advanced-domain-features
Use BayesianOptimization advanced domain features: constraints, typed parameters, custom parameter classes, TargetSpace conversions, and sequential domain reduction.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- advanced-domain-features
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use BayesianOptimization advanced domain features: constraints, typed parameters, custom parameter classes, TargetSpace conversions, and sequential domain reduction.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/constraints.md7.3 KB
- 参考文档references/domain-reduction.md7.0 KB
- 参考文档references/parameter-types.md9.5 KB
- 参考文档references/troubleshooting.md8.7 KB
- 脚本scripts/advanced_features_smoke.py11.0 KB
03 SKILL.md 正文
Advanced Domain Features
Use this sub-skill when a task involves constrained optimization, non-float
search variables, custom parameter encodings, low-level TargetSpace
conversion/debugging, or sequential domain reduction in the
bayesian-optimization package.
Route here for
- SciPy
NonlinearConstraintsetup withBayesianOptimization(..., constraint=...). ConstraintModelbehavior:eval,fit,predict,approx,allowed, multiple constraints, and registering known constrained observations.- Integer, categorical, float, or preconstructed
BayesParameterentries inpbounds. - Custom
BayesParametersubclasses and kernel-space transforms. TargetSpace.params_to_array,array_to_params, masks, bounds, and typed conversion semantics.SequentialDomainReductionTransformersetup, validation, and limitations.
Route elsewhere
- Basic optimizer lifecycle,
maximize, ask-tell loops, state save/load,predict, and ordinary HPO recipes:../optimizer-workflows/SKILL.md - Acquisition selection, exploration/exploitation tuning, custom acquisition
functions, Constant Liar, or GPHedge:
../acquisition-control/SKILL.md - Editing the repository, running native test selections, docs, lint, CI, or
release-maintenance checks:
../repo-maintenance/SKILL.md
Start with the relevant reference
- Read
references/constraints.mdwhen the user needs expensive/learned constraints, multi-output constraints, feasibility debugging, or manual registration of constrained observations. - Read
references/parameter-types.mdwhenpboundsincludes integers, categories, preconstructed parameters, or custom domain encodings. - Read
references/domain-reduction.mdwhen bounds should shrink during optimization or a bounds transformer fails. - Read
references/troubleshooting.mdwhen errors mention mismatched constraint arguments, no allowed points, unsupported constrained acquisitions, invalid categorical categories, key/dimension mismatches, experimental non-float warnings, or domain reduction with typed parameters.
Safe smoke check
Run scripts/advanced_features_smoke.py
only after the package and its normal runtime dependencies are importable:
python scripts/advanced_features_smoke.py --check all
The helper is deterministic, uses no network and no plots, and validates tiny
constraint, typed-parameter, and domain-reduction behavior. Use
python scripts/advanced_features_smoke.py --help to list narrower checks.
Working rules for future agents
- Treat constraints and typed parameters as part of the domain definition, not
as after-the-fact filters. The optimizer wraps SciPy constraints in a
ConstraintModel, and the internal GP kernels use theTargetSpacetransform for typed domains. - Keep objective and constraint keyword names identical.
TargetSpacecalls both functions with the samearray_to_params(...)result. - For known observations in a constrained optimizer, always register the raw
constraint value with
constraint_value; otherwise the target and constraint histories cannot stay aligned. - Expect a warning for non-float parameters. Integer/categorical optimization is useful, but still marked experimental by the package.
- Do not combine sequential domain reduction with integer, categorical, or
custom non-
FloatParameterdomains in the current implementation. - If a task asks both for advanced domains and acquisition strategy design, handle the domain facts here, then route acquisition-specific choices to the acquisition-control sub-skill.
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).