Cookbook

Brush mostly consists of these components:

• A dataset class that handles multiple types of data

• A program class that defines composable operations over data

• A search space that samples the components of programs

• Tweaking operators to change programs randomly

• Optimizers to learn optimal splitting thresholds and optimal parameter values within programs.

• Working with Data
• The Search Space
  • Initializing
  • Inspecting
  • Initializing sampling probabilities
  • Loading datatypes
  • Automatically updating sampling probabilities with multi-armed bandits
  • Sampling
• Working with Programs
  • Types of Programs
  • Representation
  • Generation
  • Evaluation
  • Visualizing Programs
  • Serialization
    • String
    • Quick Little Tree
    • GraphViz
    • Tweaking Graphs
    • A classification example
• JSON Interoperability
• Saving and loading populations
  • Saving just the archive
  • Using population files with classification
  • Serialization with pickle
  • Stop/resume the fitting of an estimator
• The archive
  • Loading a specific model from archive
  • Visualizing the Pareto front of the archive
  • Acessing the entire population (unique individuals)
• Using Deap