About this project

This is a fast c++ implementation of the C-DEEPSO algorithm. It was inspired by the Matlab implementation and the algorithm's original paper.

Dependencies

wup - a general purpose library with many of my reusable utilities.
clang++ - used to compile the program.

Getting started

Clone the repository and its submodule (wup).

git clone --recurse-submodules git@github.com:diegofps/cdeepso-cpp.git

Access the src folder and type make to build the example.

cd cdeepso-cpp/src
make

Running it.

# Default parameters
./main

# rastrigin function
./main -eval ras

# rosenbrock function
./main -eval ros

# griewank function
./main -eval gri

# Specify the number of threads, e.g., 16
./main -threads 16

# Same number of threads as CPU cores
./main -threads 0

# Max fitness evals
./main -maxFitEval 100000

# Max generations
./main -maxGen 50000

# Population size
./main -popSize 50

# Dimensinality
./main -dims 50

# Mutation rate
./main -mutationRate 0.5

# Communication probability
./main -communicationProbability 2.0

# Max particle velocity
./main -maxVelocity 2.0

# The number of times CDEEPSO will run (to obtain mean and average results)
./main -maxRun 50

Performance results

Results 2 (2023)

The following results were obtained in a machine running Ubuntu 22.04.2 LTS, AMD Threadripper 2990WX and 128GB of RAM. The parameters used are shown in the command lines bellow the table.

Function	Mean fitness (std)	Execution time in ms (std)
ras	4.94 (4.59)	498.23 (20.72)
ros	95.00 (1.27)	312.46 (20.74)
gri	0.00 (0.00)	454.37 (20.56)

# RAS
./main \
    -maxFitEval 100000 \
    -maxGen 50000 \
    -popSize 50 \
    -dims 100 \
    -memGBestSize 5 \
    -eval ras \
    -memStrategy POS_MEM \
    -deType RAND \
    -mutationRate 0.5 \
    -communicationProbability 0.1 \
    -printConvergenceResults 0 \
    -maxRun 1000 \
    -threads 64

# ROS
./main \
    -maxFitEval 100000 \
    -maxGen 50000 \
    -popSize 50 \
    -dims 100 \
    -memGBestSize 5 \
    -eval ros \
    -memStrategy POS_MEM \
    -deType RAND \
    -mutationRate 0.5 \
    -communicationProbability 0.1 \
    -printConvergenceResults 0 \
    -maxRun 1000 \
    -threads 64

# GRI
./main \
    -maxFitEval 100000 \
    -maxGen 50000 \
    -popSize 50 \
    -dims 100 \
    -memGBestSize 5 \
    -eval gri \
    -memStrategy POS_MEM \
    -deType RAND \
    -mutationRate 0.5 \
    -communicationProbability 0.1 \
    -printConvergenceResults 0 \
    -maxRun 1000 \
    -threads 64

Results 1 (2020)

The following results were obtained in a machine running Ubuntu 20.04.4 LTS, AMD Threadripper 2990WX and 128GB of RAM. The number of threads was 64 (-threads), the particles had 50 dimensions (-dims), and we executed CDEEPSO 100 times (-maxRun).

Function	Mean fitness	Std fitness	Execution time (ms)
ras	0	0	470.14
ros	43.46	2.31	317.91
gri	0	0	448.81