Skip to content

Autograd not working for torch.exp(1j * phase) #43349

Closed
Closed
Autograd not working for torch.exp(1j * phase)#43349
Labels
high prioritymodule: autogradRelated to torch.autograd, and the autograd engine in generalmodule: complexRelated to complex number support in PyTorchtriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
@jonashaag

Description

@jonashaag
jonashaag
opened on Aug 20, 2020

🐛 Bug

Complex tensor construction from magnitude and phase does not seem to support autograd when using mag * torch.exp(1j * phase) notation:

import torch
mag, phase = torch.tensor(5., requires_grad=True), torch.tensor(3., requires_grad=True)

complex_good = torch.view_as_complex(torch.stack([mag * torch.cos(phase), mag * torch.sin(phase)], dim=-1))
complex_good.backward()  # works

complex_bad = mag * torch.exp(1j * phase)
complex_bad.backward()

=>

.../torch/autograd/__init__.py:125: UserWarning: Complex backward is not fully supported yet and could lead to wrong gradients for functions we have not fixed yet (Triggered internally at  /opt/conda/conda-bld/pytorch_1597820903894/work/torch/csrc/autograd/python_engine.cpp:172.)
  Variable._execution_engine.run_backward(
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File ".../torch/tensor.py", line 214, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph)
  File ".../torch/autograd/__init__.py", line 125, in backward
    Variable._execution_engine.run_backward(
RuntimeError: Expected isFloatingType(grad.scalar_type()) || (input_is_complex == grad_is_complex) to be true, but got false.  (Could this error message be improved?  If so, please report an enhancement request to PyTorch.)
...

Torch version: 1.7.0.dev20200819

cc @ezyang @gchanan @zou3519 @ssnl @albanD @gqchen @anjali411 @dylanbespalko

Metadata

Metadata

Assignees

No one assigned

    Labels

    high prioritymodule: autogradRelated to torch.autograd, and the autograd engine in generalmodule: complexRelated to complex number support in PyTorchtriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions