minuet.nn.functional.scan#

Functions

compute_kernel_map_masks(num_sources, ...)

Compute the input & output masks (metadata tables) from the kernel map.

compute_kernel_map_sizes(kernel_map)

Compute the size of each weights of the kernel map.
compute_kernel_map_masks(num_sources: int, kernel_map: Tensor, kernel_map_sizes: Tensor)#

Compute the input & output masks (metadata tables) from the kernel map. This method should not be directly used by the user.

Parameters:

  • num_sources – the number of source coordinates

  • kernel_map – the tensor of shape \((N_\text{weight}, D)\) representing the kernel map

  • kernel_map_sizes – the tensor that stores the size of the kernel map for each weight

Returns:

the tensor of input masks and the tensor of output masks

compute_kernel_map_sizes(kernel_map: Tensor)#

Compute the size of each weights of the kernel map. This method should not be directly used by the user.

Parameters:

kernel_map – the tensor of shape \((N_\text{weight}, D)\) representing the kernel map

Returns:

a tensor of shape \((N_\text{weight})\) representing the size of the

kernel map