pysr
Use PySR for symbolic regression, interpretable equation discovery, equation export, runtime scaling, and PySR troubleshooting.
SKILL.mdskills/repositories/repo-skills/pysr/SKILL.md
GitHubPySR repo skill
Use this skill when a task involves PySR, SymbolicRegression.jl from Python, symbolic regression, interpretable equation discovery from numeric data, scaling-law or formula search, or troubleshooting PySR installation/search/export behavior.
PySR exposes a scikit-learn-style PySRRegressor front end backed by Julia's SymbolicRegression.jl. It returns a Pareto front of equations that trade loss against complexity. Treat PySR as a stochastic search tool: validate data and operators first, run bounded smoke fits, inspect the full Pareto front, then run longer searches only after the setup is sensible.
Install and import check
For ordinary use:
pip install pysr
# or
conda install -c conda-forge pysr
Minimal check:
import pysr
from pysr import PySRRegressor
print(pysr.__version__)
A fresh import pysr can download/resolve Julia packages and the first fit can compile Julia code. Do not misdiagnose that first-run setup cost as a failed search. For runtime/startup details, read sub-skills/runtime-and-scaling/SKILL.md.
Route map
| Task | Read |
|---|---|
Fit ordinary tabular data, choose operators, inspect model.equations_, select equations, debug trivial/slow searches |
sub-skills/fit-and-diagnose/SKILL.md |
| Define custom Julia operators, losses, full objectives, units, constraints, complexity rules, mutations, or plugins | sub-skills/customization-and-constraints/SKILL.md |
Use known expression skeletons, TemplateExpressionSpec, category parameters, shared/vector-valued templates, differential operators, or guesses |
sub-skills/structured-expressions/SKILL.md |
| Export equations to SymPy/LaTeX/NumPy/JAX/PyTorch, inspect hall-of-fame CSVs, reload checkpoints, or configure TensorBoard logging | sub-skills/export-and-artifacts/SKILL.md |
| Install, import, manage JuliaCall, choose threads/processes/Slurm, bound long runs, use CLI tests, or diagnose startup/runtime failures | sub-skills/runtime-and-scaling/SKILL.md |
Fast operating pattern
- Start with data and objective. Confirm
Xhas shape(n_samples, n_features)andyis one-dimensional for one target or two-dimensional for multi-output. Decide whether weights, feature names, units, or denoising are needed. - Choose a small search space. Keep operators minimal. Every unnecessary operator expands the evolutionary search; add custom operators/losses only when domain evidence justifies them.
- Bound first runs. Use low
niterations,timeout_in_seconds,max_evals, andinput_stream="devnull"in scripts/noninteractive contexts. - Inspect the Pareto front.
model.equations_is the primary result. Compare loss and complexity, and passindex=topredict,sympy, orlatexwhen you want a specific row rather than the default selection. - Escalate deliberately. For known structure use templates; for domain rules use finite penalties, constraints, units, or custom objectives; for large jobs use batching or Slurm after the simple case works.
- Persist durable artifacts. Save the construction code and hall-of-fame CSV. Pickle checkpoints are useful but version-sensitive.
Shared references and scripts
references/api-index.mdmaps common public APIs to the owning sub-skill.references/troubleshooting.mdcovers cross-cutting failures and points to the nearest owner.references/repo-provenance.mdrecords the source snapshot used to create this skill.references/repo-routing-metadata.jsoncontains structured routing metadata for managed repo-skill import.scripts/check_pysr_environment.pyperforms a safe package/CLI/API probe without fitting a model by default.
Boundaries
This skill is for using PySR as a package. It does not replace SymbolicRegression.jl's Julia-native documentation for deep backend development, and it does not claim optional JAX, PyTorch, TensorBoard, Slurm, or autodiff backends are installed unless the current environment is checked.