21 @@ public void testParsesUniq() {
   public void testBatchUniq() {
     List<CoreMap> sentences = buildSmallBatch();
     SemgrexPattern semgrex = SemgrexPattern.compile("{word:foo}=x > {}=y :: uniq x");
-    List<Pair<CoreMap, List<SemgrexMatch>>> matches = semgrex.matchSentences(sentences);
+    List<Pair<CoreMap, List<SemgrexMatch>>> matches = semgrex.matchSentences(sentences, false);
     // only the first foo sentence should match when using "uniq x"
     assertEquals(1, matches.size());
     assertEquals(BATCH_PARSES[0], matches.get(0).first().get(CoreAnnotations.TextAnnotation.class));
     assertEquals(1, matches.get(0).second().size());
 
     semgrex = SemgrexPattern.compile("{word:foo}=x > {}=y :: uniq");
-    matches = semgrex.matchSentences(sentences);
+    matches = semgrex.matchSentences(sentences, false);
     // same thing happens when using "uniq" and no nodes - only one match will occur
     assertEquals(1, matches.size());
     assertEquals(BATCH_PARSES[0], matches.get(0).first().get(CoreAnnotations.TextAnnotation.class));
     assertEquals(1, matches.get(0).second().size());
 
     semgrex = SemgrexPattern.compile("{word:foo}=x > {}=y :: uniq y");
-    matches = semgrex.matchSentences(sentences);
+    matches = semgrex.matchSentences(sentences, false);
     // now it should match both foo>bar and foo>baz
     assertEquals(2, matches.size());
     assertEquals(BATCH_PARSES[0], matches.get(0).first().get(CoreAnnotations.TextAnnotation.class));
@@ -1558,7 +1558,7 @@ public void testBatchUniq() {
     assertEquals(1, matches.get(1).second().size());
 
     semgrex = SemgrexPattern.compile("{}=x > {}=y :: uniq x y");
-    matches = semgrex.matchSentences(sentences);
+    matches = semgrex.matchSentences(sentences, false);
     // now it should batch each of foo>bar, bar>baz, foo>baz
     assertEquals(3, matches.size());
     assertEquals(BATCH_PARSES[0], matches.get(0).first().get(CoreAnnotations.TextAnnotation.class));
@@ -1570,7 +1570,7 @@ public void testBatchUniq() {
   }
 
   public static void outputBatchResults(SemgrexPattern pattern, List<CoreMap> sentences) {
-    List<Pair<CoreMap, List<SemgrexMatch>>> matches = pattern.matchSentences(sentences);
+    List<Pair<CoreMap, List<SemgrexMatch>>> matches = pattern.matchSentences(sentences, false);
     for (Pair<CoreMap, List<SemgrexMatch>> sentenceMatch : matches) {
       System.out.println("Pattern matched at:");
       System.out.println(sentenceMatch.first());