<html>
<head>
<title>
NearTree -- function library efficiently solving the Nearest Neighbor Problem
</title>
</head>
<body>
<font face="Arial,Helvetica,Times" size="3">
<table border=2>
<tr>
<td valign="center" width="88" align="center" height="35" bgcolor="cyan">
<font face="Arial,Helvetica,Times" size="2">
<a href="http://sf.net/projects/cvector" style="text-decoration:none">CVector</a>
</font>
</td>
<td valign="center" width="88" align="center" height="35">
<a href="http://sourceforge.net/projects/neartree"><img src="http://sflogo.sourceforge.net/sflogo.php?group_id=237121&type=10" width="80" height="15" alt="Get NearTree at SourceForge.net. Fast, secure and Free Open Source software downloads" /></a>
</td>
</tr>
</table>
<hr>
<center>
<h2 align=center>NearTree</h2>
Release 2.3.2<br />
<font size="-1">
30 October 2010<br />
© Copyright 2001, 2008, 2009, 2010 Larry Andrews. All rights reserved
<br />
based on<br />
Larry Andrews, "<i>A template for the nearest neighbor problem</i>",<br />
C/C++ Users Journal, Volume 19 , Issue 11 (November 2001), 40 - 49 (2001),
ISSN:1075-2838,<br />
<a href="http://www.ddj.com/architect/184401449" style="text-decoration:none">www.ddj.com/architect/184401449</a>
<p>
Revised 12 Dec 2008, for sourceforge release, Larry Andrews and Herbert J. Bernstein<br />
8 Jan 2009 Release 1.0 LCA and HJB<br />
11 Jan 2009 Release 1.0.1 LCA and HJB<br />
21 March 2009 Release 2.0 LCA and HJB<br />
30 May 2009 Release 2.1 LCA and HJB<br />
4 June 2009 Release 2.1.1 LCA and HJB<br />
7 June 2009 Release 2.1.2 LCA and HJB<br />
7 July 2009 Release 2.1.3 LCA and HJB<br />
29 November 2009 Release 2.1.4 LCA<br />
23 April 2010 Release 2.1.5 LCA and HJB<br />
18 July 2010 Release 2.2 HJB<br />
25 July 2010 Release 2.2.1 HJB<br />
31 August 2010 Release 2.3 LCA<br />
7 September 2010 Release 2.3.1 LCA<br />
30 October 2010 Release 2.3.2 LCA<br />
</font>
<P>
<h4>YOU MAY REDISTRIBUTE NearTree UNDER THE TERMS OF THE <a href=lgpl.txt style="text-decoration:none">LGPL</a></h4>
<p>
</center>
<center>
<table border=1>
<tr>
<td><font face="Arial,Helvetica,Times" size="2">
<h4 align=center>LGPL NOTICES</h4>
<p>
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
<p>
This library 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*
Lesser General Public License for more details.
<p>
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free
Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston,
MA 02110-1301 USA
</td>
</table>
</center>
<P>
This is a release of an API for finding nearest neighbors among points in spaces of
arbitrary dimensions. This release provides a C++ template, TNear.h, and a
C library, CNearTree.c, with example/test programs.
<font size="-1">
<p>Release 2.3.2 adds optional returns of vectors of ordinals of found objects
<p>Release 2.3.1 adds Centroid method for Lloyd clustering.
<p>Release 2.3 added methods for clustering.
<p>Release 2.2.1 was a minor revision to Release 2.2 to add an include of limits.h to TNear.h,
primarily for MINGW use.
<p>Release 2.2 added support for C code for fixed length string searches using a hamming distance norm,
and for spherical and hemispherical geodesic norm based searches. Because of the
addition of new type and norm flags, the version 2.2 shared libraries cannot be used
to support binaries copiled against earlier headers and vice-versa.
<p>Release 2.1.5 was a cleanup update to the 2.1 release of 30 May 2009 to increase
portability, in five stages (2.1.1 on 4 June 2009, 2.1.2 on 7 June 2009, 2.1.3 on 7 July 2009,
2.1.4 on 29 November 2009 and 2.1.5 on 23 April 2010)
dealing with the following issues:
<ul>
<li>Convert to use of a self-contained portable random-number generator from Rob Harrison (2.1.1)
<li>Ensure wider use of const where appropriate (2.1.1)
<li>Correct typos and unclear wording in the README (2.1.2)
<li>Reorganize use of USE_LOCAL_HEADERS in CNearTreeTest.cpp (2.1.2)
<li>Change FAR macro to CNEARTREE_FAR (2.1.3)
<li>Add BelongsToPoints and SeparateByRadius (2.1.4)
<li>Fix dimensions for rhrand (2.1.5)
</ul>
<p>The 2.1 release was a minor update to the 2.0 release of 21 March 2009 to deal with the
following issues:
<ul>
<li>Make delayed insertion the default
<li>Complete the containerization of TNear.h
<li>Add code for K-nearest/farthest in TNear.h and in CNearTree.c
<li>Correct the InAnnulus search filter
</ul>
<P>Release 2.0 was a major update to the 1.0 release of 8 January 2009 to deal with the
following issues:
<ul>
<li>Replace use recursion with a stack, except in insertion logic
<li>Replace use of double with templated DistanceType (usually double)
<li>Provide constuctors to build NearTree from vectors, lists or sets
<li>Change "Insert" to "insert" for consistentcy with
other containers
<li>Add access function "at" or array type references [], and provide
contents of a neartree as a vector
<li>Add iterator support
<li>Provide delayed insertion logic
<li>Functions added for searches outside of a sphere or in an annular region
</ul>
</font>
<P>
Our thanks to Nicolas Brodu for suggesting the more general handling of the
distance type.
<p>
<b>Note:</b> As Nicolas Brodu has noted, CNearTree is particularly well-suited
to multi-threaded applications. However, if the same CNearTree is to be searched
in multiple threads, it is important to complete all insertions and/or delayed
insertions before parallel execution of parallel searches.
<hr>
<h4>Contents</h4>
<P>
<ul>
<li><a href="#Installation">Installation</a>
<li><a href="#Tnear.h">The C++ template: TNear.h</a>
<li><a href="#CNearTree.c">The C NearTree API: CNearTree.c</a>
<li><a href="#rhrand.h">A Portable pseudo-random number generator: rhrand.h</a>
</ul>
<hr>
<hr>
<h4><a name="Installation">Installation</a></h4>
<p>
The NearTree package is available at <a href="http://www.sourceforge.net/projects/neartree">www.sourceforge.net/projects/neartree</a>.
A source tarball is available at <a href="http://downloads.sourceforge.net/neartree/NearTree-2.3.2.tar.gz">downloads.sourceforge.net/neartree/NearTree-2.3.2.tar.gz</a>.
Later tarballs may be available.
<p>
If you decide to simply use the TNear.h header to add nearest neighbor support to C++ code under Visual
Studio, be sure to also use the rhrand.h and triple.h headers, and to define USE_LOCAL_HEADERS. For
unix or MINGW, you will need to use the Makefile and to have libtool on your system. Be warned that
the default libtool under Mac OS X will not work for this installation.
<P>
When the source tarball is downloaded and unpacked, you should have a directory NearTree-2.3.2. To see the current settings for a build
execute
<p>
make
</p>
which should give the following information:
<p>
<pre><tt>
PLEASE READ README_NearTree.txt and lgpl.txt
Before making the NearTree libraries and example programs, check
that the chosen settings are correct
The current C++ and C compile commands are:
libtool --mode=compile g++ -g -O2 -Wall -ansi -pedantic \
-DCNEARTREE_SAFE_TRIANG=1 -I. -c
libtool --mode=compile gcc -g -O2 -Wall -ansi -pedantic \
-DCNEARTREE_SAFE_TRIANG=1 -I. -c
The C API, CNearTree.c, depends on the sourceforge project CVector
You are currently setup to use the system defaults for CVector
If that is not correct, define the variable CVECTOR_INCLUDE
The current library link command is:
libtool --mode=link gcc -version-info 3:0:0 \
-no-undefined -rpath /usr/local/lib
The current C++ and C library local, and C dynamic and static build commands are:
libtool --mode=link g++ -no-undefined -g -O2 -Wall -ansi -pedantic \
-DCNEARTREE_SAFE_TRIANG=1 -I.
libtool --mode=link gcc -g -O2 -Wall -ansi -pedantic \
-DCNEARTREE_SAFE_TRIANG=1 -I.
libtool --mode=link gcc -no-undefined -g -O2 -Wall -ansi -pedantic \
-DCNEARTREE_SAFE_TRIANG=1 -shared -I/usr/local/include
libtool --mode=link gcc -g -O2 -Wall -ansi -pedantic \
-DCNEARTREE_SAFE_TRIANG=1 -static-libtool-libs -I/usr/local/include
Before installing the NearTree library and example programs, check
that the install directory and install commands are correct:
The current values are :
/usr/local
libtool --mode=install cp
To compile the NearTree library and example programs type:
make clean
make all
To run a set of tests type:
make tests
To clean up the directories type:
make clean
To install the library and headers type:
make install
</tt>
</pre>
<p>
If these settings need to be changed, edit Makefile. On some systems, e.g. Mac OS X, the default
libtool is not appropriate. In that case you should install a recent version of libtool. The
CVector kit has been tested with libtool versions 1.3.5 and 1.5.4. For MINGW, libtool version 2.2.6
and gcc version 4 are needed to work with shared libraries (DLLs). If the system libtool is
not to be used, define the variable LIBTOOL to be the path to the libtool executable, e.g.
in bash
<p>
export LIBTOOL=$HOME/bin/libtool
<p>
or in the Makefie
<p>
LIBTOOL = $(HOME)/bin/libtool
<p>
If you need to include local header files using #include "..." instead of #include <...>,
define the variable USE_LOCAL_HEADERS
<p>
The triangle inequality that must be evaluated in building trees and
retrieving data may not be evaluated correctly if the range of the three values
is extremely large (>10**15 or so for doubles) or may be evaluated differently
by some compilers in different parts of a program (due to differing usage of registers).
The default in this API is to do the triangle inequality three different ways
under the control of CNEARTREE_SAFE_TRIANG
<p>
<pre><tt>
#ifdef CNEARTREE_SAFE_TRIANG
#define TRIANG(a,b,c) ( (((b)+(c))-(a) >= 0) \
|| ((b)-((a)-(c)) >= 0) \
|| ((c)-((a)-(b)) >= 0))
#else
#define TRIANG(a,b,c) ( (((b)+(c))-(a) >= 0))
#endif
</tt></pre>
<p>
Problems with the unsafe definition of TRIANG have been seen in Linux under gcc version 4
and in MS Window under VS 2003. There is a slight performance hit from the triple test.
If maximal speed is critical and misidentification of nearest points by relative distance
errors of about 1 part in 10**15 is not a serious problem, the definition of
-DCNEARTREE_SAFE_TRIANG=1 can be removed from the definition of CFLAGS in the Makefile.
<p>
NOTE: A range of 10**15 is comparable to the diameter of the earth
vs. the separation of two bonded atoms.
<hr>
<hr>
<h4><a name="Tnear.h">The C++ template: TNear.h</a></h4>
<P>
This is a revised release of
<blockquote>
template <typename T, typename DistanceType=double, int distMinValue=-1 > class CNearTree;
</blockquote>
<P>
implementing the Nearest Neighbor algorithm after Kalantari and McDonald,
(IEEE Transactions on Software Engineering, v. SE-9, pp. 631-634,1983)
modified to use recursion for insertions and recursion (original version)
or a stack (current version) for searches
instead of a double-linked tree
and simplified so that it does a bit less checking for
things like is the distance to the right less than the
distance to the left; it was found that these checks made little
to no difference.
<p>
This template is used to contain a collection of objects. After the
collection has been loaded into this structure, it can be quickly
queried for which object is "closest" to some probe object of the
same type. The major restriction on applicability of the near-tree
is that the algorithm only works if the objects obey the triangle
inequality. The triangle rule states that the length of any side of
a triangle cannot exceed the sum of the lengths of the other two sides.
<p>
CNearTree is the root class for the neartree. The actual data of the
tree is stored in NearTreeNode objects descending from a CNearTree.
<p>
The types of objects that can be stored in the tree is quite broad. The
biggest limitation is that the objects must reside in some sort of metric
space and must obey the triangle rule. They must also be all of the same
size because they are stored in an std::vector. If your application
requires object of varying storage, then your best way to use this
code is to store pointers or handles and to write your own distance functions.
Note that std::string is a pointer type variable and so can be stored directly.
<p>
The type of the objects to be stored is the only <b>required</b> template argument.
The type of the distance measure (DistanceType) defaults to double. If your
application is for an integer type, then the type for DistanceType can be your
integer type. This has the potential for speeding the calculations by
avoiding FP computation. Other general types can be used if desired, but you
may need to also input a value of distMinValue.
