package de.lmu.ifi.dbs.elki.algorithm;
/*
This file is part of ELKI:
Environment for Developing KDD-Applications Supported by Index-Structures
Copyright (C) 2012
Ludwig-Maximilians-Universität München
Lehr- und Forschungseinheit für Datenbanksysteme
ELKI Development Team
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
import java.util.ArrayList;
import java.util.Collection;
import java.util.List;
import de.lmu.ifi.dbs.elki.data.NumberVector;
import de.lmu.ifi.dbs.elki.data.type.TypeInformation;
import de.lmu.ifi.dbs.elki.data.type.TypeUtil;
import de.lmu.ifi.dbs.elki.database.Database;
import de.lmu.ifi.dbs.elki.database.datastore.DataStore;
import de.lmu.ifi.dbs.elki.database.datastore.DataStoreFactory;
import de.lmu.ifi.dbs.elki.database.datastore.DataStoreUtil;
import de.lmu.ifi.dbs.elki.database.datastore.WritableDataStore;
import de.lmu.ifi.dbs.elki.database.ids.DBID;
import de.lmu.ifi.dbs.elki.database.ids.DBIDs;
import de.lmu.ifi.dbs.elki.database.query.DoubleDistanceResultPair;
import de.lmu.ifi.dbs.elki.database.relation.Relation;
import de.lmu.ifi.dbs.elki.distance.DistanceUtil;
import de.lmu.ifi.dbs.elki.distance.distancefunction.DistanceFunction;
import de.lmu.ifi.dbs.elki.distance.distancefunction.SpatialPrimitiveDistanceFunction;
import de.lmu.ifi.dbs.elki.distance.distancefunction.SpatialPrimitiveDoubleDistanceFunction;
import de.lmu.ifi.dbs.elki.distance.distancevalue.Distance;
import de.lmu.ifi.dbs.elki.distance.distancevalue.DoubleDistance;
import de.lmu.ifi.dbs.elki.index.tree.LeafEntry;
import de.lmu.ifi.dbs.elki.index.tree.spatial.SpatialEntry;
import de.lmu.ifi.dbs.elki.index.tree.spatial.SpatialIndexTree;
import de.lmu.ifi.dbs.elki.index.tree.spatial.SpatialNode;
import de.lmu.ifi.dbs.elki.index.tree.spatial.SpatialPointLeafEntry;
import de.lmu.ifi.dbs.elki.logging.Logging;
import de.lmu.ifi.dbs.elki.logging.progress.FiniteProgress;
import de.lmu.ifi.dbs.elki.logging.progress.IndefiniteProgress;
import de.lmu.ifi.dbs.elki.result.ResultUtil;
import de.lmu.ifi.dbs.elki.utilities.datastructures.heap.Heap;
import de.lmu.ifi.dbs.elki.utilities.datastructures.heap.KNNHeap;
import de.lmu.ifi.dbs.elki.utilities.datastructures.heap.KNNList;
import de.lmu.ifi.dbs.elki.utilities.documentation.Description;
import de.lmu.ifi.dbs.elki.utilities.documentation.Title;
import de.lmu.ifi.dbs.elki.utilities.exceptions.AbortException;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.OptionID;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.constraints.GreaterConstraint;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.parameterization.Parameterization;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.parameters.IntParameter;
/**
* Joins in a given spatial database to each object its k-nearest neighbors.
* This algorithm only supports spatial databases based on a spatial index
* structure.
*
* Since this method compares the MBR of every single leaf with every other
* leaf, it is essentially quadratic in the number of leaves, which may not be
* appropriate for large trees.
*
* @author Elke Achtert
* @author Erich Schubert
*
* @param <V> the type of FeatureVector handled by this Algorithm
* @param <D> the type of Distance used by this Algorithm
* @param <N> the type of node used in the spatial index structure
* @param <E> the type of entry used in the spatial node
*/
@Title("K-Nearest Neighbor Join")
@Description("Algorithm to find the k-nearest neighbors of each object in a spatial database")
public class KNNJoin<V extends NumberVector<V, ?>, D extends Distance<D>, N extends SpatialNode<N, E>, E extends SpatialEntry> extends AbstractDistanceBasedAlgorithm<V, D, DataStore<KNNList<D>>> {
/**
* The logger for this class.
