Module optimizers 

Numerical optimizers minimizing an Objective.

Every optimizer in this module reduces a scalar objective f: ℝⁿ → ℝ and returns an OptimizeResult reporting the located point, its objective value, the iteration count, and a ConvergenceStatus. The families are grouped into subfolders: gradient (first-order learning-rate methods and conjugate gradient), second_order (Newton and L-BFGS), and stochastic (simulated annealing and genetic / differential evolution). The shared test objectives live in objectives.

scipy.optimize mapping

Each optimizer is paired with the scipy.optimize method it is cross-checked against; methods with no faithful counterpart are documented as excluded so the comparison coverage is auditable.

stats-claw	scipy.optimize	agreement
gradient_descent	none (vanilla GD)	excluded
sgd	none	excluded
adam	none	excluded
rmsprop	none	excluded
adagrad	none	excluded
conjugate_gradient	minimize(method="CG")	compared
newton	minimize(method="Newton-CG")	compared
lbfgs	minimize(method="L-BFGS-B")	compared
simulated_annealing	dual_annealing	optimum
genetic	differential_evolution	optimum

Deterministic optimizers (exempt from the seed-variation check): gradient_descent, adam, rmsprop, adagrad, conjugate_gradient, newton, lbfgs. Stochastic optimizers (seed-variation required): sgd, simulated_annealing, genetic.

Modules

gradient

First-order optimizers: learning-rate methods (gradient descent, SGD, Adam, RMSProp, AdaGrad) and the Fletcher–Reeves conjugate gradient.

objectives

Standard test objectives for the optimizer suite.

second_order

Second-order optimizers: Newton’s method (Hessian-based) and L-BFGS (limited- memory quasi-Newton).

stochastic

Derivative-free stochastic optimizers: simulated annealing and a genetic (differential-evolution) optimizer.

Structs

OptimizeResult

The result of minimizing an Objective.

Enums

ConvergenceStatus

Outcome of an optimization run: whether the stopping criterion was satisfied.

Traits

Objective

A differentiable scalar objective f: ℝⁿ → ℝ to be minimized.

Functions

dot

Dot product of two equal-length vectors (extra elements of the longer one are ignored, which never happens for the matched-length inputs used internally).

matvec

Multiplies a square matrix (row-major Vec<Vec<f64>>) by a vector.

norm

Euclidean (L2) norm of a vector.