<p>
The template argument distMinValue must be something that your class will
understand as a negative number. The default input is negative one. Internally,
that is cast to DistanceType. Since most uses will be for DistanceType
to be double, that is a simple conversion. Obviously, for integer types,
there is no problem either. The need for this value is to have something
internally that is recognizable as smaller than the smallest "distance"
that can exist between any two objects in your type. For most users,
there is no need to input anything other than the default, -1. -1 must
be castable to DistanceType. It seems unlikely that anyone would actually
need this optional parameter, but it is here for completeness.
<p>
It is a design decision that this class cannot work for unsigned types.
Verifying the triangle rule for unsigned types is more complex. Sorry,
unsigned types are left as an exercise for the reader.
<p>
The user of this class needs to provide at least the following
functionality for the template to work. For the built-in
numerics of C++, they are provided by the system.
<p>
<center>
<table>
<tr>
<td><font face="Arial,Helvetica,Times" size="3">
DistanceType Norm( ); </font></td><td><font face="Arial,Helvetica,Times" size="3">// conversion constructor from the templated class to DistanceType<br />
// (usually will return a "length" of type double)
</font></td>
</tr>
<tr>
<td><font face="Arial,Helvetica,Times" size="3">
operator- ( ); </font></td><td><font face="Arial,Helvetica,Times" size="3">// geometrical (vector) difference of two objects
</font></td>
</tr>
<tr>
<td></td><td><font face="Arial,Helvetica,Times" size="3">// a copy constructor would be nice
</font></td>
</tr>
<tr>
<td></td><td><font face="Arial,Helvetica,Times" size="3">// a constructor would be nice
</font></td>
</tr>
<tr>
<td></td><td><font face="Arial,Helvetica,Times" size="3">// a destructor would be nice
</font></td>
</tr>
</table>
</center>
<p>
The provided interface is:
<p>
<pre><tt>
<b>#include <TNear.h></b>
<b>CNearTree</b>( <b>void</b> ) // constructor
instantiated by something like: CNearTree <T> vTree;
for some type T
<b>CNearTree</b>( <b>const ContainerType<T></b> ) // constructor from containers, e.g. ...
<b>CNearTree</b>( <b>const std::vector<T></b> ) // constructor
<b>CNearTree</b>( <b>const std::list<T></b> ) // constructor
<b>CNearTree</b>( <b>const std::set<T></b> ) // constructor
<b>CNearTree</b>( <b>const CNearTree<T></b> ) // constructor
<b>CNearTree</b>( <b>const ContainerType<T></b>, <b>const ContainerType<T></b> )
// constructor merging 2 containers, e.g. ...
<b>~CNearTree</b>( <b>void</b> ) // destructor
<b>void clear</b>( <b>void</b> ) // clear the NearTree
<b>template<typename</b> <i>InputContainer</i>>
<b>CNearTree& operator+=</b>( <b>const InputContainer&</b>, <i>o</i> )
// add a container's contents to a NearTree
<b>template<typename</b> <i>InputContainer</i>>
<b>CNearTree& operator-=</b>( <b>const InputContainer&</b>, <i>o</i> )
// remove a container's contents from a NearTree
<b>template<typename</b> <i>InputContainer</i>>
<b>CNearTree& set_symmetric_difference</b>( <b>const InputContainer&</b>, <i>o</i> )
// remove the part of a container's
// contents from a NearTree that is
// already in the Neartree and add
// in the contents of the container
// that is not already in the Neartree
// i.e. the exclusive or
<b>void insert</b>( <b>const T&</b> <i>t</i> )
where t is an object of the type T
all inserts are delayed until a search is performed or
until an explicit call to CompleteDelayedInsertions
is called or a search is called. The purpose is to distribute
the objects a bit more randomly. Excessively ordered objects
leads to less than optimal trees.
Starting with the 2.1 release, places objects in a queue for
insertion later when CompleteDelayInsert is called. In
earlier releases the default was immediate insertion.
The following additional convenience insert template
allow insertion of containers of objects
<b>template< typename</b> <i>InputContainer</i> >
<b>void insert</b>( <b>ContainerType</b> ) // e. g. ...
<b>void insert</b>( <b>const std::vector<T></b> )
<b>void insert</b>( <b>const std::list<T></b> )
<b>void insert</b>( <b>const std::set<T></b> )
<b>void insert</b>( <b>const CNearTree<T></b> )
<b>bool NearestNeighbor</b> ( <b>const DistanceType&</b> <i>dRadius</i>, <b>T&</b> <i>tClosest</i>, <b>const T&</b> <i>t</i> ) <b>const</b>
<i>dRadius</i> is the largest radius within which to search; make it
very large if you want to include every point that was loaded.
<i>tClosest</i> is returned as the object that was found closest to the probe
point (if any were within radius dRadius of the probe)
<i>t</i> is the probe point, used to search in the group of points insert'ed
return value is true if some object was found within the search radius, false otherwise.
If false is returned, tClosest is invalid (at best).
<b>iterator NearestNeighbor</b> ( <b>const DistanceType &</b> <i>dRadius</i>, <b>const T&</b> <i>t</i> ) <b>const</b>
returns an iterator to the nearest point to the probe point <i>t</i> or end() if there is none
<b>bool FarthestNeighbor</b> ( <b>T&</b> <i>tFarthest</i>, <b>const T&</b> <i>t</i> ) <b>const</b>
<i>tFarthest</i> is returned as the object that was found farthest from the probe
point
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is true if some object was found, false otherwise
If false is returned, tFarthest is invalid (at best).
<b>iterator FarthestNeighbor</b> ( <b>T&</b> <b>const T&</b> <i>t</i> ) <b>const</b>
returns an iterator to the nearest point to the probe point <i>t</i> or end() if there is none
<b>bool FarthestNeighbor</b> ( <b>T&</b> <i>tFarthest</i>, <b>const T&</b> <i>t</i> ) <b>const</b>
The following functions (BelongsToPoints, SeparateByRadius, FindInSphere, FindOutSphere,
and FindInAnnulus) all return a container (ContainerType) that can be any standard library
container (such as std::vector< T >) or CNearTree. In each case option arguments
allow a parallel vector of indices to be returned for each container, giving the
indices of the returned objects within the original NearTree.
