Parallel Asynchronous Particle Swarm Optimization

The implementation of the algorithm written in the paper of Koh et al. (see literature folder or click here). This project was created using IntelliJ 2021.2 with Cursive Plugin 1.10.3 This project is part of the course 'Praktische Algorithmen der Bioinformatik und Computerlinguistik mit Lisp - SoSe 2021'at Ulm University.

Requirements

• Clojure 1.10.3 or higher
• Lein https://leiningen.org/
• Optional: Java 1.8 or higher

Usage

Algorithm

Init example:

(require '[algorithm.swarm :as psa])
(require '[testfunction.core :as atf])

(def dim 2)      ;; Dimension
(def gCount 10)  ;; 10 neighbourhoods, 1 for global best only
(def sSize 512)  ;; Number of particles
(def nIter 100)  ;; Max iterations
(def sRange 100) ;; Range of x, e.g. 100 = [-100,100]
(defn tF [x] (atf/h1 x)) ;; Function used to evaluate fitness of a position

(psa/setSwarmProperties dim gCount sSize sRange tF) ;; Set new swarm properties
(psa/resetSwarm) ;; Create swarm with new properties
...

Until this point you haven't started the algorithm. Start with one of the following methods:

Starts algorithm single threaded:

...
(def outSwarm (psa/psSync nIter map))

Starts algorithm multi-threaded aka parallel:

...
(def outSwarm (psa/psSync nIter pmap))

Starts algorithm multi-threaded with asynchronous updates (papso):

...
(def viewSwarm (psa/ps))

This time we retrieve a reference to the swarm which we can use to monitor progress of the individuals. Use (psa/stopPs) to stop papso.

You can use (sort-by psa/fitness (map deref psa/groupBest)) to retrieve a list of swarm best positions

More examples can be found in src/benchmark/core.clj

Visual Run example

Initialize the swarm as described above, but don't execute the PSO algorithm. Instead you call visualRun as follows:

(require '[algorithm.swarm :as psa])
(require '[testfunction.core :as atf])
(require '[visualize.core :as v])
(psa/setSwarmProperties 16 10 512 600 (fn [a] (-(atf/h3 a)))) ;;dim gCount sSize nIter sRange tF
(psa/resetPs)
(v/visualRun)

Legend:

• Green dot: A single particle
• Red dot: Particle best position
• White dot: Neighbourhood best position

Dimension reduction: Given all particles are of dimensionality d > 1, the plotting routine splits the dimensions into d/2 parts and plots every part in a 2-D plane. For example, if the dimension is set to 8 the plotting routine will show 4 windows with 2 dimensions each.

Known Problems

If the number of dimension is odd, the last dimension won't be visualized.

SEND COFFEE!