Vectorized BBOB functions in torch
Project description
BBOB torch
Implementation of BBOB functions (Real-Parameter Black-Box Optimization Benchmarking) as specified by https://coco.gforge.inria.fr/downloads/download16.00/bbobdocfunctions.pdf.
All the functions are vectorized and allow to pass potentional solutions in the same (num_of_solutions, problem_dimension)
.
Problem description
All the problems are represented by Problem
class. This class allows to:
- Evaluate your solutions by directly calling it
problem(solutions)
. - Get problem dimension
problem.dim
. - Get optimal solution resp. optimal function value using
problem.x_opt
resp.problem.f_opt
. - Get boundaries of solution using
problem.min
resp.problem.max
for each dimension. - Change underlying type or device using
problem.type(torch.float16)
andproblem.ty(torch.device('cuda:0'))
.
Problem creation
You can create new instance of each problem by calling corresponding create_fxx
function. This function accepts problem dimension and can optionally accept device and seed.
import torch
import bbobtorch
problem = bbobtorch.create_f09(40, dev=torch.device('cuda:0'), seed=42)
Example
import matplotlib.pyplot as plt
import numpy as np
import torch
import bbobtorch
x = torch.arange(-5,5, 0.01, dtype=torch.float32)
grid = torch.stack(torch.meshgrid(x, x), -1)
flat_grid = torch.reshape(grid, (-1,2))
xgrid, ygrid = np.meshgrid(x.numpy(), x.numpy())
fn = bbobtorch.create_f22(2, seed=42) # two dimension with seed 42
results = fn(flat_grid)
results_grid = torch.reshape(results, xgrid.shape) - fn.f_opt
plt.figure(figsize=(6,6))
plt.pcolormesh(xgrid, ygrid, results_grid, cmap='inferno', shading='nearest')
plt.scatter(*fn.x_opt.tolist()[::-1], marker='x', c='r')
plt.show()
You can view all the functions in attached PDF.
Author: Patrik Valkovič
License: MIT
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for BBOBtorch-1.0.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 3fe277d706546e2a8597f0443266841fc175b73a3191344f3feb8c78040a3bb4 |
|
MD5 | 816ecb1ffabdf41c68cf4f46d6c010b2 |
|
BLAKE2b-256 | 99322abfa0aa45e7858988070463bb9214aa84e47ff8d53fbee5ebade692aabc |