Recipes
Short, task-focused how-tos that show OptunaSharp features in context. Where the Guides
explain a subsystem end to end, each recipe answers one practical question — "how do I persist a
study?", "how do I write my own sampler?", "how do I feed in hyperparameters I already know are
good?" — and is a direct port of the matching page in Optuna's
Recipes tutorial
(optuna/tutorial/20_recipes/), rewritten against the C# API.
The objectives here are deliberately small (a quadratic, a mock score) so the mechanism — not a machine-learning library — is what each page teaches. Swap in your real training code wherever a recipe returns a placeholder value.
The recipes
| # | Recipe | You'll learn to |
|---|---|---|
| 1 | Saving/Resuming a Study with an RDB backend | Persist a study to SQLite and resume it later |
| 2 | Multi-objective optimization | Optimize several objectives and read the Pareto front |
| 3 | User attributes | Annotate studies and trials with your own metadata |
| 4 | Command-line interface | Drive an ask/tell loop from the optunasharp tool |
| 5 | User-defined sampler | Implement your own sampling algorithm |
| 6 | User-defined pruner | Implement your own early-stopping rule |
| 7 | Callback for Study.Optimize |
Run code after every trial (and stop the study) |
| 8 | Specify hyperparameters manually | Enqueue known-good params or inject already-evaluated ones |
| 9 | Ask-and-tell interface | Own the optimization loop; define-and-run; batch evaluation |
| 10 | Re-use the best trial | Re-evaluate the best hyperparameters after tuning |
| 11 | Journal-file storage | Persist and share a study through a single log file |
| 12 | Artifacts | Attach large files (models, images) to trials |
| 13 | Wilcoxon pruner | Early-stop objectives that average many evaluations |
| 14 | Ablation study with brute force | Exhaustively sweep a conditional search space |
How these differ from Guides
A Guide is the reference for a whole area — Storage documents every backend; Samplers documents every sampler. A recipe is a walkthrough of one scenario and links back to the Guide for the full surface. If you want the complete API, follow the Guide link at the bottom of each recipe.
Relationship to Optuna
Each recipe corresponds one-to-one with a file in Optuna's
tutorial/20_recipes/. The C#
port keeps the same narrative and demonstrates the same feature; under a fixed seed the trial
sequence is bit-for-bit identical to Python wherever the path is deterministic.
See also
Study and Trials · Samplers · Pruners · Storage · Artifacts · the API reference.