Skip to main content

Heteroscedastic evolutionary bayesian optimisation

Project description

pip install HEBO

README

Bayesian optimsation library developed by Huawei Noahs Ark Decision Making and Reasoning (DMnR) lab. The winning submission to the NeurIPS 2020 Black-Box Optimisation Challenge.

Summary Ablation
Results Results

Contributors

Alexander I. Cowen-Rivers, Wenlong Lyu, Rasul Tutunov, Zhi Wang, Antoine Grosnit, Ryan Rhys Griffiths, Alexandre Max Maraval, Hao Jianye, Jun Wang, Jan Peters, Haitham Bou Ammar

Installation

You can install HEBO from pypi by

pip install HEBO

You can also build from source code to obtain our latest update:

cd HEBO
pip install -e .

Documentation

Online documentation can be seen here

You can also build the documentation by your self

pip install -r dev-requirements.txt
cd doc
make html

Demo

import pandas as pd
import numpy  as np
from hebo.design_space.design_space import DesignSpace
from hebo.optimizers.hebo import HEBO

def obj(params : pd.DataFrame) -> np.ndarray:
    return ((params.values - 0.37)**2).sum(axis = 1).reshape(-1, 1)
        
space = DesignSpace().parse([{'name' : 'x', 'type' : 'num', 'lb' : -3, 'ub' : 3}])
opt   = HEBO(space)
for i in range(5):
    rec = opt.suggest(n_suggestions = 4)
    opt.observe(rec, obj(rec))
    print('After %d iterations, best obj is %.2f' % (i, opt.y.min()))

Auto Tuning via Sklearn Estimator

from sklearn.datasets import load_boston
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import r2_score, mean_squared_error

from hebo.sklearn_tuner import sklearn_tuner

space_cfg = [
    {'name' : 'max_depth', 'type' : 'int', 'lb' : 1, 'ub' : 20},
    {'name' : 'min_samples_leaf', 'type' : 'num', 'lb' : 1e-4, 'ub' : 0.5},
    {'name' : 'max_features', 'type' : 'cat', 'categories' : ['auto', 'sqrt', 'log2']},
    {'name' : 'bootstrap', 'type' : 'bool'},
    {'name' : 'min_impurity_decrease', 'type' : 'pow', 'lb' : 1e-4, 'ub' : 1.0},
    ]
X, y   = load_boston(return_X_y = True)
result = sklearn_tuner(RandomForestRegressor, space_cfg, X, y, metric = r2_score, max_iter = 16)

Test

pytest -v test/ --cov ./bo --cov-report term-missing --cov-config ./test/.coveragerc

Reproduce Experimental Results

  • See archived_submissions/hebo, which is the exact submission that won the NeurIPS2020 Black-Box Optimsation Challenge.
  • Use run_local.sh in bbo_challenge_starter_kit to reproduce bayesmark experiments, you can just drop archived_submissions/hebo to the example_submissions directory.
  • The MACEBO in bo.optimizers.mace is the same optimiser, with same hyperparameters but a modified interface (bayesmark dependency removed).

Features

  • Continuous, integer and categorical design parameters.
  • Constrained and multi-objective optimsation.
  • Contextual optimsation.
  • Multiple surrogate models including GP, RF and BNN.
  • Modular and flexible Bayesian Optimisation building blocks.

Cite Us

Cowen-Rivers, Alexander I., et al. "An Empirical Study of Assumptions in Bayesian Optimisation." arXiv preprint arXiv:2012.03826 (2021).

BibTex

@article{Cowen-Rivers2022-HEBO,
author = {Cowen-Rivers, Alexander and Lyu, Wenlong and Tutunov, Rasul and Wang, Zhi and Grosnit, Antoine and Griffiths, Ryan-Rhys and Maravel, Alexandre and Hao, Jianye and Wang, Jun and Peters, Jan and Bou Ammar, Haitham},
year = {2022},
month = {07},
pages = {},
title = {HEBO: Pushing The Limits of Sample-Efficient Hyperparameter Optimisation},
volume = {74},
journal = {Journal of Artificial Intelligence Research}
}

Project details


Download files

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

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

HEBO-0.3.6-py3-none-any.whl (114.7 kB view details)

Uploaded Python 3

File details

Details for the file HEBO-0.3.6-py3-none-any.whl.

File metadata

  • Download URL: HEBO-0.3.6-py3-none-any.whl
  • Upload date:
  • Size: 114.7 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/5.1.1 CPython/3.9.13

File hashes

Hashes for HEBO-0.3.6-py3-none-any.whl
Algorithm Hash digest
SHA256 f3d46a106205eac5340822e5ad1aeb389109ea302dfe35f17d24e69c7c1d0665
MD5 03fd7619a08379d1d1c2ee09f29cd08e
BLAKE2b-256 462162d5e593b2b38cc1d2f148e89ad072fe973a3584492eab2d0c47c7b8c8e8

See more details on using hashes here.

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