jina.types.ndarray.sparse.pytorch

class jina.types.ndarray.sparse.pytorch.SparseNdArray(transpose_indices=True, *args, **kwargs)[source]

Bases: jina.types.ndarray.sparse.BaseSparseNdArray

Pytorch powered sparse ndarray, i.e. FloatTensor

See also

https://pytorch.org/docs/stable/sparse.html

Parameters

transpose_indices (bool) – in torch, the input to LongTensor is NOT a list of index tuples.

If you want to write your indices this way, you should transpose before passing them to the sparse constructor

Note

To comply with Tensorflow, transpose_indices is set to True by default

sparse_constructor(indices, values, shape)[source]

Sparse NdArray constructor, must be implemented by subclass

Parameters

  • indices (np.ndarray) – the indices of the sparse array

  • values (np.ndarray) – the values of the sparse array

  • shape (List[int]) – the shape of the dense array

Return type

FloatTensor

Returns

sparse_parser(value)[source]

Parsing a Sparse NdArray to indices, values and shape, must be implemented by subclass

Parameters

value (FloatTensor) – the sparse ndarray

Returns

a Dict with three entries {‘indices’: …, ‘values’:…, ‘shape’:…}