Practical overview
When it fits
Use BoTorch when you need a low-level, customizable framework for Bayesian optimization research or a non-standard optimization loop. For experiment configuration and orchestration with less direct algorithm development, the project recommends Ax.
Who it is for
- Researchers developing or customizing Bayesian optimization algorithms.
- Experienced practitioners who need control over models, acquisition functions or the optimization loop.
- PyTorch users applying Bayesian optimization to costly black-box objectives.
Documented uses
- Compose or modify Bayesian optimization algorithms from modular components.
- Create custom acquisition functions and optimizers.
- Use Gaussian-process models, including multi-task and deep-GP approaches.
- Combine differentiable models, including neural-network modules, in an optimization workflow.
Requirements & limits
- The tutorials assume familiarity with both Bayesian optimization and PyTorch.
- BoTorch exposes modular algorithm components; Ax provides a more user-oriented interface for experiment configuration and orchestration.
- BoTorch supports GPyTorch Gaussian processes and model-agnostic acquisition optimization when the model's posterior can be sampled.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Choose an approach for Bayesian optimization experiments
Select Ax or BoTorch according to whether you need experiment orchestration or direct control over Bayesian optimization details.
Before you start
- Familiarity with Bayesian optimization and PyTorch to use BoTorch tutorials.
- For introductory guidance, consult Ax documentation and tutorials if you are new to Bayesian optimization, and the PyTorch introductory tutorial if you are new to PyTorch.
Steps
For experiments using standard Bayesian optimization workflows, start with Ax for experiment configuration and orchestration.
For experiments using custom BoTorch models or acquisition functions, use Ax's Modular BoTorch Interface to integrate those objects with Ax orchestration.
If you need a non-standard setup or direct control over the optimization loop, consider working purely in BoTorch and follow the full optimization loop tutorials.
Expected outcome
A documented path is selected: Ax for end-user experiment orchestration, Ax's Modular BoTorch Interface for custom BoTorch components with orchestration, or BoTorch tutorials for full-loop control.
Guide limitations
- The guidance distinguishes approaches but does not provide a specific experiment implementation or execution procedure.
- BoTorch is presented as a low-level API for researchers and sophisticated practitioners; end-users are directed to Ax.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 12 October 2026
Related options
Ax — A higher-level adaptive experimentation platform recommended for users who want experiment configuration and orchestration rather than direct work with BoTorch’s lower-level compon