simpeg.optimization.ProjectedGNCG#

class simpeg.optimization.ProjectedGNCG(*, lower=-inf, upper=inf, cg_maxiter=5, cg_rtol=None, cg_atol=None, step_active_set=True, active_set_grad_scale=0.01, **kwargs)[source]#

Bases: Bounded, InexactGaussNewton

Attributes

`active_set_grad_scale`	Scalar to apply to the active set's gradient
`approxHinv`	The approximate Hessian inverse is used to precondition CG.
`bfgsH0`	Approximate Hessian used in preconditioning the problem.
`callback`	A used defined callback function.
`cg_atol`	Absolute tolerance for inner CG iterations.
`cg_maxiter`	Maximum number of CG iterations.
`cg_rtol`	Relative tolerance for inner CG iterations.
`lower`	The lower bound value.
`maxIterCG`	InexactCG.maxIterCG has been deprecated.
`stepActiveSet`	ProjectedGNCG.stepActiveSet has been deprecated.
`stepOffBoundsFact`	ProjectedGNCG.stepOffBoundsFact has been deprecated.
`step_active_set`	Whether to include the active set's gradient in the step direction.
`tolCG`	InexactCG.tolCG has been deprecated.
`upper`	The upper bound value.

counter
parent
print_type

Methods

`activeSet`(x)	If we are on a bound
`bfgsrec`(k, n, nn, S, Y, d)	BFGS recursion
`bindingSet`(x)	If we are on a bound and the negative gradient points away from the feasible set.
`doEndIteration`(xt)	Operation called at the end of each minimize iteration.
`doStartIteration`()	Called at the start of each minimize iteration. If you have things that also need to run in the method doStartIteration, you can create a method::.
`findSearchDirection`()	Finds the search direction based on projected CG
`finish`()	Called at the end of the optimization. If you have things that also need to run in the method finish, you can create a method::.
`inactiveSet`(x)	The free variables.
`minimize`(evalFunction, x0)	Minimizes the function (evalFunction) starting at the location x0.
`modifySearchDirection`(p)	Changes the search direction based on some sort of linesearch or trust-region criteria.
`modifySearchDirectionBreak`(p)	Called if modifySearchDirection fails to find a descent direction.
`printDone`([inLS])	Called at the end of the optimization routine.
`printInit`([inLS])	Called at the beginning of the optimization routine.
`printIter`([inLS])	Called directly after function evaluations.
`projection`(x)	Make sure we are feasible.
`scaleSearchDirection`(p)	Scales the search direction if appropriate.
`startup`(x0)	Called at the start of any new minimize call.