sparse#

torchdyno.models.initializers.sparse(shape, density=0.01, values_sampler=None, zero_diagonal=True, seed=None, dtype=torch.float32)[source]#

Sparse random tensor.

Parameters:

  • shape (torch.Size) – shape of the tensor.

  • density (float, optional) – density of the sparse tensor. Defaults to 0.01.

  • values_sampler (Optional[Callable[..., np.ndarray]], optional) – function to sample values for the sparse tensor. Defaults to None.

  • zero_diagonal (bool, optional) – whether to zero the diagonal. Defaults to True.

  • seed (Optional[int], optional) – random seed. Defaults to None.

  • dtype (dtype)

Returns:

initialized sparse random tensor.

Return type:

torch.Tensor