public void map(
LongWritable key,
Text value,
OutputCollector<IntWritable, ClusterWritable> output,
Reporter reporter)
throws IOException {
String movieIdStr = new String();
String reviewStr = new String();
String userIdStr = new String();
String reviews = new String();
String line = new String();
String tok = new String("");
long movieId;
int review, userId, p, q, r, rater, rating, movieIndex;
int clusterId = 0;
int[] n = new int[maxClusters];
float[] sq_a = new float[maxClusters];
float[] sq_b = new float[maxClusters];
float[] numer = new float[maxClusters];
float[] denom = new float[maxClusters];
float max_similarity = 0.0f;
float similarity = 0.0f;
Cluster movie = new Cluster();
ClusterWritable movies_arrl = new ClusterWritable();
StringBuffer sb = new StringBuffer();
line = ((Text) value).toString();
movieIndex = line.indexOf(":");
for (r = 0; r < maxClusters; r++) {
numer[r] = 0.0f;
denom[r] = 0.0f;
sq_a[r] = 0.0f;
sq_b[r] = 0.0f;
n[r] = 0;
}
if (movieIndex > 0) {
movieIdStr = line.substring(0, movieIndex);
sb.append(movieIdStr).append(":");
movieId = Long.parseLong(movieIdStr);
movie.movie_id = movieId;
reviews = line.substring(movieIndex + 1);
StringTokenizer token = new StringTokenizer(reviews, ",");
int attrCnt = 0;
// while (token.hasMoreTokens()) { Leo
while (token.hasMoreTokens() && attrCnt < attrNum) {
tok = token.nextToken();
int reviewIndex = tok.indexOf("_");
// userIdStr = tok.substring(0, reviewIndex); //Leo
userIdStr = String.valueOf(attrCnt);
reviewStr = tok.substring(reviewIndex + 1);
if (attrCnt > 0) {
sb.append(",");
}
sb.append(String.valueOf(attrCnt)).append("_").append(reviewStr);
userId = Integer.parseInt(userIdStr);
review = Integer.parseInt(reviewStr);
for (r = 0; r < totalClusters; r++) {
/*for (q = 0; q < centroids_ref[r].total; q++) {
rater = centroids_ref[r].reviews.get(q).rater_id;
rating = (int) centroids_ref[r].reviews.get(q).rating;
if (userId == rater) {
numer[r] += (float) (review * rating);
sq_a[r] += (float) (review * review);
sq_b[r] += (float) (rating * rating);
n[r]++; // counter
break; // to avoid multiple ratings by the same reviewer
}
}*/
// Leo
rating = (int) centroids_ref[r].reviews.get(attrCnt).rating;
numer[r] += (float) ((review - rating) * (review - rating));
n[r]++; // counter
}
attrCnt++;
}
for (p = 0; p < totalClusters; p++) {
/*denom[p] = (float) ((Math.sqrt((double) sq_a[p])) * (Math
.sqrt((double) sq_b[p])));
if (denom[p] > 0) {
similarity = numer[p] / denom[p];
if (similarity > max_similarity) {
max_similarity = similarity;
clusterId = p;
}
}*/
// Leo
similarity = 250 - numer[p];
if (similarity > max_similarity) {
max_similarity = similarity;
clusterId = p;
}
}
// movies_arrl.movies.add(line);//Leo
movies_arrl.movies.add(sb.toString());
movies_arrl.similarities.add(max_similarity);
movies_arrl.similarity = max_similarity;
output.collect(new IntWritable(clusterId), movies_arrl);
reporter.incrCounter(Counter.WORDS, 1);
}
}
public static void main(String[] args) {
// TEST MEASURE
// Point p1 = new Point(-1d, -1d);
// Point p2 = new Point(2d, 3d);
// System.out.println(measure.d(p1, p2));
// System.out.println(measure.s(p1, p2));
// return;
Double[][] data = FileHandler.readFile(fileName);
// cannot display points if dimension is > 2
if (data[0].length != 2) canDisplay = false;
// build graphic points from coords' array
buildPointsFromData(data);
Config.computeBoundingRect(points);
// init display
if (canDisplay) {
disp = new Display();
disp.setVisible(true);
for (Point p : points) {
disp.addObject(p);
}
}
testResults = new double[nbTests];
for (int t = 0; t < nbTests; ++t) {
// define K clusters and K temporary centres
clusters = new ArrayList<Cluster>();
for (int i = 0; i < K; ++i) {
clusters.add(new Cluster());
}
setRandomCenters();
for (Cluster c : clusters) {
System.out.println("center for cluster " + c + ": " + c.getCenter());
}
if (canDisplay) pause(1000);
// variables used in for loops
double minDist, currDist, diff;
Double[] prevCoords, newCoords;
Cluster alloc;
Point newCenter;
for (int i = 0; i < maxIter; ++i) {
if (canDisplay) {
disp.setLabel("[ iteration #" + (i + 1) + " ]");
} else {
System.out.println("------> iteration #" + (i + 1));
}
// allocate points to group which center is closest
for (Point p : points) {
minDist = Config.MAX;
alloc = clusters.get(0); // default initialization
for (Cluster c : clusters) {
currDist = measure.d(p, c.getCenter());
if (currDist < minDist) {
minDist = currDist;
alloc = c;
}
}
alloc.addPoint(p);
}
// recenter: calculate gravity centers for formed groups
diff = 0;
prevCoords = null;
for (Cluster c : clusters) {
// delete previous center if it not a Point of the Cluster
if (canDisplay && !c.getPoints().contains(c.getCenter())) {
disp.removeObject(c.getCenter());
}
if (stopOnConverge) {
prevCoords = c.getCenter().getCoords();
}
newCenter = c.makeGravityCenter();
if (stopOnConverge) {
newCoords = c.getCenter().getCoords();
for (int k = 0; k < prevCoords.length; ++k) {
diff += Math.abs(prevCoords[k] - newCoords[k]);
}
}
if (canDisplay) {
disp.addObject(newCenter);
} else {
// System.out.println("\tcenter for " + c + ": " + c.getCenter());
System.out.println(c.getCenter());
}
} // loop over clusters
if (canDisplay) {
disp.repaint();
}
// if Clusters' centers don't change anymore, then stop (algorithm converged)
if (diff == 0 && stopOnConverge) {
testResults[t] = (double) i;
if (canDisplay) {
disp.setLabel("[ Converged at iteration #" + (i) + " ]");
disp.repaint();
} else {
System.out.println("[ Converged at iteration #" + (i) + " ]");
}
break;
}
pause(100);
} // loop over iterations
if (testResults[t] == 0) {
System.out.println("!!!!!!!!!! Test #" + t + " did not converge.");
if (stopOnConverge) return;
}
// reset display
if (canDisplay && t + 1 < nbTests) {
for (Point p : points) p.setCluster(null);
for (Cluster c : clusters) disp.removeObject(c.getCenter());
}
} // loop over tests
// display test results and compute means
DescriptiveStatistics stats = new DescriptiveStatistics(testResults);
System.out.println("=========> Results:");
for (int t = 0; t < nbTests; ++t) {
System.out.println("--> [ " + testResults[t] + " ]");
}
System.out.println("=========> Means: " + stats.getMean());
System.out.println("=========> Std dev: " + stats.getStandardDeviation());
}
