Method randomShuffleInPlace

src/org/antlr/codebuff/Corpus.java:166–191

Feature vectors in X are lumped together as they are read in each document. In kNN, this tends to find features from the same document rather than from across the corpus since we grab k neighbors. For k=11, we might only see exemplars from a single corpus document. If all exemplars fit in k, thi

()

	 *  https://en.wikipedia.org/wiki/Fisher%E2%80%93Yates_shuffle
	 */
	public void randomShuffleInPlace() {
		Random r = new Random();
		r.setSeed(FEATURE_VECTOR_RANDOM_SEED);
		// for i from n−1 downto 1 do
		int n = featureVectors.size();
		for (int i=n-1; i>=1; i--) {
			// j ← random integer such that 0 ≤ j ≤ i
			int j = r.nextInt(i+1);
			// exchange a[j] and a[i]
			// Swap X
			int[] tmp = featureVectors.get(i);
			featureVectors.set(i, featureVectors.get(j));
			featureVectors.set(j, tmp);
			// And now swap all prediction lists
			Integer tmpI = injectWhitespace.get(i);
			injectWhitespace.set(i, injectWhitespace.get(j));
			injectWhitespace.set(j, tmpI);
			tmpI = hpos.get(i);
			hpos.set(i, hpos.get(j));
			hpos.set(j, tmpI);
			// Finally, swap documents
			InputDocument tmpD = documentsPerExemplar.get(i);
			documentsPerExemplar.set(i, documentsPerExemplar.get(j));
			documentsPerExemplar.set(j, tmpD);
		}
	}

	public void buildTokenContextIndex() {
		curAndPrevTokenRuleIndexToExemplarIndexes = new MultiMap<>();