Supported Samplers
Tunny can utilize a number of sampling techniques.
The table below summarizes the optimizations supported by each sampler. Note that Tunny's UI will automatically display the available methods depending on the problem.
| Name | Single-Objective | Multi-Objective | Constraints | Human-in-the-loop |
|---|---|---|---|---|
| AUTO Sampler | ✓ | ✓ | ✓ | |
| TPE | ✓ | ✓ | ✓ | ✓ |
| cTPE | ✓ | ✓ | ✓ | |
| GP-Optuna | ✓ | ✓ | ||
| GP-BoTorch | ✓ | ✓ | ✓ | |
| GP-Preferential | ✓ | |||
| HEBO | ✓ | |||
| NSGA-II | ✓ | ✓ | ✓ | |
| NSGA-III | ✓ | ✓ | ✓ | |
| MOEA/D | ✓ | |||
| DE | ✓ | |||
| CMA-ES | ✓ | |||
| MO-CMA-ES | ✓ | |||
| Random | ✓ | ✓ | ||
| QMC | ✓ | ✓ | ||
| BruteForce | ✓ | ✓ |
The specific characteristics of each sampling technique are as follows
| Name | Note |
|---|---|
| AUTO Sampler | The appropriate sampler is automatically selected. |
| TPE | Along with NSGA-II, it is a highly versatile method. |
| cTPE | This is an enhanced version of TPE's constraint handling. |
| GP-Optuna | It operates faster than GP-BoTorch. |
| GP-BoTorch | Highly flexible, but slow operation. |
| GP-Preferential | Sampler designed exclusively for Human-in-the-loop. |
| HEBO | Even problems with strong nonlinearities and multimodalities converge very fast. |
| NSGA-II | It is a versatile method used by Wallacei and others. |
| NSGA-III | It is a multi-objective(more than 3), enhanced version of the NSGA-II |
| MOEA/D | It is a specialized method for multi-objective optimization |
| DE | Differential Evolution Algorithm. |
| CMA-ES | Converges fast when single objective. |
| MO-CMA-ES | Multi-objective version of CMA-ES |
| Random | Random sampling. |
| QMC | Samples low-discrepancy based on a sequence of numbers. |
| BruteForce | Explore all variable combinations |