Sparse operators

reindex_sparse is the exact-support operation for decoder and context branches. It keeps target row order and coordinate identity, drops source-only rows, and fills target-only rows without routing through a pooling kernel.

prune keeps explicit row indices in caller order. prune_mask is the boolean-mask form and preserves feature gradients through selected rows.

sparse_from_coordinates(..., duplicate_reduction='mean') performs an unweighted feature mean for exact duplicate integer coordinates. Reduced coordinates retain their first-occurrence order.

torch_lattice.operators.cat(inputs, *, join='inner')

Return type:

SparseTensor

Parameters:

• inputs (list[SparseTensor])

• join (Literal['inner', 'left', 'right', 'outer'])

torch_lattice.operators.generative_add(a, b)

Return type:

SparseTensor

Parameters:

• a (SparseTensor)

• b (SparseTensor)

torch_lattice.operators.prune(input, rows)

Keep sparse rows in caller-supplied order.

Return type:

SparseTensor

Parameters:

• input (SparseTensor)

• rows (Tensor)

torch_lattice.operators.prune_mask(input, mask)

Keep sparse rows selected by a boolean mask.

Return type:

SparseTensor

Parameters:

• input (SparseTensor)

• mask (Tensor)

torch_lattice.operators.reindex_sparse(input, target, *, fill=0.0)

Gather input features onto the exact row order of target.

Return type:

SparseTensor

Parameters:

• input (SparseTensor)

• target (SparseTensor)

• fill (float)

torch_lattice.operators.sparse_add(lhs, rhs, *, join='outer', lhs_fill=0.0, rhs_fill=0.0)

Return type:

SparseTensor

Parameters:

• lhs (SparseTensor)

• rhs (SparseTensor)

• join (Literal['inner', 'left', 'right', 'outer'])

• lhs_fill (float)

• rhs_fill (float)

torch_lattice.operators.sparse_binary(lhs, rhs, op, *, join='outer', lhs_fill=0.0, rhs_fill=0.0)

Return type:

SparseTensor

Parameters:

• lhs (SparseTensor)

• rhs (SparseTensor)

• op (Literal['add', 'sub', 'mul', 'maximum', 'minimum'])

• join (Literal['inner', 'left', 'right', 'outer'])

• lhs_fill (float)

• rhs_fill (float)

torch_lattice.operators.sparse_cat(inputs, *, join='inner')

Return type:

SparseTensor

Parameters:

• inputs (list[SparseTensor])

• join (Literal['inner', 'left', 'right', 'outer'])

torch_lattice.operators.sparse_from_coordinates(coords, feats, *, stride=1, spatial_range=None, batch_counts=None, duplicate_reduction='none')

Construct a sparse tensor with explicit duplicate-row semantics.

Return type:

SparseTensor

Parameters:

• coords (Tensor)

• feats (Tensor)

• stride (int | Sequence[int])

• spatial_range (int | Sequence[int] | None)

• batch_counts (Sequence[int] | None)

• duplicate_reduction (Literal['none', 'mean'])

torch_lattice.operators.sparse_maximum(lhs, rhs, *, join='inner', lhs_fill=0.0, rhs_fill=0.0)

Return type:

SparseTensor

Parameters:

• lhs (SparseTensor)

• rhs (SparseTensor)

• join (Literal['inner', 'left', 'right', 'outer'])

• lhs_fill (float)

• rhs_fill (float)

torch_lattice.operators.sparse_minimum(lhs, rhs, *, join='inner', lhs_fill=0.0, rhs_fill=0.0)

Return type:

SparseTensor

Parameters:

• lhs (SparseTensor)

• rhs (SparseTensor)

• join (Literal['inner', 'left', 'right', 'outer'])

• lhs_fill (float)

• rhs_fill (float)

torch_lattice.operators.sparse_mul(lhs, rhs, *, join='inner', lhs_fill=0.0, rhs_fill=0.0)

Return type:

SparseTensor

Parameters:

• lhs (SparseTensor)

• rhs (SparseTensor)

• join (Literal['inner', 'left', 'right', 'outer'])

• lhs_fill (float)

• rhs_fill (float)

torch_lattice.operators.sparse_sub(lhs, rhs, *, join='outer', lhs_fill=0.0, rhs_fill=0.0)

Return type:

SparseTensor

Parameters:

• lhs (SparseTensor)

• rhs (SparseTensor)

• join (Literal['inner', 'left', 'right', 'outer'])

• lhs_fill (float)

• rhs_fill (float)