How the weekly sweeps were run
Ufinq is tested every week against PySR, Operon and gplearn on public benchmark datasets, with the rules fixed in advance and the failures published, not buried. The weekly sweep is our own comparison; it is not the SRBench protocol. The weekly sweeps compared Ufinq with PySR, gplearn and Operon under an own method, before the comparison moved to the SRBench 2025 protocol. This page describes that earlier method so the archived results can be read. The current method is under Methodology.
The setup
- 10 datasets from PMLB and the UCI repository, 10 seeds each.
- 1,800 s per fit.
- An own 75/25 split of each dataset into training and test rows, not SRBench's.
- CPU cores were not the same for every method, so the compute each method received differed.
Principles
- One declared configuration per algorithm, applied uniformly across every dataset and seed. No per-dataset tuning.
- One run per trial. Each algorithm trains on each dataset once per seed; quality is read from held-out predictions, never from re-training.
- Beyond R². Each trial also reports symbolic recovery, complexity, stability, extrapolation, and calibration, not accuracy alone.
- Negative results are recorded. Timeouts, non-finite predictions, and all-failed datasets are categorized and shown, never silently dropped.
- Calibration separates a real regression from an environment change: a fixed reference set runs every sweep and drift is checked against a per-segment baseline.
The numbers
Every sweep is a point-in-time snapshot with a stable, citable URL. To compare environments over time, see calibration & drift; to cite a result, see how to cite.
