paper.md

title	tags	authors	affiliations	date	bibliography
Noisyopt: A Python library for optimizing noisy functions.	optimization stochastic approximation spsa	name orcid affiliation Andreas Mayer 0000-0002-6643-7622 1	name index Laboratoire de Physique Théorique, École Normale Supérieure 1	30 March 2017	paper.bib

Summary

Optimization problems have great practical importance across many fields. Sometimes a precise evaluation of the function to be optimized is either impossible or exceedingly computationally expensive. An example of the former case is optimization based on a complex simulation, an example of the latter arises in machine learning where evaluating a loss function over the complete data set can be to expensive. Optimization based on noisy evaluations of a function is thus an important problem.

Noisyopt provides optimization algorithms tailored to noisy problems with a call syntax compatible with scipy.optimize [@scipy] routines. It implements an derivative-free algorithm robust to noise -- adaptive pattern search [@adaptpattern], and a stochastic approximation algorithm using simultaneous perturbations [@spsa]. Bound constraints on variables are supported. The library also has methods for finding a root of a noisy function by an adaptive bisection algorithm.

References