Get started¶
Choose the shortest route that matches what you want to accomplish. Installation, first-run code, result interpretation, custom-problem guidance, and durable-study guidance each have one canonical page; this page only routes between them.
Try VAMOS¶
- Install VAMOS from PyPI in an isolated Python environment.
- Follow T01. Your first meaningful result to solve a built-in benchmark and interpret its actual output. The Quickstart retains the complete run-to-study sequence.
- Read Understanding optimization results to connect decision rows (
X) with objective rows (F) and distinguish a returned set from its non-dominated subset. - Use the API reference when you need exact signatures rather than tutorial guidance.
If Python itself is still new to you, begin with the Minimal Python Track. If multi-objective optimization is new to you, the Foundations path explains Pareto dominance, quality indicators and NSGA-II with interactive figures.
Solve your own problem¶
Follow T02 to check a constrained model, then T03 to select a feasible alternative using explicit preferences.
Use Solve your own problem to define objectives, bounds, encodings, vectorized evaluation, and constraints with make_problem(...).
Use the Experimental vamos create-problem command when you prefer a generated file scaffold. Adding a reusable built-in problem to VAMOS is a contributor workflow and is documented separately in Adding a problem.
VAMOS 1.2.0: vamos create-problem refuses to overwrite an existing file unless you pass --force, and rejects a --n-var, --n-obj or --budget below 1. A default file name that would hide an installed or standard module, such as numpy.py, gets a _problem suffix. Any problem name produces a valid script, and the script prints result.front(), the non-dominated solutions. In VAMOS 1.1.0, the command overwrote the file and accepted any dimension. A quote in the name broke the script, and the script labelled the whole final population as non-dominated.
Run a reproducible study¶
Follow T06: plan and run, T07: measure quality, then T08: build a report. The shared setup includes standalone scripts and a recorded experiment whose metrics can be recomputed without optimization.
Use Durable studies when you need a persistent problem-by-algorithm-by-seed matrix with explicit task state, resume, retry, inspection, and summaries.
Use Run artifacts & replay for individual persisted runs and for the distinction between loading, verification, and executable replay.
Choose an interface¶
| Need | Recommended surface |
|---|---|
| A Python script or notebook | vamos.optimize(...) |
Interpret X, F, and a non-dominated subset |
OptimizationResult + Understanding results |
| A plain Python objective function | vamos.make_problem(...) + vamos.optimize(...) |
| A stable command-line workflow | The main runner, run-artifact commands, and study commands in CLI & Config |
| An interactive onboarding wizard | Experimental vamos quickstart |
| Multiple seeds in one small call | optimize(..., seed=[...]) returning StudyResult |
| A persistent experiment matrix | StudySpec, plan_study, create_study, and Study.run() |
| Exact parameters for a built-in algorithm | Public configuration objects from vamos.algorithms |
VAMOS 1.0 uses NumPy as its deterministic reference backend. Reproducibility is a same-environment promise, not a cross-platform or cross-backend bitwise guarantee. See Stability and versioning and Known limitations for the exact supported surface.
Continue from here¶
- Tutorials — follow a learning path or study an algorithm through a worked example.
- How-to Guides — find instructions for a specific operation.
- Reference — locate exact interfaces, capabilities, and contracts.
- Examples — choose maintained scripts, notebooks, or task-oriented guides.
- Troubleshooting — installation, dependency, configuration, and runtime issues.
- Algorithms & Backends — algorithm-specific parameters and backend notes.
- Analysis & Visualization — Experimental helpers to inspect and visualize optimization results.