Description:
ibcdfo 1.0.0b1
IBCDFO
Interpolation-Based Composite Derivative-Free Optimization
This page contains source code for interpolation-based optimization methods for
composite derivative-free optimization.
Relevant references include:
J. Larson and M. Menickelly. Structure-aware methods for expensive
derivative-free nonsmooth composite optimization. arXiv:2207.08264. 2022.
LINK
J. Larson, M. Menickelly, and B. Zhou. Manifold sampling for optimizing
nonsmooth nonconvex compositions. SIAM Journal on Optimization.
31(4):2638–2664, 2021
DOI
K. A. Khan, J. Larson, and S. M. Wild. Manifold sampling for optimization of
nonconvex functions that are piecewise linear compositions of smooth
components. SIAM Journal on Optimization 28(4):3001--3024, 2018,
DOI
S. M. Wild. POUNDERS in TAO: Solving Derivative-Free Nonlinear
Least-Squares Problems with POUNDERS. Advances and Trends in Optimization with
Engineering Applications. SIAM. 529--539, 2017.
DOI
J. Larson, M. Menickelly, and S. M. Wild. Manifold sampling for l1 nonconvex
optimization. SIAM Journal on Optimization. 26(4):2540–2563, 2016.
DOI
Contributing to IBCDFO
Contributions are welcome in a variety of forms; please see CONTRIBUTING.
Installation & Updating
Note that this repository depends on one or more submodules. After cloning
this repository, from within the clone please run
git submodule update --init --recursive
to fetch all files contained in the submodules. This must be done before
attempting to use the code in the clone. Issuing the command git pull will
update the repository, but not the submodules. To update the clone and all its
submodules simultaneously, run
git pull --recurse-submodules.
The ibcdfo python package can be installed by setting up a terminal with the
target python and pip pair and executing
> pushd ibcdfo_pypkg
> python setup.py sdist
> pip install dist/ibcdfo-<version>.tar.gz
> popd
where <version> can be determined by looking at the output of the sdist
command. The installation can be partially tested by executing
> python
>>> import ibcdfo
>>> ibcdfo.__version__
<version>
where the output <version> should be identical to the value used during
installation.
License
All code included in IBCDFO is open source, with the particular form of license contained in the top-level
subdirectories. If such a subdirectory does not contain a LICENSE file, then it is automatically licensed
as described in the otherwise encompassing IBCDFO LICENSE.
Resources
To seek support or report issues, e-mail:
poptus@mcs.anl.gov
License
For personal and professional use. You cannot resell or redistribute these repositories in their original state.
Customer Reviews
There are no reviews.
