Derivative-Free Optimization
Purpose
Provides optimization capabilities for problems where gradient information is unavailable or unreliable.
Capabilities
- Nelder-Mead simplex method
- Powell's method
- Surrogate-based optimization
- Bayesian optimization
- Pattern search methods
- Trust region methods
Usage Guidelines
- Method Selection: Choose based on problem characteristics
- Function Evaluations: Minimize expensive function calls
- Surrogate Models: Build and refine surrogate approximations
- Exploration-Exploitation: Balance search strategies
Tools/Libraries
- scipy.optimize
- Optuna
- GPyOpt