search

Orkis-Research / Pytorch-Quaternion-Neural-Networks

This repository is an update to all previous repositories with implementations of various Quaternion-valued Neural Networks in PyTorch
GNU General Public License v3.0
204 stars 54 forks source link

Doubt in "QCAE" #10

Open Dxplosion opened 3 years ago

Dxplosion commented 3 years ago

Hi, I was trying to running “cae.py” in QCAE mode but experienced error

"Traceback (most recent call last): File "E:/PyCharm Community Edition 2018.3.5/PRO/Pytorch-Quaternion-Neural-Networks-master/cae.py", line 33, in net = QCAE() File "E:\PyCharm Community Edition 2018.3.5\PRO\Pytorch-Quaternion-Neural-Networks-master\convolutional_models.py", line 33, in init self.e1 = QuaternionConv(4, 32, kernel_size=3, stride=2, padding=1) File "E:\PyCharm Community Edition 2018.3.5\PRO\Pytorch-Quaternion-Neural-Networks-master\quaternion_layers.py", line 127, in init self.r_weight = Parameter(torch.Tensor(*self.w_shape)) TypeError: 'tuple' object cannot be interpreted as an integer"

I think maybe my versions of Python and Pytorch are not suitable so I want to know what versions of Python and Pytorch are you using and whether the code are needed to be modified?

Thank you, Best, Xu

TParcollet commented 3 years ago

Hi, interesting. I suppose it was pytorch 1.0.0 or something like this. I have to check what must be done.

Dxplosion commented 3 years ago

Thanks for reply. I have also tried pytorch 1.0.0 and there is still the same error. According to the error messages, the error happens because "self.w_shape" is a tuple whose pos 3 is not a int. In the default mode, "self.w_shape" is a tuple with shape of (8, 1, (3, 3)) and this results in failure in converting tuple to Tensor.