NewtonInfo

class t3toolbox.backend.optimizers.NewtonInfo

Per-iteration diagnostics passed to a newton_cg callback – everything one Newton step produces, so a custom callback (or the display in t3toolbox.backend.optimizer_display) can report anything without the loop anticipating it. x_cores (the point before the step) and lm (its LocalModel – residual / sample / frame) are carried so a callback can compute per-block errors or evaluate a validation forward. The step-related fields are None on the final converged line (no CG / line search ran). The scalar subset (all but x_cores / lm) is what lands in stats['history'].

iteration: int

objective: float

gnorm: float

g0norm: float

converged: bool

x_cores: t3toolbox.backend.common.typ.Any = None

lm: t3toolbox.backend.common.typ.Any = None

misfit: t3toolbox.backend.common.typ.Optional[float] = None

regularization: t3toolbox.backend.common.typ.Optional[float] = None

forcing_eta: t3toolbox.backend.common.typ.Optional[float] = None

cg_tol: t3toolbox.backend.common.typ.Optional[float] = None

cg_iters: t3toolbox.backend.common.typ.Optional[int] = None

cg_maxiter: t3toolbox.backend.common.typ.Optional[int] = None

cg_resid: t3toolbox.backend.common.typ.Optional[float] = None

cg_converged: t3toolbox.backend.common.typ.Optional[bool] = None

cg_truncated: t3toolbox.backend.common.typ.Optional[bool] = None

ls_steps: t3toolbox.backend.common.typ.Optional[int] = None

alpha: t3toolbox.backend.common.typ.Optional[float] = None

slope: t3toolbox.backend.common.typ.Optional[float] = None

pHp: t3toolbox.backend.common.typ.Optional[float] = None

delta_f: t3toolbox.backend.common.typ.Optional[float] = None

rho: t3toolbox.backend.common.typ.Optional[float] = None

step_rel: t3toolbox.backend.common.typ.Optional[float] = None

wall_time: t3toolbox.backend.common.typ.Optional[float] = None