make_newton_display#

t3toolbox.backend.optimizer_display.make_newton_display(problem, val_sample=None, val_data=None, print_fn=print, fmt='%.1e', record=True)#
def make_newton_display(
        problem,                                  # backend.optimizers.Problem (holds the kind + full data)
        val_sample: typ.Any = None,               # optional validation sample (same kind/layout as problem.sample)
        val_data:   typ.Any = None,               # optional validation data; both given -> a train|val table
        print_fn:   typ.Optional[typ.Callable] = print,   # where to send each iteration's text (None = silent)
        fmt:        str  = '%.1e',
        record:     bool = True,                  # accumulate per-iteration records (scalars + err matrices)
) -> typ.Tuple[typ.Callable, list]:               # (callback for newton_cg, records list filled as it runs)

Build a callback(NewtonInfo) (+ its records list) that displays each Newton iteration.

Precomputes the constant per-block data norms block_sumsq(data) once (the relative-error denominators). Each call: recomputes the residual block norms, forms the train (and, if val_data is given, validation – one extra point_forward, no transpose/sweep) relative-error matrices, prints via print_fn, and appends a self-contained record (the scalar fields + train_err / val_err). Requires problem.kind.block_sumsq (all built-in kinds have it).

Parameters:

  • val_sample (t3toolbox.backend.common.typ.Any)

  • val_data (t3toolbox.backend.common.typ.Any)

  • print_fn (t3toolbox.backend.common.typ.Optional[t3toolbox.backend.common.typ.Callable])

  • fmt (str)

  • record (bool)

Return type:

t3toolbox.backend.common.typ.Tuple[t3toolbox.backend.common.typ.Callable, list]