A Distributed particle swarm optimization (PSO) implementation in erlang. Nicolas D. Mevlüt T.
Implemented for the project of the IN.5022 Concurrent and Distributed Computing [HS 18] course.
This project aims to implement a high performance, distributed particle swarm optimizer. For a short description on particle swarm algorithms, I recommend you take a look at
The latest version of the project can be downloaded by cloning the repository or directly downloading the latest release
The distributed PSO implementations can be found in
src/pso_global_topo.erl (global swarm topology)
src/pso_social_topo.erl (social swarm topology)
The sequential PSO implementation can be found in
src/pso_seq.erl
The implementations are meant to be run in the TEDA environment. The user should create a folder in the TEDA apps directory prior to deployment. For example, if a user wants to test the distributed swarm topology, he would need to create a directory apps/pso_global_topo and move the files pso_global_topo.erl and quickdeploy-global-topo.sh to the newly created directory.
After doing so, the user should update the variables in the script to match his environment. Once done, the script can be run:
./quickdeploy-global-topo.sh
To use the optimizer, you need to do the following
- Customize the cost function
Modify the function cost_function in the implementation file you want to run. Several test functions are already availble.
The main implementation is pso_global_topo.erl, which implements a global swarm topology.
The file pso_social_topo.erl is a slight variation, implementing a social swarm topology
- Call
pso_global_topo:start(N, W_s, W_c, Phi, Dim, Lo, Hi, Epochs), where
| Parameter name | Meaning | Recommended testing values |
|---|---|---|
N |
Number of particles | 10-100 |
W_s |
Social weight. Determines how much particles are attracted to the global minimum. | 1-2 |
W_c |
Cognitive weight. Determines how much particles are attracted to the local minimum. | 1-2 |
Phi |
Inertia of the particles | (0,1] |
Dim |
The dimension of the problem to optimize. | problem dependent |
Lo |
The lower search bound | problem dependent. |
Hi |
The upper search bound | problem dependent |
Epochs |
Number of epochs | 100-1000 |
If the particle swarm algorithm is used with a high amount of particles, then distributing the particles on different erlang nodes results in a massive performance boost, as can be seen in the following plot:
Both version where started with following parameters (Variable number of particles):
W_s = 2
w_c = 2
Phi = 0.75
Dim = 2
Lo = -5
Hi = 5
Epochs = 200
The function used to test the PSO implementation is Himmelblau's function:
