espnet2.enh.layers.complex_utils.complex_norm

source

espnet2.enh.layers.complex_utils.complex_norm(c: Tensor | ComplexTensor, dim=-1, keepdim=False) → Tensor

Compute the norm of a complex tensor.

This function calculates the norm of a complex tensor along a specified dimension. If the input tensor is not complex, a TypeError is raised. The function can handle both PyTorch’s native complex tensors and custom ComplexTensor objects.

• Parameters:
  • c (Union *[*torch.Tensor , ComplexTensor ]) – The input tensor for which to compute the norm. It must be either a complex tensor or a compatible type.
  • dim (int , optional) – The dimension along which to compute the norm. Default is -1, which means the last dimension.
  • keepdim (bool , optional) – Whether to retain the dimensions of the input tensor in the output. Default is False.
• Returns: The computed norm of the input tensor.
• Return type: torch.Tensor
• Raises:TypeError – If the input tensor is not a complex tensor.

Examples

>>> import torch
>>> from torch_complex import ComplexTensor
>>> c_tensor = torch.tensor([[1.0, 2.0], [3.0, 4.0]], dtype=torch.complex64)
>>> complex_norm(c_tensor)
tensor(5.4772)

>>> c_tensor_custom = ComplexTensor(torch.tensor([1.0, 2.0]),
...                                   torch.tensor([3.0, 4.0]))
>>> complex_norm(c_tensor_custom, dim=0)
tensor([3.1623, 4.4721])

NOTE

This function adds a small epsilon value to the result to avoid division by zero errors.