/* (non-Javadoc)
* @see org.apache.hadoop.mapreduce.Reducer#reduce(KEYIN, java.lang.Iterable, org.apache.hadoop.mapreduce.Reducer.Context)
*/
protected void reduce(Tuple key, Iterable<Tuple> values, Context context)
throws IOException, InterruptedException {
sum = 0;
count = 0;
for (Tuple val : values) {
count += val.getInt(0);
sum += val.getInt(1);
}
avg = count > 0 ? sum / count : 0;
stBld.delete(0, stBld.length());
stBld.append(key.toString()).append(fieldDelim);
stBld.append(count).append(fieldDelim).append(sum).append(fieldDelim).append(avg);
outVal.set(stBld.toString());
context.write(NullWritable.get(), outVal);
}
/* (non-Javadoc)
* @see org.apache.hadoop.mapreduce.Reducer#reduce(KEYIN, java.lang.Iterable, org.apache.hadoop.mapreduce.Reducer.Context)
*/
protected void reduce(Tuple key, Iterable<Tuple> values, Context context)
throws IOException, InterruptedException {
if (stBld.length() > 0) {
stBld.delete(0, stBld.length() - 1);
}
boolean first = true;
count = 0;
latestTimeStamp = 0;
for (Tuple value : values) {
eventType = value.getInt(0);
timeStamp = value.getLong(1);
if (first) {
mostEngagingEventType = eventType;
++count;
first = false;
} else {
// all occurences of the first event type
if (eventType == mostEngagingEventType) {
++count;
}
}
// latest time stamp
if (timeStamp > latestTimeStamp) {
latestTimeStamp = timeStamp;
}
}
rating = ratingMapper.scoreForEvent(mostEngagingEventType, count);
stBld
.append(key.getString(0))
.append(fieldDelim)
.append(key.getString(1))
.append(fieldDelim)
.append(rating)
.append(fieldDelim)
.append(latestTimeStamp);
if (outputDetail) {
stBld.append(fieldDelim).append(mostEngagingEventType).append(fieldDelim).append(count);
}
valOut.set(stBld.toString());
context.write(NullWritable.get(), valOut);
}
/* (non-Javadoc)
* @see org.apache.hadoop.mapreduce.Reducer#reduce(KEYIN, java.lang.Iterable, org.apache.hadoop.mapreduce.Reducer.Context)
*/
protected void reduce(TextInt key, Iterable<Tuple> values, Context context)
throws IOException, InterruptedException {
ratingCorrelations.clear();
++logCounter;
ratingStat = null;
for (Tuple value : values) {
if (((Integer) value.get(value.getSize() - 1)) == 0) {
// in rating correlation
ratingCorrelations.add(value.createClone());
context.getCounter("Predictor", "Rating correlation").increment(1);
} else if (((Integer) value.get(value.getSize() - 1)) == 1) {
// rating stat
ratingStat = value.createClone();
} else {
// in user rating
if (!ratingCorrelations.isEmpty()) {
String userID = value.getString(0);
rating = value.getInt(1);
if (userRatingWithContext) {
ratingContext = value.getString(2);
}
// all rating correlations
for (Tuple ratingCorrTup : ratingCorrelations) {
context.getCounter("Predictor", "User rating").increment(1);
itemID = ratingCorrTup.getString(0);
ratingCorr = ratingCorrTup.getInt(1);
weight = ratingCorrTup.getInt(2);
modifyCorrelation();
int predRating =
linearCorrelation
? (rating * ratingCorr) / maxRating
: (rating * correlationScale + ratingCorr) / maxRating;
if (predRating > 0) {
// userID, itemID, predicted rating, correlation length, correlation coeff, input
// rating std dev
ratingStdDev = ratingStat != null ? ratingStat.getInt(0) : -1;
if (userRatingWithContext) {
valueOut.set(
userID
+ fieldDelim
+ itemID
+ fieldDelim
+ ratingContext
+ fieldDelim
+ predRating
+ fieldDelim
+ weight
+ fieldDelim
