| 1 | /* |
| 2 | * Copyright 2010 Savarese Software Research Corporation |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * https://www.savarese.com/software/ApacheLicense-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | package com.savarese.spatial; |
| 18 | |
| 19 | import java.util.PriorityQueue; |
| 20 | import java.util.Map; |
| 21 | import java.util.Arrays; |
| 22 | import java.util.Comparator; |
| 23 | |
| 24 | /** |
| 25 | * NearestNeighbors implements an algorithm for finding the k-nearest |
| 26 | * neighbors to a query point within the set of points contained by a |
| 27 | * {@link KDTree} instance. The algorithm can be specialized with a custom |
| 28 | * distance-finding function by passing a {@link Distance} instance to its |
| 29 | * constructor. |
| 30 | */ |
| 31 | public class NearestNeighbors<Coord extends Number & Comparable<? super Coord>, |
| 32 | P extends Point<Coord>, V> |
| 33 | { |
| 34 | /** |
| 35 | * The Entry interface makes accessible the results of a |
| 36 | * {@link NearestNeighbors} search. An Entry exposes both the |
| 37 | * point-value mapping and its distance from the query point. |
| 38 | */ |
| 39 | public interface Entry<Coord extends Number & Comparable<? super Coord>, |
| 40 | P extends Point<Coord>, V> |
| 41 | { |
| 42 | /** |
| 43 | * Returns the distance from result to the query point. This |
| 44 | * will usually be implemented by dynamically taking the square root |
| 45 | * of {@link #getDistance2}. Therefore, repeated calls may be |
| 46 | * expensive. |
| 47 | * |
| 48 | * @return The distance from result to the query point. |
| 49 | */ |
| 50 | public double getDistance(); |
| 51 | |
| 52 | /** |
| 53 | * Returns the square of the distance from result to the query point. |
| 54 | * This will usually be implemented as returning a cached value used |
| 55 | * during the nearest neighbors search. |
| 56 | * |
| 57 | * @return The square of the distance from result to the query point. |
| 58 | */ |
| 59 | public double getDistance2(); |
| 60 | |
| 61 | /** |
| 62 | * Returns the point-value mapping stored in this query result. |
| 63 | * |
| 64 | * @return The point-value mapping stored in this query result. |
| 65 | */ |
| 66 | public Map.Entry<P,V> getNeighbor(); |
| 67 | } |
| 68 | |
| 69 | private final class NNEntry |
| 70 | implements Entry<Coord, P, V>, Comparable<Entry<Coord, P, V>> |
| 71 | { |
| 72 | double _distance2; |
| 73 | Map.Entry<P,V> _neighbor; |
| 74 | |
| 75 | NNEntry(double distance2, Map.Entry<P,V> neighbor) { |
| 76 | _distance2 = distance2; |
| 77 | _neighbor = neighbor; |
| 78 | } |
| 79 | |
| 80 | public double getDistance() { |
| 81 | return StrictMath.sqrt(_distance2); |
| 82 | } |
| 83 | |
| 84 | public double getDistance2() { |
| 85 | return _distance2; |
| 86 | } |
| 87 | |
| 88 | public Map.Entry<P,V> getNeighbor() { |
| 89 | return _neighbor; |
| 90 | } |
| 91 | |
| 92 | public int compareTo(Entry<Coord, P, V> obj) { |
| 93 | final double d = obj.getDistance2(); |
| 94 | |
| 95 | if(_distance2 < d) |
| 96 | return -1; |
| 97 | else if(_distance2 > d) |
| 98 | return 1; |
| 99 | |
| 100 | return 0; |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | private final class EntryComparator |
| 105 | implements Comparator<Entry<Coord, P, V>> |
| 106 | { |
| 107 | // Invert relationship so priority queue keeps highest on top. |
| 108 | public int compare(Entry<Coord, P, V> n1, Entry<Coord, P, V> n2) { |
| 109 | final double d1 = n1.getDistance2(); |
| 110 | final double d2 = n2.getDistance2(); |
| 111 | |
| 112 | if(d1 < d2) |
| 113 | return 1; |
| 114 | else if(d1 > d2) |
| 115 | return -1; |
| 116 | |
| 117 | return 0; |
| 118 | } |
| 119 | |
| 120 | public boolean equals(Object obj) { |
| 121 | return (obj != null && obj == this); |
| 122 | } |
| 123 | } |
| 124 | |
| 125 | private boolean __omitQueryPoint; |
| 126 | private int __numNeighbors; |
| 127 | private double __minDistance; |
| 128 | private Distance<Coord, P> __distance; |
| 129 | private PriorityQueue<Entry<Coord, P, V>> __pq; |
| 130 | private P __query; |
| 131 | |
| 132 | private void find(KDTree<Coord,P,V>.KDNode node) { |
| 133 | if(node == null) |
| 134 | return; |
| 135 | |