<b>template<typename</b> <i>ContainerType</i>>
<b>void BelongsToPoints</b> ( <b>const T&</b> <i>t1</i>, <b>const T&</b> <i>t2</i>,
<b>ContainerType&</b> <i>group1</i>, <b>ContainerType&</b> <i>group2</i> )
<b>template<typename</b> <i>ContainerType</i>>
<b>void BelongsToPoints</b> ( <b>const T&</b> <i>t1</i>, <b>const T&</b> <i>t2</i>,
<b>ContainerType&</b> <i>group1</i>, <b>ContainerType&</b> <i>group2</i>,
<b>std::vector<size_t>& <i>group1_ordinals</i>, <b>std::vector<size_t>& <i>group2_ordinals</i>)
returns the points closer to t1 than to t2 in group1 and the rest in group 2
if group1_ordinals and group2_ordinals are provided the ordinals of the
found objects in the object store are put into those vectors.
<b>template<typename</b> <i>ContainerTypeInside</i>, <b>typename</b> <i>ContainerTypeOutside</i>>
<b>void SeparateByRadius</b> ( <b>const DistanceType</b> <i>radius</i>, <b>const T&</b> <i>probe</i>,
<b>ContainerTypeInside&</b> <i>inside</i>, <b>ContainerTypeOutside&</b> <i>outside</i> )
<b>template<typename</b> <i>ContainerTypeInside</i>, <b>typename</b> <i>ContainerTypeOutside</i>>
<b>void SeparateByRadius</b> ( <b>const DistanceType</b> <i>radius</i>, <b>const T&</b> <i>probe</i>,
<b>ContainerTypeInside&</b> <i>inside</i>, <b>ContainerTypeOutside&</b> <i>outside</i>,
<b>std::vector<size_t>& <i>inside_ordinals</i>, <b>std::vector<size_t>& <i>outside_ordinals</i>)
return the points within radius of the probe in inside and the rest in outside
if inside_ordinals and outside_ordinals are provided the ordinals of the
found objects in the object store are put into those vectors.
<b>long FindInSphere</b> ( <b>const DistanceType&</b> <i>dRadius</i>,
<b>ContainerType&</b> <i>tInside</i>, <b>const T&</b> t ) <b>const</b>
<b>long FindInSphere</b> ( <b>const DistanceType&</b> <i>dRadius</i>,
<b>ContainerType&</b> <i>tInside</i>,
<b>std::vector<size_t>& tIndices, <b>const T&</b> t ) <b>const</b>
<i>dRadius</i> is the radius within which to search; make it very large if you want to
include every point that was loaded;
<i>tInside</i> is returned as the container of objects that were found within a radius dRadius
of the probe point
if the <i>tIndices</i> argument is given it will ve returned as a parallel vector
of indices in the near tree of the objects returned.
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the search radius
<b>long FindInSphere</b> ( <b>const DistanceType&</b> <i>dRadius</i>,
<b>CNearTree< T >&</b> <i>tInside</i>, <b>const T&</b> t ) <b>const</b>
<i>dRadius</i> is the radius within which to search; make it very large if you want to
include every point that was loaded;
<i>tInside</i> is returned as the CNearTree of objects that were found within a radius dRadius
of the probe point
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the search radius
<b>long FindOutSphere</b> ( <b>const DistanceType&</b> <i>dRadius</i>,
<b>ContainerType&</b> <i>tOutside</i>, <b>const T&</b> t ) <b>const</b>
<b>long FindOutSphere</b> ( <b>const DistanceType&</b> <i>dRadius</i>,
<b>ContainerType&</b> <i>tOutside</i>,
<b>std::vector<size_t>& tIndices, <b>const T&</b> t ) <b>const</b>
<i>dRadius</i> is the radius outside of which to search
<i>tOutside</i> is returned as the container of objects that were found at or outside of radius dRadius
of the probe point
if the <i>tIndices</i> argument is given it will ve returned as a parallel vector
of indices in the near tree of the objects returned.
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the search radius
<b>long FindOutSphere</b> ( <b>const DistanceType&</b> <i>dRadius</i>,
<b>CNearTree< T >&</b> <i>tOutside</i>, <b>const T&</b> t ) <b>const</b>
<i>dRadius</i> is the radius outside of which to search.
<i>tClosest</i> is returned as the CNearTree of objects that were found at or outside of a radius dRadius
of the probe point
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the search radius
<b>long FindInAnnulus</b> ( <b>const DistanceType&</b> <i>dRadius1</i>,
<b>const DistanceType&</b> <i>dRadius2</i>,
<b>ContainerType&</b> <i>tInRing</i>, <b>const T&</b> t ) <b>const</b>
<b>long FindInAnnulus</b> ( <b>const DistanceType&</b> <i>dRadius1</i>,
<b>const DistanceType&</b> <i>dRadius2</i>,
<b>ContainerType&</b> <i>tInRing</i>,
<b>std::vector<size_t>& tIndices, <b>const T&</b> t ) <b>const</b>
<i>dRadius1</i> and <i>dRadius2</i> are the two radii between which to find data points
<i>tInRing</i> is returned as the container of objects that were found at or outside of a radius dRadius1
and at or inside of radius dRadius2 of the probe point
if the <i>tIndices</i> argument is given it will ve returned as a parallel vector
of indices in the near tree of the objects returned.
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the annulus
<b>long FindInAnnulus</b> ( <b>const DistanceType&</b> <i>dRadius1</i>,
<b>const DistanceType&</b> <i>dRadius2</i>,
<b>CNearTree< T >&</b> <i>tInRing</i>, <b>const T&</b> t ) <b>const</b>
<i>dRadius1</i> and <i>dRadius2</i> are the two radii between which to find data points
<i>tInRing</i> is returned as the CNearTree of objects that were found at or outside of a radius dRadius1
and at or inside of radius dRadius2 of the probe point
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the annulus
<b>long FindK_NearestNeighbors</b> ( <b>const size_t</b> <i>k</i>, <b>const DistanceType&</b> <i>dRadius</i>,
<b>ContainerType&</b> <i>tClosest</i>, <b>const T&</b> t )
<b>long FindK_NearestNeighbors</b> ( <b>const size_t</b> <i>k</i>, <b>const DistanceType&</b> <i>dRadius</i>,
<b>ContainerType&</b> <i>tClosest</i>,
<b>std::vector<size_t>& tIndices, <b>const T&</b> t )
<i>k</i> is the maximum number of nearest neighbors to return. Finds this many if possible
<i>dRadius</i> within a sphere defined by dRadius, search for the k-nearest-neighbors
<i>tClosest</i> is returned ContainerType of the objects found within the sphere
if the <i>tIndices</i> argument is given it will ve returned as a parallel vector
of indices in the near tree of the objects returned.
