Version History

This section lists the different versions of DFO-LS and the updates between them.

Version 1.0 (6 Feb 2018)

Minor bug fix to trust region subproblem solver (the output crvmin is calculated correctly) - this has minimal impact on the performance of DFO-LS.

Extra optional input args which passes through arguments for objfun.
Bug fixes: default parameters for reduced initialization cost regime, returning correct value if exiting from within a safety step, retrieving dependencies during installation.

Use different default reduced initialization cost method for inverse problems to ensure whole space is searched correctly.
Bug fixes: default trust region radius when scaling feasible region, exit correctly when no Jacobian returned, handling overflow at initial value

Use deterministic initialisation by default (so it is no longer necessary to set a random seed for reproducibility of DFO-LS results).
Full model Hessian stored rather than just upper triangular part - this improves the runtime of Hessian-based operations.
Faster trust-region and geometry subproblem solutions in Fortran using the trustregion package.
Faster interpolation solution for multiple right-hand sides.
Don’t adjust starting point if it is close to the bounds (as long as it is feasible).
Option to stop default logging behavior and/or enable per-iteration printing.
Bugfix: correctly handle 1-sided bounds as inputs, avoid divide-by-zero warnings when auto-detecting restarts.

Minor update to remove NumPy deprecation warnings - no changes to the DFO-LS algorithm.

Minor update to customise handling of NaNs in objective evaluations - no changes to the DFO-LS algorithm.

Handle finitely many arbitrary convex constraints in addition to simple bound constraints.
Add module-level logging for more informative log outputs.
Only new functionality is added, so there is no change to the solver for unconstrained/bound-constrained problems.

Require newer SciPy version (at least 1.11) as a dependency - avoids occasional undetermined behavior but no changes to the DFO-LS algorithm.
Gracefully handle NaN objective value from evaluation at a new trial point (trust-region step).

Migrate package setup to pyproject.toml (required for Python version 3.12)
Drop support for Python 2.7 and <=3.8 in line with SciPy >=1.11 dependency (introduced v1.4.0)

Add support for (possibly nonsmooth) regularizer term.
Drop warning about infeasible initial point if the point is on an upper/lower bound.

Add return values soln.xmin_eval_num and soln.jacmin_eval_nums
Allow option for parallel initial evaluations for non-random directions if npt not too large