Added thirdparty PSO

README.md

@@ -49,4 +49,5 @@ $ git clone https://github.com/GallVp/emgGO
 
 emgGO uses following third party libraries. The licenses for these libraries can be found next to source files in their respective libs/thirdpartlib folders.
 
-1. `energyop` Copyright (c) 2014, Hooman Sedghamiz. Source is available [here](https://au.mathworks.com/matlabcentral/fileexchange/45406-teager-keiser-energy-operator-vectorized).
+1. `energyop` Copyright (c) 2014, Hooman Sedghamiz. Source is available [here](https://au.mathworks.com/matlabcentral/fileexchange/45406-teager-keiser-energy-operator-vectorized).
+2. `PSOt` Copyright (c) 2005, Brian Birge. Source is available [here](https://au.mathworks.com/matlabcentral/fileexchange/7506-particle-swarm-optimization-toolbox).

libs/PSOt/DemoPSOBehavior.m

@@ -0,0 +1,148 @@
+% demopsobehavior.m
+% demo of the pso.m function
+% the pso tries to find the minimum of the f6 function, a standard
+% benchmark
+%
+% on the plots, blue is current position, green is Pbest, and red is Gbest
+
+% Brian Birge
+% Rev 3.0
+% 2/27/06
+
+clear all
+close all
+clc
+help demopsobehavior
+warning off
+
+functnames = {'ackley','alpine','DeJong_f2','DeJong_f3','DeJong_f4',...
+              'Foxhole','Griewank','NDparabola',...
+              'Rastrigin','Rosenbrock','f6','f6mod','tripod',...
+              'f6_bubbles_dyn','f6_linear_dyn','f6_spiral_dyn'};
+disp('Static test functions, minima don''t change w.r.t. time/iteration:');
+disp(' 1) Ackley');
+disp(' 2) Alpine');
+disp(' 3) DeJong_f2');
+disp(' 4) DeJong_f3');
+disp(' 5) DeJong_f4');
+disp(' 6) Foxhole');
+disp(' 7) Griewank');
+disp(' 8) NDparabola (for this demo N = 2)');
+disp(' 9) Rastrigin');
+disp('10) Rosenbrock');
+disp('11) Schaffer f6');
+disp('12) Schaffer f6 modified (5 f6 functions translated from each other)');
+disp('13) Tripod');
+disp(' ');
+disp('Dynamic test functions, minima/environment evolves over time/iteration:');
+disp('14) f6_bubbles_dyn');
+disp('15) f6_linear_dyn');
+disp('16) f6_spiral_dyn');
+
+functchc=input('Choose test function ? ');
+functname = functnames{functchc};
+
+disp(' ');
+disp('1) Intense graphics, shows error topology and surfing particles');
+disp('2) Default PSO graphing, shows error trend and particle dynamics');
+disp('3) no plot, only final output shown, fastest');
+plotfcn=input('Choose plotting function ? ');
+if plotfcn == 1
+   plotfcn = 'goplotpso4demo';
+   shw     = 1;   % how often to update display
+elseif plotfcn == 2
+   plotfcn = 'goplotpso';
+   shw     = 1;   % how often to update display
+else
+   plotfcn = 'goplotpso';
+   shw     = 0;   % how often to update display
+end
+
+
+% set flag for 'dynamic function on', only used at very end for tracking plots
+dyn_on = 0;
+if functchc==15 | functchc == 16 | functchc == 17
+   dyn_on = 1;
+end
+
+%xrng=input('Input search range for X, e.g. [-10,10] ? ');
+%yrng=input('Input search range for Y ? ');
+xrng=[-30,30];
+yrng=[-40,40];
+disp(' ');
+% if =0 then we look for minimum, =1 then max
+  disp('0) Minimize')
+  disp('1) Maximize')
+  minmax=input('Choose search goal ?');
+ % minmax=0;
+  disp(' ');
+  mvden = input('Max velocity divisor (2 is a good choice) ? '); 
+  disp(' ');
+  ps    = input('How many particles (24 - 30 is common)? ');
+  disp(' ');
+  disp('0) Common PSO - with inertia');
+  disp('1) Trelea model 1');
+  disp('2) Trelea model 2');
+  disp('3) Clerc Type 1" - with constriction');
+  modl  = input('Choose PSO model ? ');
+ % note: if errgoal=NaN then unconstrained min or max is performed
+  if minmax==1
+    %  errgoal=0.97643183; % max for f6 function (close enough for termination)
+      errgoal=NaN;
+  else
+     % errgoal=0; % min
+      errgoal=NaN;
+  end
+  minx = xrng(1);
+  maxx = xrng(2);
+  miny = yrng(1);
+  maxy = yrng(2);
+
+%--------------------------------------------------------------------------
+  dims=2;
+  varrange=[];
+  mv=[];
+  for i=1:dims
+      varrange=[varrange;minx maxx];
+      mv=[mv;(varrange(i,2)-varrange(i,1))/mvden];
+  end
+
+  ac      = [2.1,2.1];% acceleration constants, only used for modl=0
+  Iwt     = [0.9,0.6];  % intertia weights, only used for modl=0
+  epoch   = 400; % max iterations
+  wt_end  = 100; % iterations it takes to go from Iwt(1) to Iwt(2), only for modl=0
+  errgrad = 1e-99;   % lowest error gradient tolerance
+  errgraditer=100; % max # of epochs without error change >= errgrad
+  PSOseed = 0;    % if=1 then can input particle starting positions, if= 0 then all random
+  % starting particle positions (first 20 at zero, just for an example)
+   PSOseedValue = repmat([0],ps-10,1);
+
+  psoparams=...
+   [shw epoch ps ac(1) ac(2) Iwt(1) Iwt(2) ...
+    wt_end errgrad errgraditer errgoal modl PSOseed];
+
+ % run pso
+ % vectorized version
+  [pso_out,tr,te]=pso_Trelea_vectorized(functname, dims,...
+      mv, varrange, minmax, psoparams,plotfcn,PSOseedValue);
+
+
+%-------------------------------------------------------------------------- 
+% display best params, this only makes sense for static functions, for dynamic
+% you'd want to see a time history of expected versus optimized global best
+% values.
+disp(' ');
+disp(' ');
+disp(['Best fit parameters: ']);
+disp([' cost = ',functname,'( [ input1, input2 ] )']);
+disp(['---------------------------------']);
+disp(['      input1 = ',num2str(pso_out(1))]);
+disp(['      input2 = ',num2str(pso_out(2))]);
+disp(['        cost = ',num2str(pso_out(3))]);
+disp(['   mean cost = ',num2str(mean(te))]);
+disp([' # of epochs = ',num2str(tr(end))]);
+
+%% optional, save picture
+%set(gcf,'InvertHardcopy','off');
+%print -dmeta
+%print('-djpeg',['demoPSOBehavior.jpg']);

libs/PSOt/ReadME.txt

@@ -0,0 +1,112 @@
+-------------------------------------------------------------
+-------------------------------------------------------------
+PSOt, particle swarm optimization toolbox for matlab.
+
+May be distributed freely as long as none of the files are 
+modified. 
+
+Send suggestions to [email protected] 
+
+Updates will be posted periodically at the Mathworks User 
+Contributed Files website (www.mathworks.com) under the 
+Optimization category.
+
+To install:
+Extract into any directory you want but make sure the matlab 
+path points to that directory and the subdirectories 
+'hiddenutils' and 'testfunctions'. 
+
+Enjoy! - Brian Birge
+
+-------------------------------------------------------------
+-------------------------------------------------------------
+
+INFO
+Quick start: just type ... out = pso_Trelea_vectorized('f6',2) 
+and watch it work!
+
+This is a PSO toolbox implementing Common, Clerc 1", and 
+Trelea types along with an alpha version of tracking changing
+environments. It can search for min, max, or 'distance' of 
+user developed cost function. Very easy to use and hack with 
+reasonably good documentation (type help for any function and
+it should tell you what you need) and will take advantage of 
+vectorized cost functions. It uses similar syntax to Matlab's
+optimization toolbox. Includes a suite of static and dynamic 
+test functions. It also includes a dedicated PSO based neural 
+network trainer for use with Mathwork's neural network toolbox.
+
+Run 'DemoPSOBehavior' to explore the various functions, options, 
+and visualizations. 
+
+Run 'demoPSOnet' to see a neural net trained with PSO 
+(requires neural net toolbox).
+
+
+This toolbox is in constant development and I welcome 
+suggestions. The main program 'pso_Trelea_vectorized.m' lists 
+various papers you can look at in the comments.
+
+Usage ideas: to find a global min/max, to optimize training of 
+neural nets, error topology change tracking, teaching PSO, 
+investigate Emergence, tune control systems/filters, paradigm 
+for multi-agent interaction, etc.
+
+-------------------------------------------------------------
+-------------------------------------------------------------
+
+
+Files included:
+
+
+** in main directory:
+
+0) ReadMe.txt - this file, duh
+1) A Particle Swarm Optimization (PSO) Primer.pdf  -  powerpoint converted to pdf presentation explaining the very basics of PSO
+2) DemoPSOBehavior.m - demo script, useful to see how the pso main function is called
+3) goplotpso4demo.m - plotting routine called by the demo script, useful to see how custom plotting can be developed though this routine slows down the PSO a lot
+4) goplotpso.m - default plotting routine used by pso algorithm
+5) pso_Trelea_vectorized.m - main PSO algorithm function, implements Common, Trelea 1&2, Clerc 1", and an alpha version of tracking environmental changes.
+
+
+
+** in 'hiddenutils'
+
+1) forcerow, forcecol.m - utils to force a vector to be a row or column, superseded by Matlab 7 functions I believe but I think they are still called in the main algo
+2) normmat.m - takes a matrix and reformats the data to fit between a new range, very flexible
+3) linear_dyn, spiral_dyn.m - helpers for the dynamic test functions listed in the 'testfunctions' directory
+
+
+
+** in 'testfunctions'
+
+A bunch of useful functions (mostly 2D) for testing. See help for each one for specifics. Here's a list of the names:
+
+Static test functions, minima don't change w.r.t. time/iteration:
+ 1) Ackley
+ 2) Alpine
+ 3) DeJong_f2
+ 4) DeJong_f3
+ 5) DeJong_f4
+ 6) Foxhole
+ 7) Griewank
+ 8) NDparabola
+ 9) Rastrigin
+10) Rosenbrock
+11) Schaffer f6
+12) Schaffer f6 modified (5 f6 functions translated from each other)
+13) Tripod
+
+Dynamic test functions, minima/environment evolves over time (NOT iteration, though easily modifed to do so):
+14) f6_bubbles_dyn
+15) f6_linear_dyn
+16) f6_spiral_dyn
+
+
+
+** in 'nnet' (all these require Matlab's Neural Net toolbox)
+
+ 1) demoPSOnet - standalone demo to show neural net training
+ 2) trainpso   - the neural net toolbox plugin, set net.trainFcn to this
+ 3) pso_neteval - wrapper used by trainpso to call the main PSO optimizer, this is the cost function that PSO will optimize
+ 4) goplotpso4net - default graphing plugin for trainpso, shows net architecture, relative weight indications, error, and PSO details on run