*/
private static final Logging logger = Logging.getLogger(KNNJoin.class);
/**
* Parameter that specifies the k-nearest neighbors to be assigned, must be an
* integer greater than 0. Default value: 1.
*/
public static final OptionID K_ID = OptionID.getOrCreateOptionID("knnjoin.k", "Specifies the k-nearest neighbors to be assigned.");
/**
* The k parameter
*/
int k;
/**
* Constructor.
*
* @param distanceFunction Distance function
* @param k k parameter
*/
public KNNJoin(DistanceFunction<? super V, D> distanceFunction, int k) {
super(distanceFunction);
this.k = k;
}
/**
* Joins in the given spatial database to each object its k-nearest neighbors.
*
* @throws IllegalStateException if not suitable {@link SpatialIndexTree} was
* found or the specified distance function is not an instance of
* {@link SpatialPrimitiveDistanceFunction}.
*/
@SuppressWarnings("unchecked")
public WritableDataStore<KNNList<D>> run(Database database, Relation<V> relation) throws IllegalStateException {
if(!(getDistanceFunction() instanceof SpatialPrimitiveDistanceFunction)) {
throw new IllegalStateException("Distance Function must be an instance of " + SpatialPrimitiveDistanceFunction.class.getName());
}
Collection<SpatialIndexTree<N, E>> indexes = ResultUtil.filterResults(database, SpatialIndexTree.class);
if(indexes.size() != 1) {
throw new AbortException("KNNJoin found " + indexes.size() + " spatial indexes, expected exactly one.");
}
// FIXME: Ensure were looking at the right relation!
SpatialIndexTree<N, E> index = indexes.iterator().next();
SpatialPrimitiveDistanceFunction<V, D> distFunction = (SpatialPrimitiveDistanceFunction<V, D>) getDistanceFunction();
DBIDs ids = relation.getDBIDs();
// Optimize for double?
final boolean doubleOptimize = (getDistanceFunction() instanceof SpatialPrimitiveDoubleDistanceFunction);
// data pages
List<E> ps_candidates = new ArrayList<E>(index.getLeaves());
// knn heaps
List<List<KNNHeap<D>>> heaps = new ArrayList<List<KNNHeap<D>>>(ps_candidates.size());
Heap<Task> pq = new Heap<Task>(ps_candidates.size() * ps_candidates.size() / 10);
// Initialize with the page self-pairing
for(int i = 0; i < ps_candidates.size(); i++) {
E pr_entry = ps_candidates.get(i);
N pr = index.getNode(pr_entry);
heaps.add(initHeaps(distFunction, doubleOptimize, pr));
}
// Build priority queue
final int sqsize = ps_candidates.size() * (ps_candidates.size() - 1) / 2;
if(logger.isDebuggingFine()) {
logger.debugFine("Number of leaves: " + ps_candidates.size() + " so " + sqsize + " MBR computations.");
}
FiniteProgress mprogress = logger.isVerbose() ? new FiniteProgress("Comparing leaf MBRs", sqsize, logger) : null;
for(int i = 0; i < ps_candidates.size(); i++) {
E pr_entry = ps_candidates.get(i);
List<KNNHeap<D>> pr_heaps = heaps.get(i);
D pr_knn_distance = computeStopDistance(pr_heaps);
for(int j = i + 1; j < ps_candidates.size(); j++) {
E ps_entry = ps_candidates.get(j);
List<KNNHeap<D>> ps_heaps = heaps.get(j);
D ps_knn_distance = computeStopDistance(ps_heaps);
D minDist = distFunction.minDist(pr_entry, ps_entry);
// Resolve immediately:
if(minDist.isNullDistance()) {
N pr = index.getNode(ps_candidates.get(i));
N ps = index.getNode(ps_candidates.get(j));
processDataPagesOptimize(distFunction, doubleOptimize, pr_heaps, ps_heaps, pr, ps);
}
