User Guide

Backwards Compatibility and Deprecations

BayBE is in a constant state of development. As part of this, interfaces and objects might change in ways breaking existing code. We aspire to provide backwards support for deprecated code of the last three minor versions. After this time, old code will generally be removed. Both the moment of deprecation and full removal (deprecation expiration) will be noted in the changelog.

The most commonly used interface BayBE provides is the central Campaign object, which suggests new measurements and administers the current state of your experimental operation. The diagram below explains how the Campaign can be used to perform the bayesian optimization loop, how it can be configured and how the results can be post-analysed.

Detailed examples of how to use individual API components can be found below:

• Getting Recommendations
  • The recommend Call
  • Excluding Configurations
    • Permanent Exclusion
    • Dynamic Exclusion
• Campaigns
  • Creating a Campaign
    • Basic Creation
    • Creation From a JSON Config
  • Getting Recommendations
    • Basics
    • Candidate Control in Discrete Spaces
    • Caching of Recommendations
  • Adding Measurements
  • Predictive Statistics
  • Acquisition Function Values
  • Serialization
  • Further Information
• Active Learning
  • Local Uncertainty Reduction
  • Global Uncertainty Reduction
• Asynchronous Workflows
  • Marking Experiments as Pending
  • Adding Partial Results
• Constraints
  • Continuous Constraints
    • ContinuousLinearConstraint
    • ContinuousCardinalityConstraint
  • Conditions
    • ThresholdCondition
    • SubSelectionCondition
  • Discrete Constraints
    • DiscreteExcludeConstraint
    • DiscreteSumConstraint and DiscreteProductConstraint
    • DiscreteNoLabelDuplicatesConstraint
    • DiscreteLinkedParametersConstraint
    • DiscreteDependenciesConstraint
    • DiscretePermutationInvarianceConstraint
    • DiscreteCardinalityConstraint
    • DiscreteCustomConstraint
• Environment Vars
  • Basic Instructions
  • Polars
  • Disk Caching
  • EXPERIMENTAL: Floating Point Precision
  • Parallel Runs in Scenario Simulations
  • FPS Sampling Implementation
• Insights
  • Parameter Importance via SHAP
    • Basic Usage
    • Plots
    • Explainers
    • Experimental and Computational Representations
    • Multi-Output Surrogate Models
• Objectives
  • SingleTargetObjective
  • DesirabilityObjective
  • ParetoObjective
• Parameters
  • Continuous Parameters
    • NumericalContinuousParameter
  • Discrete Parameters
    • NumericalDiscreteParameter
    • CategoricalParameter
    • SubstanceParameter
    • CustomDiscreteParameter
    • TaskParameter
• Recommenders
  • General Information
  • Pure Recommenders
    • Bayesian Recommenders
    • Clustering Recommenders
    • Sampling Recommenders
  • Meta Recommenders
• Search Spaces
  • Discrete Subspaces
    • Building from the Product of Parameter Values
    • Constructing from a Dataframe
    • Creating a Simplex-Bound Discrete Subspace
    • Representation of Data within Discrete Subspaces
  • Continuous Subspaces
    • Using Explicit Bounds
    • Constructing from a Dataframe
  • Constructing Full Search Spaces
    • From the Default Constructor
    • Building from the Product of Parameter Values
    • Constructing from a Dataframe
  • Restricting Search Spaces Using Constraints
• Serialization
  • JSON (de-)serialization
  • Deserialization from configuration strings
    • Basic string assembly
    • Using default values
    • Automatic field conversion
    • The type field
    • Using abbreviations
    • Nesting objects
    • Invoking alternative constructors
    • Dataframe deserialization
• Simulation
  • Terminology: What do we mean by “Simulation”?
  • The Lookup Mechanism
    • Using a Callable
    • Using a Dataframe
    • Using None
  • Simulating a Single Experiment
  • Simulating Multiple Scenarios
  • Simulating Transfer Learning
• Surrogates
  • Available Models
  • Multi-Output Modeling
    • Surrogate Replication
    • Composite Surrogates
  • Extracting the Model for Advanced Study
  • Using Custom Models
• Targets
  • NumericalTarget
    • MIN and MAX mode
    • MATCH mode
  • Limitations
• Transfer Learning
  • Unlocking Data Treasures Through Transfer Learning
  • The Role of the TaskParameter
  • Seeing Transfer Learning in Action
• Utilities
  • Search Space Memory Estimation
  • Reproducibility
  • Adding Fake Target Measurements and Parameter Noise