How donlan7524/esa-implementation works
ESA is a research implementation for minimizing expensive benchmark functions: an experiment run initializes evaluated samples, lets an agent choose an optimization strategy, evaluates proposed candidates, updates the sample history and agent, and records results. The main workflow is visible in experiment.py and documented in the README; strategies use surrogate models to guide proposals, while benchmark functions supply the objective. Alternative agents, strategies, and surrogate implementations are preserved as distinct capabilities. Some source files were not excerpted, so their internal behavior and connections are not inferred.
Main entry points: experiment.py Experiment runner · RBF.py Surrogate selector · main.py Main entry · database.py Sample database
Components 21
Experiment orchestration
- Experiment runner
experiment.py - Main entry
main.py
Agents
- Q-learning
code/Qagent.py - Thompson sampling
code/TSagent.py - UCB1 agent
code/UCBagent.py
Search strategies
- Strategy base
code/strategy/base_Strategy.py - Differential evolution
code/strategy/DE_strategy.py - Local search
code/strategy/SLS_strategy.py - Crossover search
code/strategy/crossover_strategy.py - Trust-region search
code/strategy/TRLS_Strategy.py - L-BFGS search
code/strategy/LBFGS_Strategy.py
Surrogate models
- Surrogate selector
code/RBF.py - RBF with HPO
code/RBF_hpo.py - Simple RBF
code/RBF_simple.py - Clustered KNN
code/KNN_surrogate.py - MLP surrogate
code/MLP_hpo.py - Gaussian process
code/GP_surrogate.py - K-means helper
code/kmeans.py
Data and objectives
- Sample database
code/database.py - Benchmark functions
code/brenchmarks/functions.py
Other components
- Experiment user
Connections 26
- Experiment user → Experiment runner: starts runs
README.md - Experiment user → Main entry: starts program
README.md - Experiment runner → Sample database: initializes samples
experiment.py - Experiment runner → Benchmark functions: selects objective
experiment.py - Experiment runner → Q-learning: creates agent
experiment.py - Experiment runner → Thompson sampling: creates agent
experiment.py - Experiment runner → UCB1 agent: creates agent
experiment.py - Experiment runner → Differential evolution: configures strategy
experiment.py - Experiment runner → Local search: configures strategy
experiment.py - Experiment runner → Crossover search: configures strategy
experiment.py - Experiment runner → Trust-region search: configures strategy
experiment.py - Experiment runner → L-BFGS search: configures strategy
experiment.py - Experiment runner → Sample database: updates samples
experiment.py - Main entry → Sample database: initializes samples
main.py - Main entry → Benchmark functions: selects objective
main.py - Differential evolution → Surrogate selector: fits surrogate
code/strategy/DE_strategy.py - Local search → Surrogate selector: fits surrogate
code/strategy/SLS_strategy.py - Crossover search → Surrogate selector: fits surrogate
code/strategy/crossover_strategy.py - Trust-region search → Surrogate selector: fits surrogate
code/strategy/TRLS_Strategy.py - L-BFGS search → Surrogate selector: fits surrogate
code/strategy/LBFGS_Strategy.py - Surrogate selector → RBF with HPO: selects implementation
code/RBF.py - Surrogate selector → Simple RBF: selects implementation
code/RBF.py - Surrogate selector → Clustered KNN: selects implementation
code/RBF.py - Surrogate selector → MLP surrogate: selects implementation
code/RBF.py - Surrogate selector → Gaussian process: selects implementation
code/RBF.py - Clustered KNN → K-means helper: uses clustering
code/KNN_surrogate.py