v0.26.0

0.26.0 (2026-04-03)

⚠ BREAKING CHANGES

• remove top-level rust helpers and split python init/config API
• split algorithm init from config
• add mixed Python package integration
• add sampled MCMC chain retention
• add rolling MCMC chain storage
• simplify MCMC walker storage
• standardize bounds handling modes
• validate sampler walkers and custom simplex shapes
• validate weighted MCMC move inputs
• Status: This strongly types status messages and removes convergence flags in favor of matching the message type

Features

• Add CMA-ES optimizer (9ad2122)
• Add convenience methods for gradient and gradient-free fits and MCMC sampling using reasonable default settings (66e4561)
• Add debug derives for swarm types (ccd37eb)
• Add differential evolution optimizer (166181e)
• Add generic error type and use Results rather than panics in algorithm configuration (8dc1e6a)
• Add interval progress observer (def4775)
• Add MCMC diagnostics helpers (3a145ed)
• Add mixed Python package integration (9050360)
• Add multistart minimization orchestration (7b4553e)
• Add nonlinear conjugate gradient optimizer (550d493)
• Add numpy-backed python conversions (7da54cb)
• Add python config schema exports (eb92f4f)
• Add python feature scaffolding (d072c43)
• Add python LBFGSB config wrapper (756aaf3)
• Add python mcmc summary dict export (9ee8b6c)
• Add python mcmc summary wrapper (d8790e0)
• Add python minimization summary dict export (b8bcabe)
• Add python minimization summary wrapper (693af90)
• Add python multistart summary wrapper (cb1ef76)
• Add python run options wrappers (beb34b4)
• Add python simulated annealing config wrapper (e722b04)
• Add python simulated annealing summary exports (d78232b)
• Add python wrappers for cmaes and differential evolution configs (128cebf)
• Add python wrappers for core optimizer configs (629c8b4)
• Add recoverable numerical helper errors (769da30)
• Add rolling MCMC chain storage (76f0dc3)
• Add sampled MCMC chain retention (ac6d174)
• Add scale transform helpers (72dba5c)
• Add step-boundary checkpoint and resume support (1bce7f2)
• Add summary export helpers (fe6a292)
• Add trust-region optimizer (23b611f)
• Add typed python error mapping (2fdd5b9)
• Propagate parameter names from config (54cc088)
• Standardize bounds handling modes (6f97c00)
• Status: Replace status String with a StatusMessage type and remove convergence fields in favor of StatusMessage::success() (a28a3a3)

Bug Fixes

• ci: Restore release publishing workflows (9e24122)
• Ensure all Status-like structs are correctly and fully reset on reset() calls (29ee51c)
• Faithfully record function evaluations in NM init, MCMC, and PSO init (e14cf24)
• Guard ESS adaptive scaling against zero updates (0a13ad4)
• Harden bound containment and infinite sampling (4aa3dd8)
• Properly compute free indices in L-BFGS-B (589cfde)
• Scale volume by delta^n rather than delta^{n+1} in shrink steps for Nelder-Mead (cb40999)
• Use best neighbor rather than worst in PSO ring topology (32e49dd)
• Use best simplex point instead of worst in Diameter and Higham x-terminator methods of Nelder-Mead (34d9d19)
• Use correct random variable draw range in PSO (0effd12)
• Use correct update step in simulated annealing (08201fe)
• Use proper sign in stretch move proposal (bcdd5b4)
• Validate sampler walkers and custom simplex shapes (d6b9243)
• Validate weighted MCMC move inputs (6ca7561)

Performance Improvements

• Add lightweight benchmark matrix (7929116)
• Avoid walker clones in ensemble proposals (ecdbfb3)
• Cache resolved Nelder-Mead bounds state (8da188e)
• Reduce status formatting overhead in progress paths (0462dd7)
• Simplify MCMC walker storage (7a7cdba)
• Use fused objective and derivative evaluations (7bf83c1)

Code Refactoring

• Remove top-level rust helpers and split python init/config API (3f5959c)
• Split algorithm init from config (2c82cf3)