optimparallel 0.0.6

A parallel version of the L-BFGS-B optimizer of scipy.optimize.minimize().

optimparallel - A parallel version of scipy.optimize.minimize(method='L-BFGS-B')

Using optimparallel.minimize_parallel() can significantly reduce the optimization time. For an objective function with an execution time of more than 0.1 seconds and p parameters the optimization speed increases by up to factor 1+p when no analytic gradient is specified and 1+p processor cores with sufficient memory are available.

A similar extension of the L-BFGS-B optimizer exists in the R package optimParallel:

• optimParallel on CRAN
• R Journal article

Installation

To install the package run:

$ pip install optimparallel

Usage

Replace scipy.optimize.minimize(method='L-BFGS-B') by optimparallel.minimize_parallel() to execute the minimization in parallel:

from optimparallel import minimize_parallel
from scipy.optimize import minimize
import numpy as np
import time

## objective function
def f(x, sleep_secs=.5):
    print('fn')
    time.sleep(sleep_secs)
    return sum((x-14)**2)

## start value
x0 = np.array([10,20])

## minimize with parallel evaluation of 'fun' and
## its approximate gradient.
o1 = minimize_parallel(fun=f, x0=x0, args=.5)
print(o1)

## test against scipy.optimize.minimize()
o2 = minimize(fun=f, x0=x0, args=.5, method='L-BFGS-B')
print(all(np.isclose(o1.x, o2.x, atol=1e-10)),
      np.isclose(o1.fun, o2.fun, atol=1e-10),
      all(np.isclose(o1.jac, o2.jac, atol=1e-10)))

The evaluated x values, fun(x), and jac(x) can be returned:

o1 = minimize_parallel(fun=f, x0=x0, args=.5, parallel={'loginfo': True})
print(o1.loginfo)

More examples are given in example.py and example_extended.ipynb.

Note for Windows users: It may be necessary to run minimize_parallel() in the main scope. See example_windows_os.py.

Author

• Florian Gerber, [e-mail], https://user.math.uzh.ch/gerber.

Contributions

Contributions via pull requests are welcome.

• https://github.com/florafauna/optimParallel-python
• https://pypi.org/project/optimparallel

Thanks to contributors:

• Lewis Blake