move failsafe that checks for no votes < threshold into the classifier itself and update the analysis string as well.

Author: parrt

java/src/org/antlr/codebuff/Formatter.java

@@ -1,6 +1,5 @@
 package org.antlr.codebuff;
 
-import org.antlr.codebuff.misc.HashBag;
 import org.antlr.v4.runtime.CommonToken;
 import org.antlr.v4.runtime.CommonTokenStream;
 import org.antlr.v4.runtime.ParserRuleContext;
@@ -17,14 +16,12 @@
 import static org.antlr.codebuff.CollectFeatures.CAT_ALIGN_WITH_ANCESTOR_CHILD;
 import static org.antlr.codebuff.CollectFeatures.CAT_INDENT;
 import static org.antlr.codebuff.CollectFeatures.CAT_INDENT_FROM_ANCESTOR_FIRST_TOKEN;
-import static org.antlr.codebuff.CollectFeatures.CAT_NO_ALIGNMENT;
 import static org.antlr.codebuff.CollectFeatures.FEATURES_ALIGN;
 import static org.antlr.codebuff.CollectFeatures.FEATURES_INJECT_NL;
 import static org.antlr.codebuff.CollectFeatures.FEATURES_INJECT_WS;
 import static org.antlr.codebuff.CollectFeatures.INDEX_FIRST_ON_LINE;
 import static org.antlr.codebuff.CollectFeatures.INDEX_PREV_END_COLUMN;
 import static org.antlr.codebuff.CollectFeatures.MAX_CONTEXT_DIFF_THRESHOLD;
-import static org.antlr.codebuff.CollectFeatures.MAX_CONTEXT_DIFF_THRESHOLD2;
 import static org.antlr.codebuff.CollectFeatures.earliestAncestorStartingWithToken;
 import static org.antlr.codebuff.CollectFeatures.getNodeFeatures;
 import static org.antlr.codebuff.CollectFeatures.getRealTokens;
@@ -126,13 +123,6 @@ public void processToken(int indexIntoRealTokens, int tokenIndexInStream) {
 		features[INDEX_FIRST_ON_LINE] = injectNewline; // use \n prediction to match exemplars for alignment
 
 		int align = alignClassifier.classify(k, features, corpus.align, MAX_CONTEXT_DIFF_THRESHOLD);
-		if ( align==CAT_NO_ALIGNMENT ) {
-			HashBag<Integer> votes = alignClassifier.votes(k, features, corpus.align, MAX_CONTEXT_DIFF_THRESHOLD);
-			if ( votes.size()==0 ) {
-				// try with less strict match threshold to get some indication of alignment
-				align = alignClassifier.classify(k, features, corpus.align, MAX_CONTEXT_DIFF_THRESHOLD2);
-			}
-		}
 
 		int ws = wsClassifier.classify(k, features, corpus.injectWS, MAX_CONTEXT_DIFF_THRESHOLD);
 

java/src/org/antlr/codebuff/kNNClassifier.java

@@ -7,6 +7,8 @@
 import java.util.Arrays;
 import java.util.List;
 
+import static org.antlr.codebuff.CollectFeatures.MAX_CONTEXT_DIFF_THRESHOLD2;
+
 /** A kNN (k-Nearest Neighbor) classifier */
 public abstract class kNNClassifier {
 	protected final Corpus corpus;
@@ -50,6 +52,10 @@ public int[] classify(int k, int[] unknown, double distanceThreshold) {
 	 */
 	public int classify(int k, int[] unknown, List<Integer> Y, double distanceThreshold) {
 		HashBag<Integer> votes = votes(k, unknown, Y, distanceThreshold);
+		if ( votes.size()==0 ) {
+			// try with less strict match threshold to get some indication of alignment
+			votes = votes(k, unknown, Y, MAX_CONTEXT_DIFF_THRESHOLD2);
+		}
 		return getCategoryWithMostVotes(votes);
 	}
 
@@ -94,6 +100,11 @@ public HashBag<Integer> getVotesBag(Neighbor[] kNN, int k, int[] unknown, List<I
 	public String getPredictionAnalysis(InputDocument doc, int k, int[] unknown, List<Integer> Y, double distanceThreshold) {
 		Neighbor[] kNN = kNN(unknown, k, distanceThreshold);
 		HashBag<Integer> votes = getVotesBag(kNN, k, unknown, Y);
+		if ( votes.size()==0 ) {
+			// try with less strict match threshold to get some indication of alignment
+			kNN = kNN(unknown, k, MAX_CONTEXT_DIFF_THRESHOLD2);
+			votes = getVotesBag(kNN, k, unknown, Y);
+		}
 
 		StringBuilder buf = new StringBuilder();
 		buf.append(CollectFeatures.featureNameHeader(FEATURES));