ut3_pad_ranks#

t3toolbox.backend.ut3_operations.ut3_pad_ranks(data, n, r)#
def ut3_pad_ranks(
        data: UT3Data,
        n:    int,          # target padded Tucker rank (>= the current padded n)
        r:    int,          # target padded TT rank    (>= the current padded r)
) -> UT3Data:               # the same tensor, ranks and masks, stored at padded dims (n, r)

Zero-pad the rank dims of a uniform T3 to (n, r) – the storage changes, the represented tensor, the ranks and the mask CONTENT do not (the new slots are padding: mask False, value 0). The inverse of the pad-shrinking ut3svd does (which slices to the max rank it kept); utv_retract uses it to return a point at the frame’s padded dims, so the optimizers’ loop-invariant masks stay valid.

Parameters:

  • data (UT3Data)

  • n (int)

  • r (int)

Return type:

UT3Data