This is a repository of black-box-optimization using meta-heuristics.

Quick Start:

git clone [email protected]:pfnet-research/batch_metaheuristics.git
cd batch_metaheuristics
pip install ./batchopt
python example.py

Output (Black box optimization involves random elements, so the results will be different each time):

> Epoch: 1, Best fit: 884.7695325180321
> Epoch: 2, Best fit: -63.122234265670585
> Epoch: 3, Best fit: -85.04995732445039
> Epoch: 4, Best fit: -85.04995732445039
> Epoch: 5, Best fit: -111.70023680346456
> Epoch: 6, Best fit: -127.10859058276085
> Epoch: 7, Best fit: -131.96113684754317
> Epoch: 8, Best fit: -135.051712626202
> Epoch: 9, Best fit: -137.46424504437974
> Epoch: 10, Best fit: -138.53373567844406

Entire code of example.py is like this:

from batchopt.benchfunctions import batch_styblinski_tang
from batchopt.optimizers.whale_optimization_algorithm import WhaleOptimizationAlgorithm

opt = WhaleOptimizationAlgorithm(
    domain_range=[(-20, 20), (-20, 20), (-20, 20), (-20, 20)],
    pop_size=50,
    epoch=10,
    log=True,
)
objective_function = batch_styblinski_tang
# must be function 2-dimension-array to 1 dimension array

opt.optimize(objective_function)

This example uses WhaleOptimizationAlgorithm as optimizer. You can use other 20+ algorithms. Optimization algorithms are initialized with the following arguments:

• domain_range: List of domain of the objective function.
• pop_size: Number of candidate points to generate at one time.
• epoch: Number of times to repeat the optimization process.
• log: Whether to display logs.

Once you have created an instance of Optimizer, optimize with the optimize method. The argument of the optimize method is the objective function. The Objective function needs to be a mapping from a 2D-array to a 1D-array.

Reference implementation: https://github.com/thieu1995/mealpy