PyCRO-SL

About

This project is a Python implementation of the Coral Reef Optimization with Substrate Layers algorithm, both in its original version (CRO-SL) and in its probabilistic versions (PCRO-SL and DPCRO-SL). The goal is to provide an easy-to-use, versatile Python package that can be used off-the-shelf for a wide array of optimization problems. Among its features, we highlight:

• Operators included: the package includes more than 30 solution operators, such as Gaussian noise addition, BLX Alpha, Multi-point crossover, etc.
• Multiple conditions: the algorithm can be configured to stop after a given number of evaluations, generations, or time, or after reaching a given fitness value. These stopping conditions can also be combined using any logical expression.
• Parallelization: the package allows for parallelization of the fitness function evaluations, which can significantly speed up optimization.
• Automatic report: the implementation produces a graphical report at the end of each simulation, summarizing the behavior of the most important metrics.
• Dynamic variant: the algorithm has a dynamic variant, which allows it to reward the operators that perform better in the current problem.

For an explanation of how the CRO-SL algorithm works, we refer to the paper.

How to install

We recommend installing the project as a package. To do so, use the following commands:

git clone https://github.com.jperezaracil/PyCROSL.git
cd PyCROSL
pip intall -r requirements.txt
pip install -e .

Then, you can import the algorithm as a package, for example

from PyCROSL.CRO_SL import CRO_SL

How to use it

To use PyCRO-SL in your own optimization problems, take a look at our tutorial.

Parameters

To configure the hyperparameters a dictionary will have to be given to the class CRO_SL. This dictionary contains the following parameters:

• Basic hyperparameters:
  • popSize: maximum number of corals in the reef
  • rho: percentage of initial occupation of the reef
  • Fb: broadcast spawning proportion
  • Fd: depredation proportion
  • Pd: depredation probability
  • k: maximum attempts for larva setting
  • K: maximum number of corals with duplicate solutions
  • group_subs: if True, corals reproduce only within the same substrate, if False they reproduce within the whole population
• Dynamic variant hyperparameters:
  • dynamic: boolean value that determines whether to use the dynamic variant of the algorithm
  • method: string that determines how to determine the probability of choosing each substrate. Possible values:
    • "fitness": uses the fitness of the individuals of each substrate.
    • "diff": uses the difference between the fitness of the previous generation and the current one.
    • "success": uses the ratio of successful larvae in each generation.
  • dyn_metric: string that determines how to aggregate the values of each substrate to get the metric of each. Possible values:
    • "best": takes the best fitness
    • "avg": takes the average fitness
    • "med": takes the median fitness
    • "worse": takes the worse fitness
  • dyn_steps: specifies the number of times the substrates will be evaluated. If dyn_steps = -1, the substrates will be evaluated every generation.
  • prob_amp: float that determines how the differences between substrate metrics affect the probability of each one. A lower value means more amplification
• Stopping conditions (you only need to include the ones that will be used!):
  • Neval: number of evaluations of the fitness function
  • Ngen: number of generations
  • time_limit: execution time limit given in seconds (real time, not CPU time)
  • fit_target: value of the fitness function we want to reach
  • stop_cond: a string that determines the stopping condition. It can be simply the name of the criterion to be used (e.g. Ngen, Neval, etc), or also a logical expression that combines these different criteria (e.g. Ngen or Neval, time_limit and fit_target, etc). Must be included even if only a single criterion is used.
• Parallelization:
  • Njobs: the number of jobs to run in parallel. If Njobs = 1, the algorithm will run in sequential mode.
• Display options:
  • verbose: shows a periodic report of the algorithm's performance
  • v_timer: amount of time between each report

For examples of how to fill the parameters dictionary, please take a look at main.py.

Cite

If you use this implementation, please cite it as:

@article{perez2023,
  title={New Probabilistic, Dynamic Multi-Method Ensembles for Optimization Based on the CRO-SL},
  author={P{\'e}rez-Aracil, Jorge and Camacho-G{\'o}mez, Carlos and Lorente-Ramos, Eugenio and Marina, Cosmin M and Cornejo-Bueno, Laura M and Salcedo-Sanz, Sancho},
  journal={Mathematics},
  volume={11},
  number={7},
  pages={1666},
  year={2023},
  publisher={MDPI}
}