pymoo
Use pymoo for single-, multi-, and many-objective optimization, custom problems, evolutionary operators, performance tuning, and Pareto analysis.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- pymoo
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use pymoo for single-, multi-, and many-objective optimization, custom problems, evolutionary operators, performance tuning, and Pareto analysis.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-entrypoints.md3.7 KB
- 参考文档references/repo-provenance.md3.1 KB
- 参考文档references/repo-routing-metadata.json332 B
- 参考文档references/troubleshooting.md4.0 KB
- 脚本scripts/pymoo_quickstart_smoke.py2.0 KB
03 SKILL.md 正文
pymoo
Use this repo skill when a task involves pymoo, evolutionary computation, single-objective optimization, multi-objective or many-objective optimization, Pareto fronts, custom optimization problems, genetic operators, reference directions, constraint handling, performance indicators, or pymoo visualization.
pymoo is a Python package for optimization workflows. It minimizes all
objectives, represents inequality constraints as G <= 0, and centers common
usage around a Problem, an Algorithm, a termination criterion, and
pymoo.optimize.minimize.
Quick install and smoke
Public install path:
python -m pip install -U pymoo
Minimal import/runtime check:
python - <<'PY'
from pymoo.algorithms.moo.nsga2 import NSGA2
from pymoo.optimize import minimize
from pymoo.problems import get_problem
res = minimize(get_problem("zdt1"), NSGA2(pop_size=20), ("n_gen", 3), seed=1, verbose=False)
print(res.F.shape, res.algorithm.evaluator.n_eval)
PY
For a bundled self-check after installing pymoo, run scripts/pymoo_quickstart_smoke.py.
Route by task
| User task signal | Read |
|---|---|
Run NSGA-II/NSGA-III/GA/DE/PSO/MOEA/D, choose an algorithm, set termination, use minimize, inspect Result, callbacks, history, ask-and-tell, external evaluation loop |
optimization-workflows |
Define a custom problem, choose Problem vs ElementwiseProblem vs FunctionalProblem, use get_problem, fix F/G/H shapes, convert maximization or constraints, handle bounds/Pareto fronts |
problem-modeling |
| Configure real/integer/binary/choice variables, mixed-variable search, sampling/crossover/mutation/repair, duplicate elimination, initial populations, custom object/permutation operators, hyperparameter helper APIs | operators-and-variables |
| Speed up slow evaluations, vectorize objectives, add thread/process starmap, use optional joblib/dask/ray, check Cython compiled extensions, reason about optional GPU/backend paths, control history/progress overhead | performance-and-parallelization |
| Compute hypervolume, GD/IGD/epsilon/KKTPM/R-metric, generate reference directions, use decomposition or MCDM, analyze convergence/history, save headless Pareto plots | analysis-and-visualization |
Operating rules that prevent common pymoo mistakes
- Minimization only: negate objectives that are originally maximized.
- Constraint sign: inequality constraints are feasible when
out["G"] <= 0; equality residuals belong inout["H"]withn_eq_constr. - Validate problem shapes first: check tiny
problem.evaluate(...)samples before tuning algorithms. - Use explicit budgets: pass termination tuples such as
("n_gen", 50)or("n_evals", 5000)for reproducible scripts. - Seed stochastic runs: pass
seed=...and control randomness inside custom objective code, especially under parallel evaluation. - Match algorithm to search space: mixed/discrete/permutation variables need matching variables/operators or repair, not just a float-coded algorithm.
- Keep optional extras explicit: base pymoo includes matplotlib and core numerical dependencies; joblib/dask/ray, Optuna, COMO-CMA-ES, pyrecorder, and GPU frameworks are optional task-specific additions.
- Use headless plotting in agents: set a non-interactive Matplotlib backend
and save figures instead of relying on GUI
show().
Shared references
- API entrypoints: cross-sub-skill import map, version/extras notes, and first checks.
- Troubleshooting: install/import, compiled extension, optional dependency, and cross-workflow failure routing.
- Repo provenance: source version and evidence baseline for refresh decisions.
- Repo routing metadata: structured router metadata consumed during managed import.
Bundled self-check scripts
- scripts/pymoo_quickstart_smoke.py: base
import, compiled-extension status, and tiny
NSGA2/ZDT1run. - Sub-skill scripts cover custom-problem shape validation, mixed-variable search, parallel elementwise evaluation, indicator checks, compiled-extension checks, and headless scatter plotting.
Avoid this skill when
- The task is generic mathematical optimization with SciPy and does not use pymoo APIs, evolutionary operators, Pareto fronts, or pymoo problem classes.
- The task is hyperparameter optimization with Optuna/Ray Tune alone and no pymoo algorithm/problem integration.
- The task requires training deep learning models or GPU tensor code unrelated to pymoo's optional vectorized objective pattern.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).