neighborhood

Travis CI:

Python 3 implementation of “neighborhood algorithm” direct-search optimization and Bayesian ensemble appraisal. In short, a nearest-neighbor interpolant based on Voronoi polygons is used to interpolate the misfit (search) and posterior probability (appraisal) to allow efficient sampling and integration for high-dimensional problems. Details on theory and implementation are supplied in the references.

https://raw.githubusercontent.com/keithfma/neighborhood/master/example_rosenbrock_4d.png — Example search population for 4D Rosenbrock objective function. Image include 10,000 samples collected in 1,000 iterations of the neighborhood algorithm direct search, with num_samp=10 and num_resamp=5. The true minimum is 0 at (1, 1, 1, 1), while the best sample is 0.0113 at ((0.976, 0.953, 0.908, 0.824). This result continues to converge for larger sample size (but the plot is less interesting since the density converges to a point!)

To generate the example figure above, you can run the internal demo, like so:

import neighborhood as nbr

nbr.demo_search(ndim=4, nsamp=10, nresamp=5, niter=500)

Equivalently, you can do the following:

import neighborhood as nbr

num_dim = 4
srch = nbr.Searcher(
    objective=nbr.rosenbrock,
    limits=[(-1.5, 1.5) for _ in range(num_dim)],
    num_samp=10,
    num_resamp=5,
    maximize=False,
    verbose=True
    )
srch.update(500)
srch.plot()

Status

Optimization is implemented, ensemble appraisal is in progress.

Testing

This project uses pytest for unit testing. The aim is not to be exhuastive, but to provide reasonable assurances that everything works as advertised. To run, simply call pytest --verbose from somewhere in this package.

Release

Release versions are tagged in the repository, built as distributions, and uploaded to PyPI. The minimal commands to do this are:

# update PyPI-readable README
pandoc --from=markdown --to=rst --output=README.rst README.md
# build with setuptools
python3 setup.py sdist bdist_wheel
# upload to PyPI test server (then check it out)
twine upload --repository-url https://test.pypi.org/legacy/ dist/*
# upload to PyPI
twine upload dist/*
# tag release in git repo
git tag -a X.X.X -m "vX.X.X"
git push origin --tags

References

Sambridge, M. (1999). Geophysical inversion with a neighbourhood algorithm - I. Searching a parameter space. Geophysical Journal International, 138(2), 479–494. http://doi.org/10.1046/j.1365-246X.1999.00876.x
Sambridge, M. (1999). Geophysical inversion with a neighborhood algorithm -

Appraising the ensemble. Geophys, J. Int., 138, 727–746. http://doi.org/10.1046/j.1365-246x.1999.00900.x

Metadata

Release files for neighborhood 0.1.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for neighborhood 0.1.1
File	Size	Uploaded
neighborhood-0.1.1.tar.gz	8.1 kB	Apr 12, 2018	Details

Built distribution (wheel)

Table of built distributions (wheels) for neighborhood 0.1.1
File	Interpreter	ABI	Platform	Reset
neighborhood-0.1.1-py3-none-any.whl 11.1 kB Apr 12, 2018	Python 3	none	any	Details

Total release size: 19.2 kB

Release files / neighborhood-0.1.1.tar.gz

Download URL	neighborhood-0.1.1.tar.gz
Size	8.1 kB
Tags	Source
SHA-256 checksum How to use checksums	`0355ec5e0fe292814caeee3fbdff51bbe18d34f79ae490cd0574bff290b871d3`
BLAKE2b-256 checksum How to use checksums	`7f5fa211bd7d7801e3fbe7576b79baa80afa3806cdf49dbf4735a41396913f91`
Upload date	Apr 12, 2018
Uploaded using Trusted Publishing? What is trusted publishing?	No

Release files / neighborhood-0.1.1-py3-none-any.whl

Download URL	neighborhood-0.1.1-py3-none-any.whl
Size	11.1 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`eb4fea44253f0521f2953231e3122fb1358141d1064704093df72d52a82b9a35`
BLAKE2b-256 checksum How to use checksums	`876ad6fb49e1da702779d58940df16834c54bdae1b4f62f0f0b7410d0c54033f`
Upload date	Apr 12, 2018
Uploaded using Trusted Publishing? What is trusted publishing?	No