<i>t</i> is the probe point, used to search in the group of points insert'ed
return value is the count of the number of points found within the sphere
<b>long FindK_NearestNeighbors</b> ( <b>const size_t</b> <i>k</i>, <b>const DistanceType&</b> <i>dRadius</i>,
<b>CNearTree< T >&</b> <i>tClosest</i>, <b>const T&</b> t )
<i>k</i> is the maximum number of nearest neighbors to return. Finds this many if possible
<i>dRadius</i> within a sphere defined by dRadius, search for the k-nearest-neighbors
<i>tClosest</i> is returned as the CNearTree of objects found within the sphere
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the sphere
<b>long FindK_FarthestNeighbors</b> ( <b>const size_t</b> <i>k</i>,
<b>ContainerType&</b> <i>tFarthest</i>, <b>const T&</b> t )
<b>long FindK_FarthestNeighbors</b> ( <b>const size_t</b> <i>k</i>,
<b>ContainerType&</b> <i>tFarthest</i>,
<b>std::vector<size_t>& tIndices, <b>const T&</b> t )
<i>k</i> is the maximum number of farthest neighbors to return. Finds this many if possible
<i>tFarthest</i> is returned ContainerType of the objects found
if the <i>tIndices</i> argument is given it will ve returned as a parallel vector
of indices in the near tree of the objects returned.
<i>t</i> is the probe point, used to search in the group of points insert'ed
return value is the count of the number of points found within the sphere
<b>long FindK_NearestNeighbors</b> ( <b>const size_t</b> <i>k</i>, <b>CNearTree< T >&</b> <i>tFarthest</i>, <b>const T&</b> t )
<i>k</i> is the maximum number of farthest neighbors to return. Finds this many if possible
<i>tFarthest</i> is returned as the CNearTree of objects found
<i>t</i> is the probe point, used to search in the group of points Insert'ed
return value is the count of the number of points found within the sphere
<hr>
Access Functions:
<b>T at</b> ( <b>const size_t</b> <i>n</i> ) <b>const</b>
returns the n'th item of the internal data store
<b>T operator[]</b> ( <b>const size_t</b> <i>n</i> )
returns the n'th item of the internal data store
<b>template<typename</b> <i>ContainerType</i>>
<b>operator ContainerType</b> ( <b>void</b> ) <b>const</b>
returns all of the inserted objects in the tree in a container of type ContainerType.
ContainerType can be std::vector<T>, etc, or other containers, including CNearTree<T>.
The returned vector contents are not guaranteed to be returned in the order loaded.
<b>iterator begin</b> ( <b>void</b> ) <b>const</b>
returns an iterator to the beginning of the internal data store
<b>iterator end</b> ( <b>void</b> ) <b>const</b>
returns an iterator to the end of the data store (one beyond the last item)
<b>iterator back</b> ( <b>void</b> ) <b>const</b>
returns an iterator to the last data item of the internal data store
<hr>
Information and special operation functions:
<b>void ImmediateInsert</b>( <b>const</b> <b>T</b>&aamp; <i>t</i> )
insert places objects in a queue for insertion later when CompleteDelayInsert
is called or a search is called. ImmediateInsert inserts the data immediately
into the tree (with the potential of a less balanced tree). ImmediateInsert is not
intended for the ordinary user.
<b>void CompleteDelayedInsert</b> ( <b>void</b> )
completes insertion for all delayed objects. sqrt(n) of them are inserted by
random choice. The rest are inserted in linear order as originally queued.
CompleteDelayedInsert is invoked at the beginning of all searches, so the
average user will never need to call it.
<b>size_t GetDeferredSize</b> ( <b>void</b> )
returns the number of delayed objects that have not yet completed insertion. This is
mainly for information about details of the tree.
<b>size_t GetTotalSize</b> ( <b>void</b> )
returns the number of objects that have been insert'ed plus those DelayInsert'ed
<b>size_t size</b> ( <b>void</b> )
identical to GetTotalSize
<b>size_t GetDepth</b> ( <b>void</b> )
returns the maximum tree layers from the root. This is mainly for information about
details of the tree.
<b>T Centroid</b> ( <b>void</b> )
returns the centroid of a neartree.
<b>bool empty</b> ( <b>void</b> )
returns true if the tree is empty, otherwise false
<hr>
Iterators:
Random access iterators are provided for accessing the data in a CNearTree. The most important
expected use is to retrieve the objects returned from one of the sphere search functions that
can return a CNearTree. However, they can be used with any CNearTree.
They should function in a fashion essentially the same as STL iterators. There is no assurance
that data will be returned in the order it was loaded, just that it is accessible. This is the
list of iterators. The same set is avaiable for const_iterator.
<b>iterator</b> ( <b>void</b> ) { }; // constructor
<b>iterator</b> ( <b>const const_iterator&</b> <i>s</i> )
<b>iterator& operator= </b> ( <b>const iterator&</b> <i>s</i> )
<b>iterator operator++ </b> ( <b>const int</b> <i>n</i> )
<b>iterator operator-- </b> ( <b>const int</b> <i>n</i> )
<b>iterator& operator++ </b> ( <b>void</b> )
<b>iterator& operator-- </b> ( <b>void</b> )
<b>iterator operator+ </b> ( <b>const long</b> <i>n</i> ) <b>const</b>
<b>iterator operator- </b> ( <b>const long</b> <i>n</i> ) <b>const</b>
<b>iterator& operator+= </b> ( <b>const long</b> <i>n</i> )
<b>iterator& operator-= </b> ( <b>const long</b> <i>n</i> )
<b>T operator* </b> ( <b>void</b> ) <b>const</b>
<b>bool operator== </b> ( <b>const iterator&</b> <i>t</i> ) <b>const</b>
<b>bool operator!= </b> ( <b>const iterator&</b> <i>t</i> ) <b>const</b>
<b>bool operator== </b> ( <b>const const_iterator&</b> </i>t</i> ) <b>const</b>
<b>bool operator!= </b> ( <b>const const_iterator&</b> </i>t</i> ) <b>const</b>
<b>bool operator> </b> ( <b>const iterator&</b> </i>t</i> ) <b>const</b>
<b>bool operator> </b> ( <b>const const_iterator&</b> </i>t</i> ) <b>const</b>
<b>bool operator< </b> ( <b>const iterator&</b> </i>t</i> ) <b>const</b>
<b>bool operator< </b> ( <b>const const_iterator&</b> </i>t</i> ) <b>const</b>
<b>const T * const operator-> </b> ( <b>void</b> ) <b>const</b>
<hr>
</tt>
</pre>
<p>
So a complete program is:
<p>
<pre>
<tt>
#include "TNear.h"
#include <cstdio>
void main()
{
CNearTree< double > dT;
double dNear;
dT.Insert( 1.5 );
if ( dT.NearestNeighbor( 10000.0, dNear, 2.0 )) printf( "%f\n",double(dNear-2.0) );
}
</tt>
</pre>
and it should print 0.5 (that's how for 2.0 is from 1.5). For more examples of
the use of TNear.h, see <a href="main.cpp">main.cpp</a> and <a href="CNearTreeTest.cpp">CNearTreeTest.cpp</a>.
