PyXAB 0.3.0

PyXAB - A Python Library for X-Armed Bandit and Online Blackbox Optimization Algorithms.

PyXAB - Python X-Armed Bandit

PyXAB is a Python open-source library for X-armed bandit algorithms, a prestigious set of optimizers for online black-box optimization and hyperparameter optimization.

PyXAB contains the implementations of 10+ optimization algorithms, including the classic ones such as Zooming, StoSOO, and HCT, and the most recent works such as GPO, StroquOOL and VHCT. PyXAB also provides the most commonly-used synthetic objectives to evaluate the performance of different algorithms and the implementations for different hierarchical partitions

PyXAB is featured for:

• User-friendly APIs, clear documentation, and detailed examples
• Comprehensive library of optimization algorithms, partitions and synthetic objectives
• High standard code quality and high testing coverage
• Low dependency for flexible combination with other packages such as PyTorch, Scikit-Learn

Reminder: The algorithms are maximization algorithms!

Quick Links

• Quick Example
• Documentations
• Installation
• Features
  • X-armed bandit algorithms
  • Hierarchical partition
  • Synthetic objectives
• Contributing
• Citations

Quick Example

PyXAB follows a natural and straightforward API design completely aligned with the online blackbox optimization paradigm. The following is a simple 6-line usage example.

First, we define the parameter domain and the algorithm to run. At every round t, call algo.pull(t) to get a point and call algo.receive_reward(t, reward) to give the algorithm the objective evaluation (reward)

domain = [[0, 1]]               # Parameter is 1-D and between 0 and 1
algo = T_HOO(rounds=1000, domain=domain) 
for t in range(1000):
    point = algo.pull(t)
    reward = 1                  #TODO: User-defined objective returns the reward
    algo.receive_reward(t, reward)

More detailed examples can be found here

Documentations

• The most up-to-date documentations

• The roadmap for our project

• Our manuscript for the library

Installation

To install via pip, run the following lines of code

pip install PyXAB                 # normal install
pip install --upgrade PyXAB       # or update if needed

To install via git, run the following lines of code

git clone https://github.com/WilliamLwj/PyXAB.git
cd PyXAB
pip install .

Features:

X-armed bandit algorithms

• Algorithm starred are meta-algorithms (wrappers)

Algorithm	Research Paper	Year
Zooming	Multi-Armed Bandits in Metric Spaces	2008
T-HOO	X-Armed Bandit	2011
DOO	Optimistic Optimization of a Deterministic Function without the Knowledge of its Smoothness	2011
SOO	Optimistic Optimization of a Deterministic Function without the Knowledge of its Smoothness	2011
StoSOO	Stochastic Simultaneous Optimistic Optimization	2013
HCT	Online Stochastic Optimization Under Correlated Bandit Feedback	2014
POO*	Black-box optimization of noisy functions with unknown smoothness	2015
GPO*	General Parallel Optimization Without A Metric	2019
PCT	General Parallel Optimization Without A Metric	2019
SequOOL	A Simple Parameter-free And Adaptive Approach to Optimization Under A Minimal Local Smoothness Assumption	2019
StroquOOL	A Simple Parameter-free And Adaptive Approach to Optimization Under A Minimal Local Smoothness Assumption	2019
VROOM	Derivative-Free & Order-Robust Optimisation	2020
VHCT	Optimum-statistical Collaboration Towards General and Efficient Black-box Optimization	2023
VPCT	N.A. (GPO + VHCT)	N.A.

Hierarchical partition

Partition	Description
BinaryPartition	Equal-size binary partition of the parameter space, the split dimension is chosen uniform randomly
RandomBinaryPartition	Random-size binary partition of the parameter space, the split dimension is chosen uniform randomly
DimensionBinaryPartition	Equal-size partition of the space with a binary split on each dimension, the number of children of one node is 2^d
KaryPartition	Equal-size K-ary partition of the parameter space, the split dimension is chosen uniform randomly
RandomKaryPartition	Random-size K-ary partition of the parameter space, the split dimension is chosen uniform randomly

Synthetic objectives

• Some of these objectives can be found on Wikipedia

Objectives	Image
Garland
DoubleSine
DifficultFunc
Ackley
Himmelblau
Rastrigin

Contributing

We appreciate all forms of help and contributions, including but not limited to

• Star and watch our project
• Open an issue for any bugs you find or features you want to add to our library
• Fork our project and submit a pull request with your valuable codes

Please read the contributing instructions before submitting a pull request.

Citations

If you use our package in your research or projects, we kindly ask you to cite our work

@misc{Li2023PyXAB,
  doi = {10.48550/ARXIV.2303.04030},
  url = {https://arxiv.org/abs/2303.04030},
  author = {Li, Wenjie and Li, Haoze and Honorio, Jean and Song, Qifan},
  title = {PyXAB -- A Python Library for $\mathcal{X}$-Armed Bandit and Online Blackbox Optimization Algorithms},
  publisher = {arXiv},
  year = {2023},
}

We would also appreciate it if you could cite our related works.

@article{li2023optimumstatistical,
    title={Optimum-statistical Collaboration Towards General and Efficient Black-box Optimization},
    author={Wenjie Li and Chi-Hua Wang and Guang Cheng and Qifan Song},
    journal={Transactions on Machine Learning Research},
    issn={2835-8856},
    year={2023},
    url={https://openreview.net/forum?id=ClIcmwdlxn},
    note={}
}

@misc{li2022Federated,
  doi = {10.48550/ARXIV.2205.15268},
  url = {https://arxiv.org/abs/2205.15268}, 
  author = {Li, Wenjie and Song, Qifan and Honorio, Jean and Lin, Guang},
  title = {Federated X-Armed Bandit},
  publisher = {arXiv},
  year = {2022},
}