def get_queries(self, queries: torch.Tensor):
"""Compute embedding and biases of queries."""
lhs_e = givens_rotations(self.rel_diag(queries[:, 1]),
self.entity(queries[:, 0])) + self.rel(
queries[:, 1])
lhs_biases = self.bh(queries[:, 0])
return lhs_e, lhs_biases
def get_queries(self, queries):
"""Compute embedding and biases of queries."""
c = F.softplus(self.c[queries[:, 1]])
head = expmap0(self.entity(queries[:, 0]), c)
rel1, rel2 = torch.chunk(self.rel(queries[:, 1]), 2, dim=1)
rel1 = expmap0(rel1, c)
rel2 = expmap0(rel2, c)
lhs = project(mobius_add(head, rel1, c), c)
res1 = givens_rotations(self.rel_diag(queries[:, 1]), lhs)
res2 = mobius_add(res1, rel2, c)
return (res2, c), self.bh(queries[:, 0])
def get_queries(self, queries):
"""Compute embedding and biases of queries."""
c = F.softplus(self.c[queries[:, 1]])
head = self.entity(queries[:, 0])
rot_mat, ref_mat = torch.chunk(self.rel_diag(queries[:, 1]), 2, dim=1)
rot_q = givens_rotations(rot_mat, head).view((-1, 1, self.rank))
ref_q = givens_reflection(ref_mat, head).view((-1, 1, self.rank))
cands = torch.cat([ref_q, rot_q], dim=1)
context_vec = self.context_vec(queries[:, 1]).view((-1, 1, self.rank))
att_weights = torch.sum(context_vec * cands * self.scale,
dim=-1,
keepdim=True)
att_weights = self.act(att_weights)
att_q = torch.sum(att_weights * cands, dim=1)
lhs = expmap0(att_q, c)
rel, _ = torch.chunk(self.rel(queries[:, 1]), 2, dim=1)
rel = expmap0(rel, c)
res = project(mobius_add(lhs, rel, c), c)
return (res, c), self.bh(queries[:, 0])
def get_rotation_queries(self, queries):
lhs_rot_e = givens_rotations(self.rot(queries[:, 1]),
self.entity(queries[:, 0]))
return lhs_rot_e