<hr>
<hr>
<h4><a name="CNearTree.c">The C NearTree API: CNearTree.c</a></h4>
<h4>Synopsis</h4>
<P>
<blockquote>
<font size="-1">
<b>#include <CNearTree.h></b>
<P>
<p>
<b>double</b> <b>CNearTreeDist</b> ( <b>CNearTreeHandle</b> <i>treehandle</i>, <b>void *</b> <i>coord1</i>,
<b>void *</b> <i>coord2</i> );
<p>
<b>int</b> <b>CNearTreeSetNorm</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>, <b>int</b> <i>treenorm</i> );
<p>
<b>int</b> <b>CNearTreeNodeCreate</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>CNearTreeNodeHandle *</b> <i>treenodehandle</i> )
<p>
<b>int</b> <b>CNearTreeCreate</b> ( <b>CNearTreeHandle *</b> <i>treehandle</i>,
<b>size_t</b> <i>treedim</i>,
<b>int</b> <i>treetype</i> );
<p>
<b>int</b> <b>CNearTreeFree</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i> );
<p>
<b>int</b> <b>CNearTreeClear</b> ( <b>CNearTreeHandle *</b> <i>treehandle</i> );
<p>
<b>int</b> <b>CNearTreeNodeFree</b> ( <b>CNearTreeNodeHandle *</b> <i>treenodehandle</i> );
<p>
<b>int</b> <b>CNearTreeInsert</b>( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>const</b> <b>void *</b> <i>obj</i> );
<p>
<b>int</b> <b>CNearTreeNodeInsert</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>CNearTreeNodeHandle</b> <i>treenodehandle</i>,
<b>size_t</b> <i>index</i>;
<b>size_t *</b> <i>depth</i> );
<p>
<b>int</b> <b>CNearTreeImmediateInsert</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>const</b> <b>void *</b> <i>obj</i> );
<p>
<b>int</b> <b>CNearTreeDelayedInsert</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>const</b> <b>void *</b> <i>obj</i> ); /* ***DEPRECATED*** */
<p>
<b>int</b> <b>CNearTreeCompleteDelayedInsert</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i> )
<b>int</b> <b>CNearTreeZeroIfEmpty</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i> );
<p>
<b>int</b> <b>CNearTreeGetSize</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>, <b>size_t *</b> <i>size</i> );
<p>
<b>int</b> <b>CNearTreeGetDelayedSize</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>, <b>size_t *</b> <i>size</i> );
<p>
<b>int</b> <b>CNearTreeGetDepth</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>, <b>size_t *</b> <i>depth</i> )
<p>
<b>int</b> <b>CNearTreeNearestNeighbor</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>double</b> dRadius,
<b>void * *</b> <i>coordClosest</i>,
<b>void * *</b> <i>objClosest</i>,
<b>const</b> <b>void *</b> <i>coord</i> );
<p>
<b>int</b> <b>CNearTreeFarthestNeighbor</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>void *</b> * <i>coordFarthest</i>,
<b>void *</b> * <i>objFarthest</i>,
<b>const</b> <b>void *</b> <i>coord</i> );
<p>
<b>int</b> <b>CNearTreeFindInSphere</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CVectorHandle</b> <i>coordInside</i>,
<b>CVectorHandle</b> <i>objInside</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> );
<p>
<b>int</b> <b>CNearTreeFindTreeInSphere</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CNearTreeHandle</b> <i>foundInside</i>,
<b>const void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> )
<p>
<b>int</b> <b>CNearTreeFindOutSphere</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CVectorHandle</b> <i>coordOutside</i>,
<b>CVectorHandle</b> <i>objOutside</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> );
<p>
<b>int</b> <b>CNearTreeFindTreeOutSphere</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CNearTreeHandle</b> <i>foundOutside</i>,
<b>const void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> )
<p>
<b>int</b> <b>CNearTreeFindInAnnulus</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>double</b> <i>dRadiusInner</i>,
<b>const</b> <b>double</b> <i>dRadiusOuter</i>,
<b>CVectorHandle</b> <i>coordInRing</i>,
<b>CVectorHandle</b> <i>objInRing</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> );
<p>
<b>int</b> <b>CNearTreeFindTreeInAnnulus</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>double</b> <i>dRadiusInner</i>,
<b>const</b> <b>double</b> <i>dRadiusOuter</i>,
<b>CNearTreeHandle</b> <i>foundInRing</i>,
<b>const void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> )
<p>
<b>int</b> <b>CNearTreeFindKNearest</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>size_t</b> <i>k</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CVectorHandle</b> <i>coordClosest</i>,
<b>CVectorHandle</b> <i>objClosest</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> );
<p>
<b>int</b> <b>CNearTreeFindKTreeNearest</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>size_t</b> <i>k</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CNearTreeHandle</b> <i>foundClosest</i>,
<b>const void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> )
<p>
<b>int</b> <b>CNearTreeFindKFarthest</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>size_t</b> <i>k</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CVectorHandle</b> <i>coordFarthest</i>,
<b>CVectorHandle</b> <i>objFarthest</i>,
<b>const</b> <b>void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> );
<p>
<b>int</b> <b>CNearTreeFindKTreeFarthest</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>const</b> <b>size_t</b> <i>k</i>,
<b>const</b> <b>double</b> <i>dRadius</i>,
<b>CNearTreeHandle</b> <i>foundFarthest</i>,
<b>const void *</b> <i>coord</i>,
<b>int</b> <i>resetcount</i> )
<p>
<b>int</b> <b>CNearTreeNearest</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>double</b> * dRadius,
<b>void *</b> * <i>coordClosest</i>,
<b>void *</b> * <i>objClosest</i>,
<b>const</b> <b>void *</b> <i>coord</i> );
<p>
<b>int</b> <b>CNearTreeFindFarthest</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>double</b> * dRadius,
<b>void *</b> * <i>coordFarthest</i>,
<b>void *</b> * <i>objFarthest</i>,
<b>const</b> <b>void *</b> <i>coord</i> );
<p>
