validate_sharing#

t3toolbox.backend.sharing.validate_sharing(sharing, shape)#
def validate_sharing(
        sharing:    typ.Sequence,       # len=d, static; one hashable group label per mode, e.g. (0, 1, 1)
        shape:      typ.Sequence[int],  # (N0, ..., N(d-1)); modes sharing a label need equal sizes
) -> typ.Tuple[typ.Tuple[int, ...], ...]:  # groups, static; mode indices per group (canonical form)

Validate a sharing partition and return its canonical groups form.

sharing assigns one hashable group label per mode; modes with equal labels share one Tucker factor. The canonical form lists each group’s mode indices (ascending), groups ordered by first mode, singleton groups included – static structure (a jit aux / closure constant), never traced. Structural problems – wrong length, unhashable labels, unequal mode sizes within a group – raise unconditionally (both safety modes).

Examples

Labels are arbitrary hashables; groups come back as mode-index tuples, ordered by first mode, singletons included:

>>> import t3toolbox.backend.sharing as sharing
>>> sharing.validate_sharing((0, 1, 1, 2, 2, 2), (4, 5, 5, 6, 6, 6))
((0,), (1, 2), (3, 4, 5))
>>> sharing.validate_sharing(('in', 'out', 'in'), (7, 5, 7))   # non-adjacent groups are fine
((0, 2), (1,))

Gotcha: modes in a group must have equal mode sizes (a shared factor needs one ambient dimension) – a mismatch is a structural error:

>>> sharing.validate_sharing((0, 0), (4, 5))
Traceback (most recent call last):
    ...
ValueError
Parameters:

  • sharing (t3toolbox.backend.common.typ.Sequence)

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

Return type:

t3toolbox.backend.common.typ.Tuple[t3toolbox.backend.common.typ.Tuple[int, …], …]