dmi_aboutneighborhoods.java

使用java编写的关于通讯及串口编程所必需的类库

/*
 * DMINeighborhoods.java
 *
 * Created on March 7, 2003, 1:10 PM
 */

package com.adaptiveview.ospso.dmi;

/** Documentation: About <b>PSO Neighborhoods</b>.
 * <p>Neighbors provide the "sociality factor" in the PSO algorithm. How a particle
 * "decides" to move is not only based upon the best (most fit) coordinates it has found
 * so far, but the best coordinates found by any particle in its neighborhood. In
 * a sense, the particle adapts its behaviour according to the successes of its
 * neighbors. This is very much in line with the sociocognitive view that mind and
 * intelligence emerge from an individual's social immersion.
 * <p>The neighborhood memberships are set <i>prior</i> to the <code>Swarm</code>
 * object distributing the particles into the solution space. That is, membership
 * in a neighborhood has nothing to do with how close particles are (or get) in the
 * solution space but to how they were arranged before they began hunting for good
 * solutions.
 * <p>How particles are arranged into neighborhoods is referred to as the <B>neighborhood
 * topology</B>. There are three <i>common</i> topologies. The first is a <B>Global Topology</B> in
 * which every particle is a neighbor of every other particle (i.e., there is just
 * one neighborhood and every particle is in it). 
 * The second topology is the <B>Star Topology</B> in which one particle is selected as a "hub"
 * and its neighborhood includes all other particles. All of the other particles, though, include
 * <I>only the hub particle</I> in their neighborhood.
 * The third is a <B>Circle Topology</B> in which the particles are arranged in a circle and then each particle includes
 * some (usually small) number of particles on its left and right in its neighborhood.
 * In this case, there is one neighborhood per particle and all adjacent neighborhoods
 * overlap (they are not identical but they share common members). 
 * <p><B>These Topologies</B> (using 7 particles: a, b, c, d, e, f and g):
 * <pre>
 *        Global                        Star                       Circle
 *        ------                       ------                       ----
 *
 *     a b c d e f g                     g                         g    a       
 *                                   f       b                 f           b   
 *                                       a                         
 *                                   e       c                  e         c    
 *                                       d                           d          
 * </pre>
 * <B>Would result in these Neighborhoods:</B>
 * <pre>
 *       abcdefg                 abcdefg, ba, ca,              abc, bcd, cde
 *                               da, ea, fa, ga             def, efg, fga, gab
 *                                                     (with neighborhood size = 3) 
 * </pre>
 *<p>The choice of neighborhood topology effects how quickly the influence of a
 * potential best (optimal) solution is propogated to all of the particles. In the
 * Global Topology, the propogation takes one search iteration. That is, the best
 * solution found by any particle at iteration <i>i</i> influences all particles at iteration <i>i+1</i>.
 * In the Star Topology, the influence is delayed one or more iterations. If, using
 * the above illustrations, particle <i>b</i> locates a new global best this influences
 * particle <i>a</i> at the next iteration but it is not until particle <i>a</i> moves to that 
 * location (assuming it is still the global best) that it influences the remaining
 * particles.
 *<p>In the Circle Topology, the global best may need to propogate through several
 * neighborhoods before all particles are influenced by it. If particle <i>b</i> locates a
 * global best, it won't influence particle <i>e</i> until particle <i>f</i> or <i>d</i> has moved to that
 * location. With larger, more typical swarm sizes the global best must become the neighborhood best across
 * many neighborhoods before it influences all particles (e.g., with a swarm of 40 particles and 
 * a neighborhood size = 5, a global best may have to be propogated across 8 or more neigborhoods before
 * it influences all particles).
 *<p>The significance of how a topology propogates a global best to all
 * neighborhoods is in what the delay in propogation allows
 * to happen: <b>more searching of the solution space</b>. A "global best"
 * is simply the best (most fit) location found by any particle <i>so far</i>. It may or may not be
 * the best location in the solution space. Delaying propogation means more exploration as
 * most particles will be oblivious to a newly found global best for many iterations
 * and, during those iterations, they'll be moving under the influence of different 
 * "neighborhood bests." This increases the probability that an optimum solution 
 * in the solution space will be found (rather than just the first good, but 
 * possibly sub-optimal solution some particle happens upon).
 *
 * @author AdaptiveView.com
 */
public interface DMI_AboutNeighborhoods extends DMI {}