ut3_corewise_randn#

t3toolbox.backend.ut3_constructors.ut3_corewise_randn(shape, tucker_ranks, tt_ranks, stack_shape=(), use_jax=False)#
def ut3_corewise_randn(
        shape:        typ.Sequence[int],               # (N0,...,N(d-1))
        tucker_ranks,                                  # int | len-d seq | (d,)+stack array
        tt_ranks,                                      # int | len-(d+1) seq | (d+1,)+stack array
        stack_shape:  typ.Tuple[int, ...] = (),
        use_jax:      bool = False,                    # constructor: chooses the SUPERCORE type
) -> UT3Data:

Uniform Tucker tensor train with random N(0,1) supercores (padded regions masked to zero).

tucker_ranks/tt_ranks may vary per stack element (the variety) – a full (d,)+stack / (d+1,)+stack array sets per-element ranks while keeping one padded supercore shape.

Parameters:

  • shape (t3toolbox.backend.common.typ.Sequence[int])

  • stack_shape (t3toolbox.backend.common.typ.Tuple[int, Ellipsis])

  • use_jax (bool)

Return type:

UT3Data