aquakin.OptimizerConfig#

class aquakin.OptimizerConfig(method='lbfgsb', n_starts=1, jitter=0.5, jitter_schedule=None, seed=0, max_iter=500, tol=1e-06, param_halfwidth=None)[source]#

Bases: object

Optimiser and multistart settings for calibration.

Parameters:

  • method ({"lbfgsb", "gauss_newton"}) – Optimisation backend. "lbfgsb" (default) is SciPy L-BFGS-B on the scalar loss; "gauss_newton" is a least-squares solve on the residual.

  • n_starts (int) – Number of optimiser starts. 1 (default) is a single start from the initial parameters; >1 adds jittered restarts and keeps the best.

  • jitter (float) – Std dev of the Gaussian restart perturbation in unconstrained space (ignored when jitter_schedule is given).

  • jitter_schedule (tuple of float, optional) – Explicit per-restart jitter scales; overrides jitter and fixes the restart sequence for reproducibility.

  • seed (int) – Seed for the restart perturbations.

  • max_iter (int) – Maximum optimiser iterations per start.

  • tol (float) – Optimiser convergence tolerance.

  • param_halfwidth (float, optional) – If given, box-bound each free rate parameter to theta0 +/- param_halfwidth in unconstrained space; otherwise the rate parameters are unbounded.

__init__(method='lbfgsb', n_starts=1, jitter=0.5, jitter_schedule=None, seed=0, max_iter=500, tol=1e-06, param_halfwidth=None)#
Parameters:

  • method (str)

  • n_starts (int)

  • jitter (float)

  • jitter_schedule (tuple | None)

  • seed (int)

  • max_iter (int)

  • tol (float)

  • param_halfwidth (float | None)

Return type:

None

Methods

__init__([method, n_starts, jitter, ...])

Attributes

jitter

jitter_schedule

max_iter

method

n_starts

param_halfwidth

seed

tol