def ne(self, y, _scale=True):
"""Returns self != y"""
if comm.get().get_world_size() == 2:
return 1 - self.eq(y, _scale=_scale)
difference = self - y
difference.share = torch_stack([difference.share, -(difference.share)])
return difference._ltz(_scale=_scale).sum(0)
def square(size, device=None):
"""Generate square double of given size"""
r = generate_random_ring_element(size, device=device)
r2 = r.mul(r)
# Stack to vectorize scatter function
stacked = torch_stack([r, r2])
stacked = ArithmeticSharedTensor(stacked, precision=0, src=0)
return stacked[0], stacked[1]
def stack(seq, *args, **kwargs):
"""Stacks a list of tensors along a given dimension"""
assert isinstance(seq, list), "Stack input must be a list"
assert isinstance(
seq[0], BinarySharedTensor
), "Sequence must contain BinarySharedTensors"
result = seq[0].shallow_copy()
result.share = torch_stack(
[BinarySharedTensor.share for BinarySharedTensor in seq], *args, **kwargs
)
return result
def stack(tensors, *args, **kwargs):
"""Perform tensor stacking"""
for i, tensor in enumerate(tensors):
if is_tensor(tensor):
tensors[i] = ArithmeticSharedTensor(tensor)
assert isinstance(
tensors[i], ArithmeticSharedTensor
), "Can't stack %s with ArithmeticSharedTensor" % type(tensor)
result = tensors[0].shallow_copy()
result.share = torch_stack([tensor.share for tensor in tensors], *args,
**kwargs)
return result
def _get_additive_PRSS(self, size, remove_rank=False):
"""
Generates a plaintext value from a set of random additive secret shares
generated by each party
"""
gens = self._get_generators(device=self.device)
if remove_rank:
gens = gens[1:]
result = torch_stack([
generate_random_ring_element(size, generator=g, device=g.device)
for g in gens
])
return result.sum(0)
