symbolic-data 0.2.0

Model-agnostic symbolic-regression data layer: skeleton/expression sampling, priors, holdout, datasets.

symbolic_data

The model-agnostic symbolic-regression data layer, carved out of flash-ansr: skeleton/expression sampling, priors, (X, y) support sampling, holdout management, and dataset construction.

Both symbolic-regression methods (for training holdout) and the srbf eval framework depend on it, so training, holdout, and evaluation draw from one source of truth. Depends only on simplipy + numpy/sklearn.

Install

pip install symbolic-data

Quick start

import symbolic_data

# 1. Sample (X, y) problems from a skeleton pool (the model-agnostic seam)
pool = symbolic_data.SkeletonPool.from_config("skeleton_pool.yaml")
pool.create(100)
for sample in symbolic_data.iter_samples(pool, n_support=32, noise_level=0.01, seed=0):
    sample.x_support, sample.y_support, sample.expression  # ready to fit / tokenize

# 2. Load a benchmark (spec fetched + cached from the psaegert/ansr-data HF dataset)
fastsrb = symbolic_data.load_benchmark("fastsrb")
dataset = fastsrb.sample("II.38.3", n_points=100)

Extensibility

Distributions and benchmarks are pluggable via registries: in-process with @symbolic_data.DISTRIBUTIONS.register("name") / @symbolic_data.BENCHMARKS.register("name"), or across packages via importlib.metadata entry points (groups symbolic_data.distributions, symbolic_data.benchmarks). A registered name drops into the same config slot as a builtin.

Versioning / reproducibility

v1 guarantees leak-safety (a seeded, shipped holdout grid + a robust symbolic/numeric matcher), not cross-consumer byte-identical regeneration (the rng-Generator threading is a separate, later phase). Benchmark specs are HF-versioned and stamped on .provenance.

Status: v0.1.0. Registry, iter_samples seam, and load_benchmark (FastSRB) are in. Deferred: the MIA matched-control audit, curated benchmark loaders (Feynman/Nguyen), the canonical-v1 holdout grid mint, and cross-consumer byte-identical sampling.