Minimal Python Track¶
If you can run a command and copy-paste, this track is for you. Goal: run an experiment and get results in minutes, without learning all the jargon up front.
Status as of March 31, 2026: the quickstart wizard and the base single-run CLI are smoke-tested for the standard NSGA-II/ZDT1 path. If you want the absolute shortest path to a first script, the Python API in
docs/guide/getting-started.mdis still the lightest option.
1. Install¶
Create a virtual environment and install VAMOS:
python -m venv .venv
source .venv/bin/activate # Windows: .\.venv\Scripts\Activate.ps1
pip install -e ".[analysis]"
If you do not need plots, you can install just the core:
pip install -e .
2. Run the guided wizard¶
The quickstart wizard asks a few questions, writes a config file, and runs one experiment:
vamos quickstart
Want to see domain-flavored templates?
vamos quickstart --template list
Run a template without prompts:
vamos quickstart --template physics_design --yes --no-plot
3. Find your results¶
Results are stored under:
results/quickstart/<PROBLEM>/.../seed_<N>/
Key files:
manifest.json: requested/resolved settings, actual seed, provenance, outcome, and hashesresult.npz: objective, decision, constraint, population, and archive arraysenvironment.json: bounded runtime environment detailspareto_front_*.png: plot output outside the canonical run directory (if enabled)
4. Re-run or change settings¶
Re-run the same config:
vamos --config results/quickstart/quickstart_YYYYMMDD_HHMMSS.json
Change budget or seed without editing files:
vamos --config results/quickstart/quickstart_YYYYMMDD_HHMMSS.json --max-evaluations 8000 --seed 1
5. Quick summary¶
List recent runs:
vamos summarize --results results/quickstart
Open the latest run folder:
vamos open-results --results results/quickstart --open
Glossary (plain language)¶
- Problem: the task you want to optimize (a dataset or a math function).
- Algorithm: the search method (default is NSGA-II).
- Objective: the quantity you want to minimize (often two or more).
- Pareto front: the best trade-offs found so far (no single solution is best at everything).
- Budget: how many evaluations to spend (more budget = longer run).
- Population size: how many candidate solutions are kept each step.
- Seed: fixes randomness so runs are repeatable.
- Engine: compute backend (numpy is the default).
If something fails¶
- Run:
vamos checkto verify your install. - For biology/chemistry templates, install scikit-learn:
pip install -e ".[examples]". - For plots, install the analysis extras:
pip install -e ".[analysis]".
Next steps¶
docs/guide/getting-started.mdfor the full API overview.docs/guide/cli.mdfor CLI details and config files.docs/guide/cookbook.mdfor copy-paste recipes.