1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
|
package de.lmu.ifi.dbs.elki.index.tree.spatial.rstarvariants.strategies.bulk;
/*
This file is part of ELKI:
Environment for Developing KDD-Applications Supported by Index-Structures
Copyright (C) 2015
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.Arrays;
import java.util.List;
import de.lmu.ifi.dbs.elki.data.spatial.SpatialComparable;
import de.lmu.ifi.dbs.elki.data.spatial.SpatialSingleMeanComparator;
import de.lmu.ifi.dbs.elki.utilities.datastructures.QuickSelect;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.AbstractParameterizer;
/**
* This is variation of the original STR bulk load for non-rectangular data
* spaces. Instead of iterating through the dimensions and splitting each by
* (approximately) the same factor, this variation tries to adjust the factor to
* the extends of the data space. I.e. if the data set is twice as wide as high,
* this should produce twice as many partitions on the X than on the Y axis.
*
* Whether or not this offers benefits greatly depends on the distance queries
* used. But for symmetric distances, the resulting pages should be more
* rectangular, which often is beneficial.
*
* See {@link SortTileRecursiveBulkSplit} for the original STR bulk load.
*
* @author Erich Schubert
* @since 0.5.0
*/
public class AdaptiveSortTileRecursiveBulkSplit extends AbstractBulkSplit {
/**
* Static instance.
*/
public static final AdaptiveSortTileRecursiveBulkSplit STATIC = new AdaptiveSortTileRecursiveBulkSplit();
@Override
public <T extends SpatialComparable> List<List<T>> partition(List<T> spatialObjects, int minEntries, int maxEntries) {
final int dims = spatialObjects.get(0).getDimensionality();
final int p = (int) Math.ceil(spatialObjects.size() / (double) maxEntries);
List<List<T>> ret = new ArrayList<>(p);
strPartition(spatialObjects, 0, spatialObjects.size(), 0, dims, maxEntries, new SpatialSingleMeanComparator(0), ret);
return ret;
}
/**
* Recursively partition.
*
* @param objs Object list
* @param start Subinterval start
* @param end Subinterval end
* @param depth Iteration depth (must be less than dimensionality!)
* @param dims Total number of dimensions
* @param maxEntries Maximum page size
* @param c Comparison helper
* @param ret Output list
* @param <T> data type
*/
protected <T extends SpatialComparable> void strPartition(List<T> objs, int start, int end, int depth, int dims, int maxEntries, SpatialSingleMeanComparator c, List<List<T>> ret) {
final int p = (int) Math.ceil((end - start) / (double) maxEntries);
// Compute min and max:
double[] mm = new double[dims * 2];
for (int d = 0; d < mm.length; d += 2) {
mm[d] = Double.POSITIVE_INFINITY; // min <- +inf
mm[d + 1] = Double.NEGATIVE_INFINITY; // max <- -inf
}
for (int i = start; i < end; i++) {
T o = objs.get(i);
for (int d1 = 0, d2 = 0; d2 < mm.length; d1++, d2 += 2) {
mm[d2] = Math.min(mm[d2], o.getMin(d1));
mm[d2 + 1] = Math.max(mm[d2 + 1], o.getMax(d1));
}
}
// Find maximum and compute extends
double maxex = 0.0;
int sdim = depth;
double[] exts = new double[dims];
for (int d = 0; d < mm.length; d += 2) {
final double extend = mm[d + 1] - mm[d];
if (extend > maxex) {
maxex = extend;
sdim = d >>> 1;
}
exts[d >>> 1] = extend;
}
// Compute sum of the k largest extends:
Arrays.sort(exts);
double extsum = 0.;
for (int d = depth; d < exts.length; d++) {
extsum += exts[d];
}
// Chose the number of partitions:
final int s;
if (maxex > 0. && depth + 1 < dims) {
s = (int) Math.ceil(Math.pow(p, 1.0 / (dims - depth)) * (dims - depth) * maxex / extsum);
} else {
s = (int) Math.ceil(Math.pow(p, 1.0 / (dims - depth)));
}
final double len = end - start; // double intentional!
for (int i = 0; i < s; i++) {
// We don't completely sort, but only ensure the quantile is invariant.
int s2 = start + (int) ((i * len) / s);
int e2 = start + (int) (((i + 1) * len) / s);
// LoggingUtil.warning("STR " + dim + " s2:" + s2 + " e2:" + e2);
if (e2 < end) {
c.setDimension(sdim);
QuickSelect.quickSelect(objs, c, s2, end, e2);
}
if (depth + 1 == dims) {
ret.add(objs.subList(s2, e2));
} else {
// Descend
strPartition(objs, s2, e2, depth + 1, dims, maxEntries, c, ret);
}
}
}
/**
* Parameterization class.
*
* @author Erich Schubert
*
* @apiviz.exclude
*/
public static class Parameterizer extends AbstractParameterizer {
@Override
protected AdaptiveSortTileRecursiveBulkSplit makeInstance() {
return STATIC;
}
}
}
|
