Skip to content

v0.2.0

Latest

Choose a tag to compare

@github-actions github-actions released this 28 Oct 17:02
· 5 commits to main since this release

[0.2.0] - 2026-10-28

sceua now requires Python 3.13 or later. A run stops when neither the best value nor the population median improves for max_tolerant_iter iterations, or when every parameter's spread falls below x_tolerance of its bound range; x_tolerance is now that fraction (default 1e-4), and the tolerance argument is gone, so drop it from calls and rescale any absolute x_tolerance you set. minimize() honors max_evals, rejects budgets below the population size and x0 points outside the bounds, evaluates every moved point before it competes, and ranks NaN objective values below any number, with success=False when every evaluation returns NaN. Result is exported for type annotations.

Added

  • Add Result to the public API to be used for type checking on the user side. (e5f37db)

Changed

  • Name the parameter-convergence check (fb8804d)
  • Apply the new hooks' formatting (71a831c)
  • Require Python 3.13 and move tooling to dependency groups (breaking) (14760a4)
  • Satisfy ruff's full rule set in the solver (415da54)

Fixed

  • Remove tolerance, whose margin never changed a stop (breaking) (2463cad)
  • Keep progress statistics finite near the float limit (96d8f18)
  • Stop on population progress with scale-free tolerances (breaking) (e355afa)
  • Let finite candidates replace NaN values in complex evolution (78b3c0d)
  • Honor max_evals and reject budgets below the population size (0e07773)
  • Let a finite value replace a NaN best and flag all-NaN runs (a03bacb)
  • Skip evolution for single-point complexes (be7b4cb)
  • Store objective values as floats (f1623d6)
  • Stop the HYMOD test model from creating runoff without rain (bbc8175)
  • Detect lost dimensions in bound-scaled units (d5369e4)
  • Evaluate PCA-moved points and never move the best one (6230b0c)
  • List the best point once in the first complex (3d403a4)
  • Reject x0 points outside the bounds (a340e9d)
  • Cast x0 to float so offspring are not truncated (af2c9cc)
  • Skip PCA recovery for one-dimensional problems (f7fe8b3)