bg-szy Bayesian Optimization (Self-Driving Lab)
The Bayesian Optimizer allows agents to efficiently explore a parameter space to maximize a target metric (yield, purity, binding affinity) with minimal experiments. It uses Gaussian Processes to model uncertainty and the Upper Confidence Bound (UCB) acquisition function.
When to Use This Skill
- When experiments are expensive or time-consuming.
- To autonomously tune hyperparameters for a machine learning model.
- To optimize reaction conditions (temperature, pH, concentration).
Core Capabilities
- Next Step Proposal: Suggests the next best experiment parameters.
- Surrogate Modeling: Predicts outcomes for untested parameters.
- Exploration/Exploitation: Balances trying new things vs. refining known good results.
Workflow
- Input: History of past experiments (params -> results) and bounds.
- Process: Fits a Gaussian Process to the data.
- Output: Returns the parameters for the next experiment.
Example Usage
User: "Given these past results, what temperature and pH should I try next?"
Agent Action:
python3 Skills/Mathematics/Probability_Statistics/bayesian_optimization.py \
--history "[[20, 7.0, 0.5], [25, 6.5, 0.6]]" \
--bounds "[[10, 40], [5, 9]]" \
--output next_experiment.json