SamplingKind

class t3toolbox.backend.fitting.SamplingKind

Bases: t3toolbox.backend.common.ValueHashedFields

A sampling kind: the bare J / J^T for one measurement operator, plus its reductions.

The what you measure axis of a fit. Holds the kind-specific operations the geometry-generic Gauss-Newton model needs – the bare Jacobian and its transpose (no gauge Pi: that is the geometry’s), the |.|^2 reduction over the sample stack, and the point operation S(x) used to form the residual – plus the minimal layout the default minibatch draw needs.

Built-in kinds: APPLY, ENTRIES, PROBE (module singletons) and the parameterized ProbeKind, ApplyDerivativesKind, EntriesDerivativesKind, ProbeDerivativesKind. The uniform twins are in t3toolbox.backend.uniform_fitting.

Writing your own. Subclass and supply the five operation methods; you get value identity for free provided your parameters are dataclass fields (that is the point – a kind whose parameters hide in closures cannot be compared, and silently recompiles).

name: t3toolbox.backend.common.typ.ClassVar[str] = 'sampling'

has_order: t3toolbox.backend.common.typ.ClassVar[bool] = False

has_block_sumsq: t3toolbox.backend.common.typ.ClassVar[bool] = False

Methods

precompute(frame, sample)	The reusable frame sweep at frame for sample (computed once per local model).
forward(variations, sample, frame, sweep)	The bare Jacobian J v (no gauge).
transpose(residual, sample, frame, sweep)	The bare transpose J^T r, summed over the sample stack W; raw (dU, dG).
point_forward(x_cores, sample)	The point operation S(x) -- what the residual is measured against.
take(sample, data, idx)	Gather a flat W subset of (sample, data) -- the default minibatch draw's layout hook.
w_axes(sample)	The number of leading sample-stack (W) axes.
n_measurements(sample)	The flat |W| measurement count, for the default draw.
sumsq(out, n_w)	|out|^2 over W (and the order / free-mode axes), keeping the frame stack C.
block_sumsq(out, n_w)	Per-(mode, order) |.|^2 -> a 2-D matrix; UNWEIGHTED (the honest diagnostic table).