IN DEVELOPMENT: benchfuncs

The benchfuncs package offers a selection of benchmark functions that can be used to test optimisation algorithms. Below is an example that shows (a) how bencfuncs can be used to generate training points for modelling and machine learning and (b) how functions can be solved with an optimiser (in this case the L-BFGS-B solver from the scipy.optimizepackage.)

from benchfuncs import Sphere
import numpy as np
from scipy.optimize import minimize


# define benchmark function
func = Sphere(dims=4)

# sample training data
x = np.random.uniform(low=func.bounds[:, 0], high=func.bounds[:, 1], size=(3, func.dims))

# get training data outpus
y = func(x)

# print results
print("Inputs: ", x)
print("Outputs: ", y)

Inputs:  [[-2.69264719  2.34062409 -0.16280928 -0.55581425]
          [ 1.60939843  0.28119955 -0.47671281 -0.51286048]
          [ 2.32721282  0.32551319 -1.25113608 -4.29779957]]
Outputs:  [13.0643064   3.15951746 25.55830099]

# optimise benchmark function
x0 = np.random.uniform(low=func.bounds[:, 0], high=func.bounds[:, 1], size=func.dims)
results = minimize(func, x0, method='L-BFGS-B', bounds=func.bounds)

# compare solution with global minimum
print(f"Solution: \t Inputs: {results['x']} \t Output: {results['fun']}")
print(f"Optimum: \t Inputs: {func.optimum['inputs']} \t Output: {func.optimum['output']}")

Solution:        Inputs: [-4.92781912e-09 -4.96113870e-09 -4.99729987e-09 -5.03493624e-09]      Output: 9.921988747713389e-17
Optimum:         Inputs: [[0. 0. 0. 0.]]         Output: [0.]