For optimization, everyone starts out with the Scipy optimization library, but, at some point, you might want to try something else. NLOPT is a great library, but can be quite a hassle rewrite your code to use it.
This package provides a Scipy interface to the NLOPT optimization library. It's aim is not to provide a complete ecosystem which different solvers are available, but merely a way to quickly connect the NLOPT solvers, once you already have something set up with Scipy, or are familiar with how to set something up with Scipy.
import numpy as np
from spinopt import NLOptimizer
dim = 3
# Define objective in this way
def my_easy_func(x, grad):
if grad.size > 0:
grad[:] = 2 * (x - np.arange(len(x)))
x = x - np.arange(len(x))
return x.dot(x)
# Define constraints Scipy style
A = np.ones((1, dim))
b = np.ones((1, 1))
constraints = [{"type": "eq", "jac": lambda w: A, "fun": lambda w: A.dot(w) - b.squeeze()}]
# Initialize optimizer
x0=np.zeros(dim)
opt = NLOptimizer(my_easy_func, x0, constraints=constraints)
# Optimize
res = opt.minimize()
assert res.success
assert np.allclose(res.x, np.arange(dim), atol=1e-5)To install from PyPI:
pip install spinoptTo install the latest development version from github:
pip install git+https://github.com/mvds314/spinopt.gitFor development purposes, clone the repo:
git clone https://github.com/mvds314/spinopt.gitThen navigate to the folder containing setup.py and run
pip install -e .to install the package in edit mode.
Run unittests with pytest.