PopulationBasedTrainingOptimizer<T, TInput, TOutput>

Models & Types · AiDotNet.HyperparameterOptimization

Implements Population-based Training (PBT) for hyperparameter optimization.

Inheritance: HyperparameterOptimizerBase<T, TInput, TOutput> → PopulationBasedTrainingOptimizer<T, TInput, TOutput>

How It Works

For Beginners: Population-based Training is an evolutionary approach that:

  • Trains a population of models simultaneously
  • Periodically checks each model’s performance
  • Poor performers “exploit” better performers (copy their weights/hyperparameters)
  • Copies are then “explored” (slightly mutated hyperparameters)
  • This creates an online hyperparameter schedule adapted during training

Key concepts:

  • Population: Multiple models training in parallel
  • Exploit: Copy weights and hyperparameters from a better performer
  • Explore: Perturb hyperparameters to discover better values
  • Ready: A model is “ready” to exploit/explore after training for some steps

PBT can discover hyperparameter schedules that vary during training, something that grid/random/Bayesian search cannot do.

Constructors

Constructor Summary
PopulationBasedTrainingOptimizer(bool maximize, int populationSize, int readyInterval, double exploitFraction, double perturbFactor, ExploitStrategy exploitStrategy, ExploreStrategy exploreStrategy, int? seed) Initializes a new instance of the PopulationBasedTrainingOptimizer class.

Methods

Method Summary
Dictionary<string, object> SuggestNext(HyperparameterTrial<T> trial) Suggests the next hyperparameter configuration to try.
HyperparameterOptimizationResult<T> Optimize(Func<Dictionary<string, object>, T> objectiveFunction, HyperparameterSearchSpace searchSpace, int nTrials) Runs Population-based Training optimization.
List<PopulationMemberInfo> GetPopulationState() Gets the current population state.
PopulationMemberInfo GetBestMember() Gets the best member of the population.