gloss-opt 0.1.0

GLOSS: Global-Local-Unexplored Sampling Strategy for batch surrogate optimization in vast chemical search spaces

GLOSS

Global–Local–Unexplored Sampling Strategy — a multi-strategy batch recommender for surrogate-based optimization in vast chemical search spaces.

Overview

GLOSS decomposes each $q$-point batch across three complementary streams that share a single surrogate model:

• Global — UCB-driven exploitation
• Local — BallTree refinement around the current best (with an $\mathcal{O}(K)$ top-$K$ truncation that keeps it practical on $n=10^5$–$10^6$ candidate pools)
• Unexplored — maximizes geometric distance to observed points; uses no surrogate signal, providing robustness against an overfit surrogate

Install

git clone https://github.com/zbc0315/gloss.git
cd gloss
pip install -e .

Python 3.9+ required. Dependencies are pinned in requirements.txt.

Quick start

from gloss import GLOSS

g = GLOSS(
    space={"candidates": candidates},
    direction="maximize",
    ratio={"global_best": 4, "local_best": 2, "unexplored": 2},
)
batch = g.recommend(X_obs, y_obs, n_points=8)

Reproducing the benchmark

python -m benchmarks.bench_main --study all

Benchmark covers Buchwald–Hartwig ($n=3{,}955$), QM9 HOMO–LUMO gap ($n=100{,}000$), and an Arrhenius-2D virtual surface, comparing GLOSS against UCB-BO, BO(EI), GA, and Random across 5 seeds.

License

MIT.