+ ratingCorr
+ fieldDelim
+ ratingStdDev);
} else {
valueOut.set(
userID
+ fieldDelim
+ itemID
+ fieldDelim
+ predRating
+ fieldDelim
+ weight
+ fieldDelim
+ ratingCorr
+ fieldDelim
+ ratingStdDev);
}
context.write(NullWritable.get(), valueOut);
context.getCounter("Predictor", "Rating correlation").increment(1);
}
}
}
}
}
}
/* (non-Javadoc)
* @see org.apache.hadoop.mapreduce.Reducer#reduce(KEYIN, java.lang.Iterable, org.apache.hadoop.mapreduce.Reducer.Context)
*/
protected void reduce(Tuple key, Iterable<Tuple> values, Context context)
throws IOException, InterruptedException {
if (stBld.length() > 0) {
stBld.delete(0, stBld.length());
}
testEntityId = key.getString(0);
stBld.append(testEntityId);
// collect nearest neighbors
count = 0;
neighborhood.initialize();
for (Tuple value : values) {
int index = 0;
trainEntityId = value.getString(index++);
distance = value.getInt(index++);
trainClassValue = value.getString(index++);
if (classCondtionWeighted && neighborhood.IsInClassificationMode()) {
trainingFeaturePostProb = value.getDouble(index++);
if (inverseDistanceWeighted) {
neighborhood.addNeighbor(
trainEntityId, distance, trainClassValue, trainingFeaturePostProb, true);
} else {
neighborhood.addNeighbor(
trainEntityId, distance, trainClassValue, trainingFeaturePostProb);
}
} else {
Neighborhood.Neighbor neighbor =
neighborhood.addNeighbor(trainEntityId, distance, trainClassValue);
if (neighborhood.isInLinearRegressionMode()) {
neighbor.setRegrInputVar(Double.parseDouble(value.getString(index++)));
}
}
if (++count == topMatchCount) {
break;
}
}
if (neighborhood.isInLinearRegressionMode()) {
String testRegrNumFld = isValidationMode ? key.getString(2) : key.getString(1);
neighborhood.withRegrInputVar(Double.parseDouble(testRegrNumFld));
}
// class distribution
neighborhood.processClassDitribution();
if (outputClassDistr && neighborhood.IsInClassificationMode()) {
if (classCondtionWeighted) {
Map<String, Double> classDistr = neighborhood.getWeightedClassDitribution();
double thisScore;
for (String classVal : classDistr.keySet()) {
thisScore = classDistr.get(classVal);
// LOG.debug("classVal:" + classVal + " thisScore:" + thisScore);
stBld.append(fieldDelim).append(classVal).append(fieldDelim).append(thisScore);
}
} else {
Map<String, Integer> classDistr = neighborhood.getClassDitribution();
int thisScore;
for (String classVal : classDistr.keySet()) {
thisScore = classDistr.get(classVal);
stBld.append(classVal).append(fieldDelim).append(thisScore);
}
}
}
if (isValidationMode) {
// actual class attr value
testClassValActual = key.getString(1);
stBld.append(fieldDelim).append(testClassValActual);
}
// predicted class value
if (useCostBasedClassifier) {
// use cost based arbitrator
if (neighborhood.IsInClassificationMode()) {
posClassProbab = neighborhood.getClassProb(posClassAttrValue);
testClassValPredicted = costBasedArbitrator.classify(posClassProbab);
}
} else {
// get directly
if (neighborhood.IsInClassificationMode()) {
testClassValPredicted = neighborhood.classify();
} else {
testClassValPredicted = "" + neighborhood.getPredictedValue();
}
}
stBld.append(fieldDelim).append(testClassValPredicted);
if (isValidationMode) {
if (neighborhood.IsInClassificationMode()) {
confMatrix.report(testClassValPredicted, testClassValActual);
}
}
outVal.set(stBld.toString());
context.write(NullWritable.get(), outVal);
}