Probabilistic Reparametrization

About

An interface to Bayesian optimization with Probabilistic Reparametrization (ProbRep), by [Daulton et al., 2022]. This implementation uses a fork of the official implementation provided by the authors.

How to run

Warning

This solver runs in a different conda environment than base.

You can find a conda environment where this solver can run here.

If you have cloned poli-baselines locally:

conda env create --file src/poli_baselines/solvers/bayesian_optimization/pr/environment.pr.yml
conda activate poli__pr

import numpy as np

from poli import objective_factory
from poli_baselines.solvers.bayesian_optimization.pr import (
    ProbabilisticReparametrizationSolver,
)

# Define the alphabet and sequence length.
# (depending on the problem, you may find them in black_box.info)
alphabet: list[str] = load_alphabet()
sequence_length: int = load_sequence_length()

problem = objective_factory.create(
    name="rdkit_qed", string_representation="SELFIES"
)
black_box, x0 = problem.black_box, problem.x0
y0 = black_box(x0)

# Add padding to complete the original input to the sequence length
x0_ = np.array([["[nop]"] * sequence_length])
x0_[0, : x0.shape[1]] = x0

solver = ProbabilisticReparametrizationSolver(
    black_box=black_box,
    x0=x0_,
    y0=y0,
    alphabet=alphabet,
    sequence_length=sequence_length,
)

solver.solve(max_iter=10)

References

If you use this solver, we expect that you cite the following resources:

References as text

[1] Daulton, S., Wan, X., Eriksson, D., Balandat, M., Osborne, M. A., & Bakshy, E. (2022). Bayesian optimization over discrete and mixed spaces via probabilistic reparameterization. In Advances in Neural Information Processing Systems 35.

[2] González-Duque, M., Bartels, S., & Michael, R. (2024). poli: a libary of discrete sequence objectives [Computer software]. MachineLearningLifeScience/poli

References as BibTeX

@inproceedings{Daulton:ProbRep:2022,
      title={Bayesian Optimization over Discrete and Mixed Spaces via Probabilistic Reparameterization}, 
      author={Samuel Daulton and Xingchen Wan and and David Eriksson and Maximilian Balandat and Michael A. Osborne and Eytan Bakshy},
      booktitle={Advances in Neural Information Processing Systems 35},
      year={2022}
}

@software{Gonzalez-Duque:poli:2024,
author = {González-Duque, Miguel and Bartels, Simon and Michael, Richard},
month = jan,
title = {{poli: a libary of discrete sequence objectives}},
url = {https://github.com/MachineLearningLifeScience/poli},
version = {0.0.1},
year = {2024}
}