fit-and-diagnose
Fit ordinary PySR regressors, inspect Pareto fronts, select equations, and diagnose slow or poor searches.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- fit-and-diagnose
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Fit ordinary PySR regressors, inspect Pareto fronts, select equations, and diagnose slow or poor searches.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md8.0 KB
- 参考文档references/troubleshooting.md7.7 KB
- 参考文档references/workflows.md8.4 KB
- 脚本scripts/pysr_quickstart_smoke.py4.8 KB
03 SKILL.md 正文
PySR fit-and-diagnose
Use this sub-skill when the task is an ordinary PySR symbolic-regression fit: prepare tabular data, choose a small operator set, run PySRRegressor.fit, inspect model.equations_, pick equations from the Pareto front, call predict/sympy/latex, or debug a slow or low-quality search.
Route first
| User need | Use here? | Route |
|---|---|---|
Plain PySRRegressor fit, data shape, feature names, weights, denoising, feature selection, Pareto selection |
Yes | Continue below. |
| Custom Julia operators, custom elementwise/full losses, units, operator constraints, required/forbidden structure by loss penalties | No | ../customization-and-constraints/SKILL.md |
| Known expression templates, parametric expressions, shared/vector-valued templates, guesses for templates | No | ../structured-expressions/SKILL.md |
| Saved hall-of-fame files, reload/checkpoints, SymPy/LaTeX/JAX/Torch export details, TensorBoard artifacts | No | ../export-and-artifacts/SKILL.md |
| Installation, first Julia startup, threading, multiprocessing, Slurm, long-run scaling, reproducibility environment | Usually no | ../runtime-and-scaling/SKILL.md |
Minimal operating loop
- Validate data shape. Use
Xas(n_samples, n_features)andyas(n_samples,)for one target or(n_samples, n_outputs)for independent multi-output fits. Prefer a pandas DataFrame when feature names matter; otherwise pass safevariable_namestofit. - Start with a small operator set. Use only operators that are plausible for the domain. For a polynomial-like first pass, start with
binary_operators=["+", "-", "*"]; add/, trig,exp, or other operators only when justified. - Bound the first run. Start short to validate the setup, then run longer. Use
niterations,timeout_in_seconds, ormax_evalsfor budget control. Avoid interpreting the first Julia compile as a search failure. - Fit and inspect the whole front.
model.fit(X, y, weights=..., variable_names=...)stores a Pareto front inmodel.equations_. Print it and inspect loss versus complexity rather than reporting only the auto-selected row. - Select explicitly when needed.
model_selection="best"balances loss and simplicity;"accuracy"chooses minimum loss;"score"chooses the largest score. Passindex=topredict,sympy, orlatexto evaluate a specific row. - Iterate scientifically. If results are trivial or slow, simplify features/operators, adjust
maxsize, useselect_k_featuresfor high-dimensional tabular data, considerdenoise=Trueor weights for noisy data, and only then add more compute.
See references/workflows.md for recipes, references/api-reference.md for focused signatures and option semantics, and references/troubleshooting.md for failure modes.
Safe bundled smoke helper
scripts/pysr_quickstart_smoke.py is a small, deterministic helper for agents. It is help/dry-run oriented by default and imports PySR only when --run-fit is supplied.
python scripts/pysr_quickstart_smoke.py --help
python scripts/pysr_quickstart_smoke.py --run-fit --niterations 1 --timeout-seconds 20
The helper is a smoke check, not a benchmark. A successful run proves that import, a tiny fit, equations_, predict, sympy, and latex are usable in the current environment; it does not prove convergence on the user's dataset.
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).