static double[] similarity(LatentDirichletAllocation lda0, LatentDirichletAllocation lda1) {
int numTopics = lda0.numTopics();
int numPairs = numTopics * (numTopics - 1);
@SuppressWarnings({"unchecked", "rawtypes"}) // ok given use w. erasure
ScoredObject<int[]>[] pairs = (ScoredObject<int[]>[]) new ScoredObject[numPairs];
int pos = 0;
for (int i = 0; i < numTopics; ++i) {
for (int j = 0; j < numTopics; ++j) {
if (i == j) continue;
double divergence =
Statistics.symmetrizedKlDivergence(
lda0.wordProbabilities(i), lda1.wordProbabilities(j));
pairs[pos++] = new ScoredObject<int[]>(new int[] {i, j}, divergence);
}
}
Arrays.sort(pairs, ScoredObject.comparator());
boolean[] taken0 = new boolean[numTopics];
Arrays.fill(taken0, false);
boolean[] taken1 = new boolean[numTopics];
Arrays.fill(taken1, false);
double[] scores = new double[numTopics];
int scorePos = 0;
for (pos = 0; pos < numPairs && scorePos < numTopics; ++pos) {
int[] pair = pairs[pos].getObject();
if (!taken0[pair[0]] && !taken1[pair[1]]) {
taken0[pair[0]] = true;
taken1[pair[1]] = true;
scores[scorePos++] = pairs[pos].score();
}
}
return scores;
}
