Particle swarm optimization: Difference between revisions

The goal of parameter selection is to ensure that the global minimum is discriminated from any local minima, and that the minimum is accurately determined, and that convergence is achieved with acceptibleacceptable resource usage. To provide a common basis for comparing implementations, the following test cases are recommended:

=={{header|C#|C sharp|C#}}==

<langsyntaxhighlight lang="csharp">using System;

}</langsyntaxhighlight>

<langsyntaxhighlight Dlang="d">import std.math;

}</langsyntaxhighlight>

<langsyntaxhighlight lang="go">package main

<langsyntaxhighlight Jlang="j">load 'format/printf'

reportState=: 'Iteration: %j\nGlobalBestPosition: %j\nGlobalBestValue: %j\n' printf 3&{.</langsyntaxhighlight>

Apply to McCormick Function:<langsyntaxhighlight Jlang="j"> require 'trig'

GlobalBestValue: _1.91322</langsyntaxhighlight>

<langsyntaxhighlight Jlang="j"> michalewicz =: 3 : '- +/ (sin y) * 20 ^~ sin (>: i. #y) * (*:y) % pi'

GlobalBestValue: _1.8013</langsyntaxhighlight>

<langsyntaxhighlight lang="java">import java.util.Arrays;

}</langsyntaxhighlight>

<langsyntaxhighlight lang="scala">// version 1.1.51

}</langsyntaxhighlight>

<langsyntaxhighlight lang="oorexx">/* REXX ---------------------------------------------------------------

::requires rxmath library</langsyntaxhighlight>

=={{header|PhixPerl}}==

{{trans|KotlinRaku}}

use constant infPI => 2 1e308*1e308 atan2(1, 0);

maxsmy($fn, =%y) {PI,= PI}@_;

paramsmy =$p {0.3, 0.3, 0.3 = $y{'p'};

nParticlesmy $n = 1000$y{'n'};

{x,y}my $rand_g = state[GBPOS]rand;

<langsyntaxhighlight lang="racket">#lang racket/base

(PSO (Michalewitz 10) 1000 30 (box 0) (box pi) #:ω 0.3 #:φ_p 0.3 #:φ_g 0.3 #:> <)</langsyntaxhighlight>

Classic REXX doesn't have a &nbsp; '''sine''' &nbsp; function, &nbsp; so a RYO version is included here.

This REXX version supports the specifying of &nbsp; '''X''', &nbsp; '''Y''', &nbsp; and &nbsp; '''D''', &nbsp; as well as the number of particles, &nbsp; and the number of decimal digits to be displayed.

<langsyntaxhighlight lang="rexx">/*REXX program calculates Particle Swarm Optimization as it migrates through a solution.*/

numeric digits length( pi() ) - 1 length(.) /*sDigs:use the isnumber theof #decimal ofdigs displayedin digitspi.*/

if x=='' | x=="," then x= -0.5 /*Not specified? Then use the default.*/

if y=='' | y=="," then y= -1.5 /* " " " " " " */

if d=='' | d=="," then d= 1 1 /* " " " " " " */

minFold=#part; iters=0; show=sDigs+3; old= /*number#part: of particles is1e12 ≡ is one trillion. */

call refine x,y #= 0; call refine x, y /*#: [↓]REFINE iterations; sameinvoke asREFINE. ÷ by five.*/

do until refine(minX, minY); d=d /*perform .2until the mix is /*decrease"refined". the difference.*/

refine: parse arg xx,yy; h= d * .5 /*compute ½ distance. */

do x=xx-d to xx+d by h

do y=yy-d to yy+d by h; f= f(x, y); if f>=minF then iterate

new= fmt(x) fmt(y) fmt(f); if new=old then return 1

iters#= iters# + 1; say center(#,13) new /*bump interations cntiter.; show new*/

say center(iters,13) new; minF= f; minX= x; minY= y; old= new /*assign new values. */

end /*y*/

end /*x*/

fmt: ?= format(arg(1), , sDigs); L= length(?); if pos(., ?)\==0 then ?= strip( strip(?, 'T', 0), "T", .); return left(?, L)

r2r: return arg(1) // ( pi() * 2) /*normalize radians ───► a unit circle.*/

sin: procedure; arg x; x= r2r( arg(1x); ); z=x; q xx= x*x; do k=2 by 2 until p=z; p=z; x= -x*q xx/ (k*(k+1)); z= z+x; end; return z</langsyntaxhighlight>

The global minimum for f(-.54719, -1.54719) ───► -1.9132229548822735814541188

The published global minimum is: -1.9133

::::: &nbsp; <u>[http://www.sfu.ca/~ssurjano/mccorm.html http://www.sfu.ca/~ssurjano/mccorm.html]</u>