<b>int</b> <b>CNearTreeObjects</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>, <b>CVectorHandle *</b> <i>vectorhandle</i> );
<p>
<b>void *</b> <b>CNearTreeObjectAt</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>, <b>size_t</b> <i>index</i> );
<p>
<b>int</b> <b>CNearTreeCoords</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>, <b>CVectorHandle *</b> <i>vectorhandle</i> );
<p>
<b>void *</b> <b>CNearTreeCoordAt</b> ( <b>const</b> <b>CNearTreeHandle</b> <i>treehandle</i>,
<b>size_t</b> <i>index</i> );
</font>
</blockquote>
<p>
The NearTree API works with coordinate vectors in an arbitrary number of
dimensions. Each neartree is accessed by a pointer of type <b>CNearTreeHandle</b>
which points to a struct of type <b>CNearTree</b>, which points to a tree of
nodes of type CNearTreeNode:
<p>
<pre><tt>
typedef struct _CNearTreeNode {
size_t m_indexLeft; /* index of left coords in m_CoordStore
and of left object in m_ObjectStore */
size_t m_indexRight; /* index of right coords in m_CoordStore
and of right object in m_ObjectStore */
double m_dMaxLeft; /* longest distance from the left object
to anything below it in the tree */
double m_dMaxRight; /* longest distance from the right object
to anything below it in the tree */
struct _CNearTreeNode * m_pLeftBranch;
/* tree descending from the left object */
struct _CNearTreeNode * m_pRightBranch;
/* tree descending from the right object */
int m_iflags; /* flags */
} CNearTreeNode;
typedef CNearTreeNode * CNearTreeNodeHandle;
typedef struct {
CNearTreeNodeHandle m_ptTree; /* pointer to the actual tree */
size_t m_szdimension; /* dimension of the coordinates */
size_t m_szsize; /* size of this tree */
size_t m_szdepth; /* depth of this tree */
int m_iflags; /* flags */
CVectorHandle m_ObjectStore; /* all inserted objects */
CVectorHandle m_CoordStore; /* all inserted coordinates */
CVectorHandle m_DelayedIndices;/* objects queued for insertion */
} CNearTree;
typedef CNearTree FAR * CNearTreeHandle;
</pre></tt>
</p>
The internal operation of the API depends on the function <b>CNearTreeDist</b>
that returns the distance (L1, L2 or L-infinity) between two
coordinate vectors as a double according to the parameters of the given tree.
Note that the tree may store the coordinates as integers or as doubles, but
the distance is always computed as a double.
If this function is replaced by a user function, it is important that the
replacement obey the triangle inequality.
<P>
A neartree is created by <b>CNearTreeCreate</b> and freed by <b>CNearTreeFree</b>.
<i>treedim</i> is the dimension of the coordinate vectors and <i>treetype</i>
is one of the three predefined constants <b>CNEARTREE_TYPE_DOUBLE</b> for double
or <b>CNEARTREE_TYPE_INTEGER</b> for integer or <b>CNEARTREE_TYPE_STRING</b>,
optionally ORed with <b>CNEARTREE_NORM_L1</b>, <b>CNEARTREE_NORM_L2</b> or CNEARTREE_NORM_LINF</b>
for L1, L2 or L-infinity norms, <b>CNEARTREE_NORM_SPHERE</b> or <b>CNEARTREE_NORM_HSPHERE</b> for
a spherical or hemispherical norm (L1-norm combination of radial and spherical/hemisppherical
triangle distances), or <b>CNEARTREE_NORM_HAMMING</b> for the string-Hamming distance norm
(add one for each differing character position).
<p>
Starting with release 2.1, all insertions are delayed by default, unless the insertions is
done by a call to <b>CNearTreeImmediateInsert</b>. The insertions that have been queued
are completed by a call to <b>CNearTreeCompleteDelayedInsert</b> or by any search. The insertions
are actually done in a randomized order, either for an initial block of sqrt(#queue) by
default, or for the entire queue if the flag <b>CNEARTREE_DEFER_ALL</b> is ored with
<i>treetype</i>. In addition, if the flag <b>CNEARTREE_DEFER_ALL</b> is set, all inserts
will be deferred, even in calls to <b>CNearTreeImmediateInsert</b>. Alteratively, if minimal reorganization of
the order of the tree is desired on each insertion, <b>CNEARTREE_FLIP</b> should
be ored with <i>treetype</i>.
When first created, a neartree has no right or left node and with the dMax-below
set to negative values so that any match found will be stored since it will
greater than the negative value. The tree is then populated by calls to
<b>CNearTreeInsert</b>, with each call providing a coordinate vector <i>coord</i>
and an optional object pointer <i>obj</i>. The API copies the coordinate vector,
but does not copy the object. Later, when a search is requested or an explicit call
to <b>CNearTreeCompleteDelayedInsert</b> is made, the tree is populated in the order left, right
and then the nearer child, working from a randomized selection from the items queued
for insertion.
<p>
Optionally, the actual insertions may done immediately by calling <b>CNearTreeImmediateInsert</b>
instead of <b>CNearTreeInsert</b>. For upwards compatability of the library for existing
code, the deprecated <b>CNearTreeDelayedInsert</b> is provided as an deprecated alternate
call to <b>CNearTreeInsert</b>.
<p>
The neartree is searched for the nearest or farthest coordinate vector in the neartree to a given
probe coordinate vector <i>coord</i> by <b>CNearTreeNearestNeighbor</b> and
<b>CNearTreeFarthestNeighbor</b>, respectively. The given radius confines the search
to a sphere around the probe. If more than a single extremal coordinate point is
needed, <b>CNearTreeFindInSphere</b> can be used to obtain a CVector result vector of
all the coordinate vectors that satisfy the constraint of being within a specified
radius, or <b>CNearTreeFindOutSphere</b> can be used to obtain a CVector result vector
of all the coordinates that satisfy the constraint of being outside a specified radius.
<b>CNearTreeFindIn Annulus</b> can be used to obtain a CVector result vector of all the
coordinates that satisfy the constraint of being between two specified radii from the probe.