else if(minDist.compareTo(pr_knn_distance) <= 0 || minDist.compareTo(ps_knn_distance) <= 0) {
pq.add(new Task(minDist, i, j));
}
if(mprogress != null) {
mprogress.incrementProcessed(logger);
}
}
}
if(mprogress != null) {
mprogress.ensureCompleted(logger);
}
// Process the queue
FiniteProgress qprogress = logger.isVerbose() ? new FiniteProgress("Processing queue", pq.size(), logger) : null;
IndefiniteProgress fprogress = logger.isVerbose() ? new IndefiniteProgress("Full comparisons", logger) : null;
while(!pq.isEmpty()) {
Task task = pq.poll();
List<KNNHeap<D>> pr_heaps = heaps.get(task.i);
List<KNNHeap<D>> ps_heaps = heaps.get(task.j);
D pr_knn_distance = computeStopDistance(pr_heaps);
D ps_knn_distance = computeStopDistance(ps_heaps);
boolean dor = task.mindist.compareTo(pr_knn_distance) <= 0;
boolean dos = task.mindist.compareTo(ps_knn_distance) <= 0;
if(dor || dos) {
N pr = index.getNode(ps_candidates.get(task.i));
N ps = index.getNode(ps_candidates.get(task.j));
if(dor && dos) {
processDataPagesOptimize(distFunction, doubleOptimize, pr_heaps, ps_heaps, pr, ps);
}
else {
if(dor) {
processDataPagesOptimize(distFunction, doubleOptimize, pr_heaps, null, pr, ps);
}
else /* dos */{
processDataPagesOptimize(distFunction, doubleOptimize, ps_heaps, null, ps, pr);
}
}
if(fprogress != null) {
fprogress.incrementProcessed(logger);
}
}
if(qprogress != null) {
qprogress.incrementProcessed(logger);
}
}
if(qprogress != null) {
qprogress.ensureCompleted(logger);
}
if(fprogress != null) {
fprogress.setCompleted(logger);
}
WritableDataStore<KNNList<D>> knnLists = DataStoreUtil.makeStorage(ids, DataStoreFactory.HINT_STATIC, KNNList.class);
// FiniteProgress progress = logger.isVerbose() ? new
// FiniteProgress(this.getClass().getName(), relation.size(), logger) :
// null;
FiniteProgress pageprog = logger.isVerbose() ? new FiniteProgress("Number of processed data pages", ps_candidates.size(), logger) : null;
// int processed = 0;
for(int i = 0; i < ps_candidates.size(); i++) {
N pr = index.getNode(ps_candidates.get(i));
List<KNNHeap<D>> pr_heaps = heaps.get(i);
// Finalize lists
for(int j = 0; j < pr.getNumEntries(); j++) {
knnLists.put(((LeafEntry) pr.getEntry(j)).getDBID(), pr_heaps.get(j).toKNNList());
}
// Forget heaps and pq
heaps.set(i, null);
// processed += pr.getNumEntries();
// if(progress != null) {
// progress.setProcessed(processed, logger);
// }
if(pageprog != null) {
pageprog.incrementProcessed(logger);
}
}
// if(progress != null) {
// progress.ensureCompleted(logger);
// }
if(pageprog != null) {
pageprog.ensureCompleted(logger);
}
return knnLists;
}
private List<KNNHeap<D>> initHeaps(SpatialPrimitiveDistanceFunction<V, D> distFunction, final boolean doubleOptimize, N pr) {
List<KNNHeap<D>> pr_heaps;
// Create for each data object a knn heap
pr_heaps = new ArrayList<KNNHeap<D>>(pr.getNumEntries());
for(int j = 0; j < pr.getNumEntries(); j++) {
pr_heaps.add(new KNNHeap<D>(k, distFunction.getDistanceFactory().infiniteDistance()));
}
// Self-join first, as this is expected to improve most and cannot be
// pruned.
processDataPagesOptimize(distFunction, doubleOptimize, pr_heaps, null, pr, pr);
return pr_heaps;
}
/**
* Processes the two data pages pr and ps and determines the k-nearest
* neighbors of pr in ps.
*
* @param distFunction the distance to use
* @param doubleOptimize Flag whether to optimize for doubles.
