# PyBenchFCN #

A python implementation of optimization benchmarks

• How to Install
• How to Use
  • Classical Single-Objective Optimization
    • Setup Benchmark Function
    • Plot Fitness Landscape
• List of Functions
  • Classical Single-Objective Optimization
  • Discrete Optimization
  • Multi-Objective Optimization
  • Real-World Optimization
• Authors
• License
• Acknowledgement

How to Install

You can simply install with command pip install PyBenchFCN.

• Pre-request: numpy, matplotlib

How to Use

Classical Single-Objective Optimization

The input of each numerical optimization problem could be a 1-D ndarray, or 2-D ndarray.

• 1-D array
  • an example of a solution (individual) for 10D problem is np.random.uniform(0, 1, 10), where each entry is a decision variable.
  • use f() to return a fitness value (scalar).
• 2-D array
  • an example of group of solutions (population) for 10D problem is np.random.uniform(0, 1, (5, 10)), where each row (totally 5) is an individual.
  • use F() to return an array of fitness value (1-D array).

Setup Benchmark Function

To set a benchmark function, one may see the sample code in Factory.py in the repository, or follow the script below. Also, there is a sample optimization program provided in sample.py.

import numpy as np

# import single objective problems from PyBenchFCN
from PyBenchFCN import SingleObjectiveProblem as SOP

n_var = 10                                      # dimension of problem
n_pop = 3                                       # size of population

problem = SOP.ackleyfcn(n_var)                  # Ackley problem

'''same function as the code above
from PyBenchFCN import Factory
problem = Factory.set_sop("f1", n_var)
'''

print( np.round(problem.optimalF, 5) )          # show rounded optimal value

xl, xu = problem.boundaries                     # bound of problem

x = np.random.uniform(xl, xu, n_var)            # initialize a solution
print( problem.f(x) )                           # show fitness value as scalar

X = np.random.uniform( xl, xu, (n_pop, n_var) ) # initialize a population
print( problem.F(X) )                           # show fitness values as 1d-array

Plot Fitness Landscape

To plot a fitness landscape (2D space), one can use the code below. Notice, this function only works for continuous SOPs.

from PyBenchFCN import Tool
Tool.plot_sop("sphere", mode="save")            # plot and save landscape of Sphere function
Tool.plot_sop("schwefel", plot_type="contour")  # plot contour plot of Schwefel function

List of Functions

Classical Single-Objective Optimization

Totally, 61 single-objective functions are implemented based on BenchmarkFcns Toolbox. The plot of 2D versions of 59 problems are provided. Please check the homepage of BenchmarkFcns Toolbox or this website for the further documentation.

Discrete Optimization

Under development ...

Multi-Objective Optimization

Under development ...

Real-World Optimization

Under development ...

Authors

Yifan He @ Dept. of CS, UTsukuba

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgement

PyBenchFCN is maintained by Yifan He. The author of this repostory is very grateful to Mr. Mazhar Ansari Ardeh, who implemented the MatLab package BenchFCNs Toolbox.

• If you find any mistakes, please report at a new issue.
• If you want to help me implement more benchmarks (discrete optimization, multi-objective optimization), please contact at [email protected].