adrianton3 / pso.js

Particle swarm optimization library

pso.js

Particle Swarm Optimisation library written in JS. Works with RequireJS, from a WebWorker, in node.js or in a plain browser environment.

Sample applications

• simple A simple application that optimizes a one dimensional function
• simple-require The same as simple, except using RequireJS
• simple-node A simple node example
• automaton A more sophisticated application that adapts a mechanism for a specified output path. pso.js is launched in this case by web workers
• circles A simple application that optimizes a two dimensional function
• shape-fitting Optimizes the positioning of arbitrary shapes in a square
• pool Optimizes the breaking shot of a pool game
• async Example of an asynchronous objective function
• parameters Optimizer performance when varying its parameters
• meta-optimizer pso.js is used to optimize the parameters of another instance of pso which is optimizing the Rastrigin function
• walking-critter Optimizing a "walking" critter - another example of asynchronous objective functions

Usage

Basic usage case

// create the optimizer
var optimizer = new pso.Optimizer();

// set the objective function
optimizer.setObjectiveFunction(function (x) { return -(x[0] * x[0] + x[1] * x[1]); });

// set an initial population of 20 particles spread across the search space *[-10, 10] x [-10, 10]*
optimizer.init(20, [{ start: -10, end: 10 }, { start: -10, end: 10 }]);

// run the optimizer 40 iterations
for (var i = 0; i < 40; i++) {
    optimizer.step();
}

// print the best found fitness value and position in the search space
console.log(optimizer.getBestFitness(), optimizer.getBestPosition());

####Optimizer parameters

Optimizer parameters can be set by calling the setOptions method before creating a population with the init method. Otherwise, the default parameters will be used. The setOptions method takes a single map-like object - here are its default values:

	{
		inertiaWeight: 0.8,
		social: 0.4,
		personal: 0.4,
		pressure: 0.5
	}

• inertiaWeight is multiplied every frame with the previous velocity
• social dictates how much a particle should be influenced by the best performing particle in the swarm
• personal indicates how much a particle should be influenced by the best position it has been in
• pressure is the bias in selecting the best performing particle in the swarm

For more details consult the annotated source.

pso.js's Issues

Best practice to auto-halt?

I'd like to remove as much guesswork as possible from the algo, and one way seems to be halting the optimization if it seems "reasonably converged."

Is there a fast/easy heuristic that I could plop on top of the getBestFitness score that would in many cases say "yup. Looks converged. Time to bail."