| 136 | final int discriminator = node._discriminator; |
| 137 | final P point = node.getKey(); |
| 138 | double d2 = __distance.distance2(__query, point); |
| 139 | |
| 140 | if(d2 < __minDistance && (d2 != 0.0 || !__omitQueryPoint)) { |
| 141 | if(__pq.size() == __numNeighbors) { |
| 142 | __pq.poll(); |
| 143 | __pq.add(new NNEntry(d2, node)); |
| 144 | __minDistance = __pq.peek().getDistance2(); |
| 145 | } else { |
| 146 | __pq.add(new NNEntry(d2, node)); |
| 147 | if(__pq.size() == __numNeighbors) { |
| 148 | __minDistance = __pq.peek().getDistance2(); |
| 149 | } |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | double dp = |
| 154 | __query.getCoord(discriminator).doubleValue() - |
| 155 | point.getCoord(discriminator).doubleValue(); |
| 156 | |
| 157 | d2 = dp*dp; |
| 158 | |
| 159 | if(dp < 0) { |
| 160 | find(node._low); |
| 161 | if(d2 < __minDistance) { |
| 162 | find(node._high); |
| 163 | } |
| 164 | } else { |
| 165 | find(node._high); |
| 166 | if(d2 < __minDistance) { |
| 167 | find(node._low); |
| 168 | } |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | /** |
| 173 | * Constructs a new NearestNeighbors instance, using the specified |
| 174 | * distance-finding functor to calculate distances during searches. |
| 175 | * |
| 176 | * @param distance A distance-finding functor implementing |
| 177 | * the {@link Distance} interface. |
| 178 | */ |
| 179 | public NearestNeighbors(Distance<Coord, P> distance) { |
| 180 | __distance = distance; |
| 181 | } |
| 182 | |
| 183 | /** |
| 184 | * Constructs a NearestNeighbors instance using a {@link EuclideanDistance} |
| 185 | * instance to calculate distances between points. |
| 186 | */ |
| 187 | public NearestNeighbors() { |
| 188 | this(new EuclideanDistance<Coord, P>()); |
| 189 | } |
| 190 | |
| 191 | /** |
| 192 | * Sets the distance-finding functor used to calculate distances during |
| 193 | * searches. |
| 194 | * |
| 195 | * @param distance The distance-finding functor to use for distance |
| 196 | * calculations. |
| 197 | */ |
| 198 | public void setDistance(Distance<Coord, P> distance) { |
| 199 | __distance = distance; |
| 200 | } |
| 201 | |
| 202 | /** |
| 203 | * Finds the k-nearest neighbors to a query point withina KDTree instance. |
| 204 | * The neighbors are returned as an array of {@link Entry} instances, sorted |
| 205 | * from nearest to farthest. |
| 206 | * |
| 207 | * @param tree The KDTree to search. |
| 208 | * @param queryPoint The query point. |
| 209 | * @param numNeighbors The number of nearest neighbors to find. This should |
| 210 | * be a positive value. Non-positive values result in no neighbors |
| 211 | * being found. |
| 212 | * @param omitQueryPoint If true, point-value mappings at a distance of |
| 213 | * zero are omitted from the result. If false, mappings at a |
| 214 | * distance of zero are included. |
| 215 | * @return An array containing the nearest neighbors and their distances |
| 216 | * sorted by least distance to greatest distance. If no neighbors |
| 217 | * are found, the array will have a length of zero. |
| 218 | */ |
| 219 | public Entry<Coord,P,V>[] get(KDTree<Coord,P,V> tree, |
| 220 | P queryPoint, |
| 221 | int numNeighbors, |
| 222 | boolean omitQueryPoint) |
| 223 | { |
| 224 | __omitQueryPoint = omitQueryPoint; |
| 225 | __numNeighbors = numNeighbors; |
| 226 | __query = queryPoint; |
| 227 | __minDistance = Double.POSITIVE_INFINITY; |
| 228 | |
| 229 | __pq = new PriorityQueue<Entry<Coord, P, V>>(numNeighbors, |
| 230 | new EntryComparator()); |
| 231 | |
| 232 | if(numNeighbors > 0) { |
| 233 | find(tree._root); |
| 234 | } |
| 235 | |
| 236 | Entry<Coord,P,V>[] neighbors = new Entry[__pq.size()]; |
| 237 | |
| 238 | __pq.toArray(neighbors); |
| 239 | Arrays.sort(neighbors); |
| 240 | |
| 241 | __pq = null; |
| 242 | __query = null; |
| 243 | |
| 244 | return neighbors; |
| 245 | } |
| 246 | |
| 247 | /** |
| 248 | * Same as {@link #get get(tree, queryPoint, numNeighbors, true)}. |
| 249 | */ |
| 250 | public Entry<Coord,P,V>[] |
| 251 | get(KDTree<Coord,P,V> tree, P queryPoint, int numNeighbors) |
| 252 | { |
| 253 | return get(tree, queryPoint, numNeighbors, true); |
| 254 | } |
| 255 | } |
