How emdgroup/baybe works
BayBE is a Bayesian design-of-experiments library: users define parameters, a constrained search space, and an objective, then submit measurements and request candidate configurations in a repeatable campaign loop. Campaign orchestration connects these domain objects to recommendation strategies; Bayesian strategies use surrogate models and acquisition functions. The README also documents bandit optimization, multi-target objectives, transfer learning, serialization, insights, and simulation/backtesting. Edges reflect the README and SOURCE INDEX references; unsampled feature implementations are represented at their supported directory paths, without asserting unverified internal calls.
Main entry points: campaign.py Campaign · botorch Bayesian recommenders · core.py Search space · sequential.py Sequential strategy
Components 20
Campaign workflow
- Campaign
baybe/campaign.py - Measured outcomes
Problem definition
- Parameters
baybe/parameters/base.py - Search space
baybe/searchspace/core.py - Constraints
baybe/constraints/base.py - Targets
baybe/targets/numerical.py - Objectives
baybe/objectives/base.py - Target transforms
baybe/transformations/basic.py
Recommendation engine
- Recommenders
baybe/recommenders/base.py - Sequential strategy
baybe/recommenders/meta/sequential.py - Bandit and naive
baybe/recommenders/naive.py - Bayesian recommenders
baybe/recommenders/pure/bayesian/botorch - Acquisition functions
baybe/acquisition/acqfs.py - Surrogates
baybe/surrogates/base.py - Gaussian processes
baybe/surrogates/gaussian_process/core.py
Evaluation and operations
- Simulation
baybe/simulation/core.py - Transfer learning
baybe/simulation/transfer_learning.py - Campaign insights
baybe/insights/shap.py - Serialization
baybe/serialization/mixin.py
Other components
- Experimenter
Connections 21
- Experimenter → Campaign: configures and requests
- Campaign → Search space: uses
baybe/campaign.py - Campaign → Objectives: uses
baybe/campaign.py - Campaign → Recommenders: dispatches recommendations
baybe/campaign.py - Parameters → Search space: builds subspaces
baybe/parameters/base.py - Search space → Constraints: validates constraints
baybe/searchspace/core.py - Objectives → Targets: models targets
baybe/objectives/base.py - Targets → Target transforms: transforms values
baybe/targets/numerical.py - Campaign → Measured outcomes: accepts measurements
- Measured outcomes → Campaign: informs next cycle
- Sequential strategy → Bayesian recommenders: selects strategy
baybe/recommenders/meta/sequential.py - Sequential strategy → Bandit and naive: selects strategy
baybe/recommenders/meta/sequential.py - Bayesian recommenders → Surrogates: fits models
- Bayesian recommenders → Acquisition functions: optimizes acquisition
baybe/acquisition/acqfs.py - Gaussian processes → Surrogates: implements model
baybe/surrogates/gaussian_process/core.py - Experimenter → Simulation: benchmarks settings
- Simulation → Campaign: runs campaign
baybe/simulation/scenarios.py - Experimenter → Transfer learning: supplies prior data
- Experimenter → Campaign insights: analyzes campaign
- Campaign → Serialization: supports persistence
baybe/campaign.py - Experimenter → Serialization: stores and restores