<b>CNearTreeFindKNearest</b> can be used to obtain a CVector result vector of the <i>k</i>
coordinates closest to the probe point such that all results are within the specified
radius of the probe point, or <b>CNearTreeFindKFarthest</b> to obtain a CVector result vector
of the <i>k</i> coordinates farthest from the probe point such that all results are at or outside the specified
radius of the probe point.
The vectors themselves
are not copied into the result vector. If the parameter <i>resetcount</i> is true (non
zero) the result vector is cleared before the search. A parallel CVector result vector
of the matching object pointers is returned if <i>objs</i> is not NULL.
Aternatives the forms <b>CNearTreeFindTreeInSphere</b>, <b>CNearTreeFindTreeOutSphere</b>,
<b>CNearTreeFindTreeInAnnulus</b>, <b>CNearTreeFindKTreeNearest</b>, <b>CNearTreeFindKTreeFarthest</b>
can be used to obtain CNearTrees rather than CVectors of results. The functions
<b>CNearTreeNearest</b> and <b>CNearTreeFindFarthest</b> implement <b>CNearTreeNearestNeighbor</b> and
<b>CNearTreeFarthestNeighbor</b>, respectively, adjusting the radius of the search while
the search is in progress, and are not normally used by users
<P>
<h4>Returns</h4>
If <b>CNearTreeDist</b> fails, it returns -1. Except for <b>CNearTreeDist</b>, all the functions
in the API return 0 ( <b>CNEARTREE_SUCCESS</b> ) for
success. If dynamic memory allocation fails, <b>CNEARTREE_MALLOC_FAILED</b> is returned.
If a call is made with an improper argument, <b>CNEARTREE_BAD_ARGUMENT</b> is returned.
If a search fails to find anything, <b>CNEARTREE_NOT_FOUND</b> is returned. If there is
a failure in an attempt to free a CNearTree, <b>CNEARTREE_FREE_FAILED</b> is returned. If
any of the internal call to CVector fail, <b>CNEARTREE_CVECTOR_FAILED</b> is returned. For
convenience in debugging, the formerly negative values of this returns are now positive.
<p>
<h4>Examples</h4>
<P>
To create a neartree for 3-dimensional vectors of doubles:
<P>
<pre>
<tt>
#include <CNearTree.h>
CNearTreeHandle treehandle;
int bReturn;
...
bReturn = !CNearTreeCreate(&treehandle,3,CNEARTREE_TYPE_DOUBLE);
</tt>
</pre>
<p>
To insert a copy of a 3-dimensional vector of doubles into this tree,
with no associated object:
<P>
<pre>
<tt>
double v[3];
...
v[0] = 1.; v[1] = 2.; v[2] = 3.;
bReturn = !CNearTreeInsert(treehandle,&v[0],NULL);
</tt>
</pre>
<p>
To search for the nearest neighbor to a probe vector vSearch in a radius of 3.,
returning a pointer to the resulting vector in vBest:
<P>
<pre>
<tt>
double * vBest;
void * vvBest;
double vSearch[3];
double dRad = =3.;
...
if ( !CNearTreeNearestNeighbor(treehandle,dRad,&vvBest,NULL,vSearch))
{ vBest = (double *)vvBest; }
</tt>
</pre>
<p>
Note the use of a separate void * vvBest instead of a cast of &vBest to avoid compiler
type punning warnings.
<P>
For more examples of the use of CNearTree.c, see <a href="main.c">main.c</a> and <a href="CNearTreeTest.c">CNearTreeTest.c</a>
in the release kit.
<hr>
<hr>
<h4><a name="rhrand.h">A Portable pseudo-random number generator: rhrand.h</a></h4>
<p>
rhrand.h is a portable pseudo-random number generator based one by Rob Harrison, derived from
"one in J.M.Hammersley and D.C. Handscomb, 'Monte Carlo
Methods,' Methuen & Co., London and Wiley & Sons, New
York, 1964, p47". See also, D. E. Knuth "The Art of
Computer Programming", Volume 2, "Seminumerical
Alogorithms, Third Edition, Addison-Wesley, Reading MA, 1997.
<p>
rhrand.h is a header file in which a C++ class, RHrand, is defined, and a
C struct typedef CRHrand is defined.
<p>
The C++ interface is
<p>
<pre><tt>
static const int RHRAND_MAX = 32767; /* the integer range accessible as RHrand::RHRAND_MAX */
RHrand(void) /* the default constructor */
RHrand( const int iseed ) /* a constructor to start with the given seed */
~RHrand( void) /* a destructor */
void srandom( const int iseed) /* reset the generator based on the given seed */
double urand( void ) /* return a random double uniformly distributed in [0,1) */
int random ( void ) /* return a random integer uniformly distributed in [0, RHRAND_MAX-1] */
</tt></pre>
<p>
In C++ code, typical use is
<pre><tt>
#include <rhhand.h>
RHrand rhr;
...
x = rhr.urand();
</pre></tt>
<p>
The C interface is suppressed in RHRAND_NOCCODE is defined. Otherwise the C interface is based on defining
a struct of type CRHRrand and calling macros that refer to a handle of type RCRHrandHandle.
<p>
<pre><tt>
typedef struct CRHrand_ { /* the struct used in random number generattion */
double buffer[55];
int indx;
int jndx;
int kndx;
double dTemp;
} CRHrand;
typedef CRHrand * CRHrandHandle; /* the type to be used in maro calls */
#define CRHRAND_MAX 32767 /* the integer range */
#define CRHrandSrandom(randhandle,iseed) ...
/* a macro to call to initialize CHRrandHandle randhandle
using see int iseed */
#define CRHrandUrand(randhandle) ... /* a macro to return a random double uniformly distributed in [0,1) */
#define CRHrandRandom(randhandle) ((int)(CRHrandUrand(randhandle)*(double)CRHRAND_MAX))
/* a macro to return a random integer uniformly distributed in
[0, CRHRAND_MAX-1] */
</tt></pre>
<p>
Typical use is
<pre><tt>
#include <rhhand.h>
CRHrand rhr;
...
CRHrandSrandom(&rhr, 0 );
...
x = CRHrandUrand(&rhr);
</pre></tt>
<hr>
<hr>
<font size="-1">
Updated 31 September 2010<br />
<script language="javascript" type="text/javascript">
<!--
var name = "andrewsL@";
var domain = "ix.netcom.com";
var domext = ".com";
document.write ("<a href=\"mailto:" + name + domain + domext + "\">" + name + domain + domext+"</a>");
// -->
</script>
<noscript>
andrewsl@ix.netcom.com.com
</noscript>
</font>
</font>
</body>
</html>
