Skip to main content

Bayesian optimization structure search

Project description

Bayesian Optimization Structure Search (BOSS) is a general-purpose Bayesian Optimization code. It is designed to facilitate machine learning in computational and experimental natural sciences.

For a more detailed description of the code and tutorials, please consult the user guide.

Installation

BOSS is distributed as a PyPI package and can be installed using pip:

python3 -m pip install aalto-boss

We recommend installing BOSS inside a virtual environment (venv, conda…). If you are not using virtual environments, we recommend performing a user-installation instead:

python3 -m pip install --user aalto-boss

Basic usage

As an easy example, consider the optimization of a bounded 1D function. BOSS can be run either directly from Python or via a CLI interface, both these approaches are illustrated briefly below. Note that BOSS always minimizes a given function.

Python iterface

To run BOSS from Python we first define our objective function, by default BOSS expects this function to take a single 2D numpy array as argument (this behaviour can be modified) and return a scalar value. Next, we import the BOMain object and feed it the function plus any number of BOSS keywords, after which the optimization can be started. Once finished, the optimziation results are returned in a BOResults object.

""" Using BOSS to solve the minimization problem
f(x) = sin(x) + 1.5*exp(-(x-4.3)**2) , 0 < x < 7
"""
import numpy as np
from boss.bo.bo_main import BOMain
from boss.pp.pp_main import PPMain

def func(X):
    """ BOSS-compatible definition of the function. """
    x = X[0, 0]
    return np.sin(x) + 1.5*np.exp(-(x - 4.3)**2)

if __name__ == '__main__':
    bo = BOMain(
        func,
        np.array([[0., 7.]]),  # bounds
        kernel='rbf',
        initpts=5,
        iterpts=15,
    )
    res = bo.run()

    # print global minimum value and location from the last iteration
    print(res.select('mu_glmin', -1), res.select('x_glmin', -1))

    # run BOSS post-processing
    # creates a folder "postprocessing" with plots and data
    pp = PPMain(res, pp_models=True, pp_acq_funcs=True)
    pp.run()

Command-line iterface

The CLI is provided by an executable called boss. The user must provide an input file containing a list of BOSS keywords and a separate Python script that defines a function to be optimized. By default, BOSS expects this function to take a single 2D numpy array as argument (this behaviour can be modified) and return a scalar value. Below we define such a function in a Python script, arbitrarily named user_function.py:

""" user_function.py
This script contains the function definition for the minimization problem
f(x) = sin(x) + 1.5*exp(-(x-4.3)**2) ,  0 < x < 7
Note that the bounds are specified in the BOSS input file.
"""
import numpy as np

def func(X):
    """ BOSS-compatible definition of the function. """
    x = X[0, 0]
    return np.sin(x) + 1.5*np.exp(-(x - 4.3)**2)

To minimize this function subject to the constraint 0 < x < 7, we define a BOSS input file boss.in:

# boss.in
userfn        user_function.py func
bounds        0 7
yrange        -1 1
kernel        rbf
initpts       5
iterpts       15

The optimization (including post-processing) can now be started from the command line:

$ boss op boss.in

Credits

BOSS is under active development in the Materials Informatics Laboratory at the University of Turku and the Computational Electronic Structure Theory (CEST) group at Aalto University. Past and present members of development team include

  • Ville Parkkinen

  • Henri Paulamäki

  • Arttu Tolvanen

  • Ulpu Remes

  • Nuutti Sten

  • Emma Lehto

  • Tuomas Rossi

  • Manuel Kuchelmeister

  • Mikael Granit

  • Ransell D’souza

  • Armi Tiihonen

  • Matthias Stosiek

  • Joakim Löfgren (maintainer)

  • Milica Todorović (team lead)

If you wish to use BOSS in your research, please cite

Milica Todorovic, Micheal U. Gutmann, Jukka Corander, and Patrick Rinke
Bayesian inference of atomistic structure in functional materials
npj Comput Mater 5, 35 (2019)

Issues and feature requests

It is strongly encouraged to submit bug reports, questions, and feature requests via the gitlab issue tracker. The BOSS development team can be contacted by email at milica.todorovic@utu.fi

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

aalto-boss-1.9.1.tar.gz (81.4 kB view hashes)

Uploaded Source

Built Distribution

aalto_boss-1.9.1-py3-none-any.whl (98.0 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page