* @param pr the first data page
* @param ps the second data page
* @param pr_heaps the knn lists for each data object in pr
* @param ps_heaps the knn lists for each data object in ps (if ps != pr)
*/
private void processDataPagesOptimize(SpatialPrimitiveDistanceFunction<V, D> distFunction, final boolean doubleOptimize, List<KNNHeap<D>> pr_heaps, List<KNNHeap<D>> ps_heaps, N pr, N ps) {
if(doubleOptimize) {
List<?> khp = (List<?>) pr_heaps;
List<?> khs = (List<?>) ps_heaps;
processDataPagesDouble((SpatialPrimitiveDoubleDistanceFunction<? super V>) distFunction, pr, ps, (List<KNNHeap<DoubleDistance>>) khp, (List<KNNHeap<DoubleDistance>>) khs);
}
else {
for(int j = 0; j < ps.getNumEntries(); j++) {
final SpatialPointLeafEntry s_e = (SpatialPointLeafEntry) ps.getEntry(j);
DBID s_id = s_e.getDBID();
for(int i = 0; i < pr.getNumEntries(); i++) {
final SpatialPointLeafEntry r_e = (SpatialPointLeafEntry) pr.getEntry(i);
D distance = distFunction.minDist(s_e, r_e);
pr_heaps.get(i).add(distance, s_id);
if(pr != ps && ps_heaps != null) {
ps_heaps.get(j).add(distance, r_e.getDBID());
}
}
}
}
}
/**
* Processes the two data pages pr and ps and determines the k-nearest
* neighbors of pr in ps.
*
* @param df the distance function to use
* @param pr the first data page
* @param ps the second data page
* @param pr_heaps the knn lists for each data object
* @param ps_heaps the knn lists for each data object in ps
*/
private void processDataPagesDouble(SpatialPrimitiveDoubleDistanceFunction<? super V> df, N pr, N ps, List<KNNHeap<DoubleDistance>> pr_heaps, List<KNNHeap<DoubleDistance>> ps_heaps) {
// Compare pairwise
for(int j = 0; j < ps.getNumEntries(); j++) {
final SpatialPointLeafEntry s_e = (SpatialPointLeafEntry) ps.getEntry(j);
DBID s_id = s_e.getDBID();
for(int i = 0; i < pr.getNumEntries(); i++) {
final SpatialPointLeafEntry r_e = (SpatialPointLeafEntry) pr.getEntry(i);
double distance = df.doubleMinDist(s_e, r_e);
pr_heaps.get(i).add(new DoubleDistanceResultPair(distance, s_id));
if(pr != ps && ps_heaps != null) {
ps_heaps.get(j).add(new DoubleDistanceResultPair(distance, r_e.getDBID()));
}
}
}
}
/**
* Compute the maximum stop distance
*
* @param heaps
* @return the k-nearest neighbor distance of pr in ps
*/
private D computeStopDistance(List<KNNHeap<D>> heaps) {
// Update pruning distance
D pr_knn_distance = null;
for(KNNHeap<D> knnList : heaps) {
// set kNN distance of r
if(pr_knn_distance == null) {
pr_knn_distance = knnList.getKNNDistance();
}
else {
pr_knn_distance = DistanceUtil.max(knnList.getKNNDistance(), pr_knn_distance);
}
}
return pr_knn_distance;
}
@Override
public TypeInformation[] getInputTypeRestriction() {
return TypeUtil.array(TypeUtil.NUMBER_VECTOR_FIELD);
}
@Override
protected Logging getLogger() {
return logger;
}
/**
* Task in the processing queue
*
* @author Erich Schubert
*
* @apiviz.exclude
*/
private class Task implements Comparable<Task> {
final D mindist;
final int i;
final int j;
/**
* Constructor.
*
* @param mindist
* @param i
* @param j
*/
public Task(D mindist, int i, int j) {
super();
this.mindist = mindist;
this.i = i;
this.j = j;
}
@Override
public int compareTo(Task o) {
return mindist.compareTo(o.mindist);
}
}
/**
* Parameterization class.
*
* @author Erich Schubert
*
* @apiviz.exclude
*/
public static class Parameterizer<V extends NumberVector<V, ?>, D extends Distance<D>, N extends SpatialNode<N, E>, E extends SpatialEntry> extends AbstractPrimitiveDistanceBasedAlgorithm.Parameterizer<V, D> {
protected int k;
@Override
protected void makeOptions(Parameterization config) {
super.makeOptions(config);
IntParameter kP = new IntParameter(K_ID, 1);
kP.addConstraint(new GreaterConstraint(0));
if(config.grab(kP)) {
k = kP.getValue();
}
}
@Override
protected KNNJoin<V, D, N, E> makeInstance() {
return new KNNJoin<V, D, N, E>(distanceFunction, k);
}
}
}