AttributeError: 'Parameter' object has no attribute 'hess'

[samiit]

Hi David,

Thanks for this repo. I tried it as an optimizer on toy examples and it worked fine but am struggling to make it work for an object detection case. Specifically, I am trying to use this as an optimizer for Yolo v5 from Ultralytics.

When I updated the train.py file there, I am facing some issues at the time of training. It uses a scheduler and I was wondering whether this is the source of the problem.
Here is part of the error log:

 File "train.py", line 309, in train
    optimizer.step()
  File "/usr/local/lib/python3.6/dist-packages/torch/optim/lr_scheduler.py", line 67, in wrapper
    return wrapped(*args, **kwargs)
  File "/usr/local/lib/python3.6/dist-packages/torch/autograd/grad_mode.py", line 26, in decorate_context
    return func(*args, **kwargs)
  File "/yolov5/ada_hessian.py", line 100, in step
    self.zero_hessian()
  File "/yolov5/ada_hessian.py", line 59, in zero_hessian
    if not isinstance(p.hess, float) and self.state[p]["hessian step"] % self.update_each == 0:
AttributeError: 'Parameter' object has no attribute 'hess'

Do you have any suggestions?

Regards,
Sam

[davda54]

Hi,

do you overwrite the model parameters after initializing the optimizer? The error on line 59 shouldn't normally occur, because every parameters is initialized with hess when creating the optimizer:

for p in self.get_params():
    p.hess = 0.0

[samiit]

Hi David,

I am still checking this out. Kindly don't close the issue yet. Hopefully I can get back to you by today.

Sam

[SamMohel]

Did you solve it please ?