}
// run mapper
FuzzyKMeansMapper mapper = new FuzzyKMeansMapper();
mapper.config(clusterList);
DistanceMeasure measure = new EuclideanDistanceMeasure();
Configuration conf = new Configuration();
conf.set(FuzzyKMeansConfigKeys.DISTANCE_MEASURE_KEY, measure.getClass().getName());
conf.set(FuzzyKMeansConfigKeys.CLUSTER_CONVERGENCE_KEY, "0.001");
conf.set(FuzzyKMeansConfigKeys.M_KEY, "2");
conf.set(FuzzyKMeansConfigKeys.EMIT_MOST_LIKELY_KEY, "true");
conf.set(FuzzyKMeansConfigKeys.THRESHOLD_KEY, "0");
DummyRecordWriter<Text, ClusterObservations> mapWriter = new DummyRecordWriter<Text, ClusterObservations>();
Mapper<WritableComparable<?>, VectorWritable, Text, ClusterObservations>.Context mapContext =
DummyRecordWriter.build(mapper, conf, mapWriter);
mapper.setup(mapContext);
for (VectorWritable point : points) {
mapper.map(new Text(), point, mapContext);
}
// run combiner
FuzzyKMeansCombiner combiner = new FuzzyKMeansCombiner();
DummyRecordWriter<Text, ClusterObservations> combinerWriter = new DummyRecordWriter<Text, ClusterObservations>();
Reducer<Text, ClusterObservations, Text, ClusterObservations>.Context combinerContext =
DummyRecordWriter.build(combiner, conf, combinerWriter, Text.class, ClusterObservations.class);
combiner.setup(combinerContext);
for (Text key : mapWriter.getKeys()) {
List<ClusterObservations> values = mapWriter.getValue(key);
combiner.reduce(new Text(key), values, combinerContext);
}
// run reducer
FuzzyKMeansReducer reducer = new FuzzyKMeansReducer();
DummyRecordWriter<Text, SoftCluster> reducerWriter = new DummyRecordWriter<Text, SoftCluster>();
Reducer<Text, ClusterObservations, Text, SoftCluster>.Context reducerContext =
DummyRecordWriter.build(reducer, conf, reducerWriter, Text.class, ClusterObservations.class);
reducer.setup(clusterList, conf);
for (Text key : combinerWriter.getKeys()) {
List<ClusterObservations> values = combinerWriter.getValue(key);
reducer.reduce(new Text(key), values, reducerContext);
}
// run clusterMapper
List<SoftCluster> reducerClusters = new ArrayList<SoftCluster>();
for (Text key : reducerWriter.getKeys()) {
List<SoftCluster> values = reducerWriter.getValue(key);
reducerClusters.add(values.get(0));
}
for (SoftCluster softCluster : reducerClusters) {
softCluster.computeParameters();
}
FuzzyKMeansClusterMapper clusterMapper = new FuzzyKMeansClusterMapper();
DummyRecordWriter<IntWritable, WeightedVectorWritable> clusterWriter = new DummyRecordWriter<IntWritable, WeightedVectorWritable>();
Mapper<WritableComparable<?>, VectorWritable, IntWritable, WeightedVectorWritable>.Context clusterContext =
DummyRecordWriter.build(clusterMapper, conf, clusterWriter);
clusterMapper.setup(reducerClusters, conf);
for (VectorWritable point : points) {
clusterMapper.map(new Text(), point, clusterContext);
}
// compute the reference result after one iteration and compare
List<SoftCluster> reference = new ArrayList<SoftCluster>();
for (int i = 0; i < k + 1; i++) {
Vector vec = tweakValue(points.get(i).get());
reference.add(new SoftCluster(vec, i, measure));
}
Map<Integer, List<WeightedVectorWritable>> refClusters = new HashMap<Integer, List<WeightedVectorWritable>>();
Collection<Vector> pointsVectors = new ArrayList<Vector>();
for (VectorWritable point : points) {
pointsVectors.add(point.get());
}
List<List<SoftCluster>> clusters = FuzzyKMeansClusterer.clusterPoints(pointsVectors,
reference,
new EuclideanDistanceMeasure(),
0.001,
2,
1);
computeCluster(pointsVectors, clusters.get(clusters.size() - 1), new FuzzyKMeansClusterer(new EuclideanDistanceMeasure(),
0.001,
2), refClusters);
// Now compare the clustermapper results with reference implementation
assertEquals("mapper and reference sizes", refClusters.size(), clusterWriter.getKeys().size());