apply#

The apply sampling type: contract a T3 with vectors in ALL modes -> one scalar per sample.

Holds the t3 + tv ops, the ambient/tangent/corewise transposes, and the frame sweeps for applies. Specializes the general probing machinery (containment probe ⊃ apply ⊃ entries; this module imports from probing, never the reverse). Costs and the role in Riemannian least-squares: docs/entries_apply_probe.md.

Functions#

t3_apply(x, vecs)

Contract a Tucker tensor train with vectors in all indices.

t3_apply_ambient_transpose(c, ww[, sum_over_probes])

Ambient transpose of t3_apply(): back-project c into CP factors.

tv_apply(ww, variation, frame)

Apply a tangent vector in all modes: contract the dense tangent with ww in every index.

tv_apply_transpose(c, ww, frame[, sum_over_probes])

Apply the transpose of tv_apply() -- back-project a residual c into a tangent.

tv_precompute_apply_frame_sweep(frame, ww)

The all-modes apply frame sweep (lean): the frame edge variables (xi-hat, mu-hat) that

tv_apply_jacobian_from_sweep(variation, ww, frame, ...)

Forward all-modes apply of a tangent vector reusing a precomputed frame sweep -- the bare 𝒥 with

tv_apply_transpose_from_sweep(c, ww, frame, frame_sweep)

Transpose of the all-modes apply reusing a precomputed frame sweep -- the bare 𝒥ᵀ, by the

t3_apply_corewise_transpose(c, ww, core_pair[, ...])

Corewise (non-manifold) transpose of t3_apply(): gradient of the