nifty.re.optimize module#

Bases: NamedTuple

Object holding optimization results inspired by JAX and scipy.

Variables:

x (ndarray) – The solution of the optimization.
success (bool) – Whether or not the optimizer exited successfully.
status (int) – Termination status of the optimizer. Its value depends on the underlying solver. NOTE, in contrast to scipy there is no message for details since strings are not statically memory bound.
hess (fun, jac,) – Values of objective function, its Jacobian and its Hessian (if available). The Hessians may be approximations, see the documentation of the function in question.
hess_inv (object) – Inverse of the objective function’s Hessian; may be an approximation. Not available for all solvers.
nhev (nfev, njev,) – Number of evaluations of the objective functions and of its Jacobian and Hessian.
nit (int) – Number of iterations performed by the optimizer.

fun: Any#: Alias for field number 3

good_approximation: None | bool | Array#: Alias for field number 13

hess: Array | None#: Alias for field number 5

hess_inv: Array | None#: Alias for field number 6

jac: Any#: Alias for field number 4

jac_magnitude: None | float | Array#: Alias for field number 12

nfev: None | int | Array#: Alias for field number 7

nhev: None | int | Array#: Alias for field number 9

nit: None | int | Array#: Alias for field number 10

njev: None | int | Array#: Alias for field number 8

status: int | Array#: Alias for field number 2

success: bool | Array#: Alias for field number 1

trust_radius: None | float | Array#: Alias for field number 11

x: Any#: Alias for field number 0

minimize(fun: Callable[[...], float] | None, x0, args: Tuple = (), *, method: str, tol: float | None = None, options: Mapping[str, Any] | None = None) → OptimizeResults[source]#: Minimize fun.

newton_cg(fun=None, x0=None, *args, **kwargs)[source]#: Minimize a scalar-valued function using the Newton-CG algorithm.

optax_wrapper(fun=None, x0=None, *, optimizer=None, maxiter=None, miniter=None, jac: Callable | None = None, fun_and_grad=None, hessp=None, name=None, xtol=1e-05)[source]#

Minimize an objective function using an Optax optimizer.

This function wraps an optax optimizer to be used in an optimize_kl optimization in NIFTy.

Parameters:

fun (Callable) – Objective function to minimize. Must accept parameters x and return a scalar value.
x0 (array-like or pytree) – Initial parameters. Must be a JAX-compatible pytree.
optimizer (optax optimizer) – Optax optimizer instance (e.g. optax.sdg(...), optax.lbfgs(...)). Must provide init and update methods.
maxiter (int, optional) – Maximum number of iterations. Defaults to 200.
miniter (int, optional) – Minimum number of iterations before convergence is checked. Defaults to 0.
jac (Callable, optional) – Function returning the gradient of fun.
fun_and_grad (Callable, optional) – Function returning both value and gradient.
hessp (Callable, optional) – Hessian-vector product function. Included for API compatibility, but not used.
name (str, optional) – Optional label used for formatted iteration logging.
xtol (float, optional) – Convergence tolerance on the gradient norm. Internally scaled by the size of x0.

Returns:

An optimization result object containing:

Return type:

OptimizeResults

static_newton_cg(fun=None, x0=None, *args, **kwargs)[source]#: Minimize a scalar-valued function using the Newton-CG algorithm.

trust_ncg(fun=None, x0=None, *args, **kwargs)[source]#