This problem occurred when I was training. Why

[NNNEFIT3]

(torch)user@ul1:/anaconda3/envs/torch/RoReg$ python trainset.py --component GFTraceback (most recent call last): File "trainset.py" , line 1, in import parses. parses gf as parses gf NoduleNotFoundError: No module named 'Parses.parses_gf"'

[HpWang-whu]

Hi @NNNEFIT3, thanks for your interest in our work! I have already fixed the bug and updated the code. Sorry for the inconvenience.

Best wishes.

[NNNEFIT3]

Thank you very much for providing the updated code.I have already debugged it. but I passed this command when training the dataset:"PythonTrain.py --component GF". It prompts me that the data entered is not in the correct shape. The following is the output of the program board:

   train set len 2282
   val set len 187
   | 0/11410 [00:00<?, ?it/s]Traceback (most recent call last):
   File "/home/rst/anaconda3/envs/roreg/RoReg-master/network/group_feat.py", line 23, in data_process
   data=torch.reshape(data,[data.shape[0],data.shape[1],60,13])
   RuntimeError: shape '[32, 32, 60, 13]' is invalid for input of size 47923200
    | 0/11410 [00:08<?, ?it/s]

I don't know where the problem is. Some progress bars did not reach 100% when generating the validation set. I'm not sure if this has an impact or is there a problem with the code?

Looking forward to receiving your reply.

[NNNEFIT3]

I think you have done a great job! So I want to achieve it faster.

[HpWang-whu]

Hi @NNNEFIT3, The batch_size in parses/parses_train_gf.py should be set to 1, meaning use one batch file in one forward. https://github.com/HpWang-whu/RoReg/blob/cc91a0f26951af2d738428913996728224d117dc/parses/parses_train_gf.py#L62-L63 The size of training batch is set to 32 in 'train/trainset/GF_ET.py'. I will reorganize the code later to avoid such an ambiguity.

About the validation set generation, is there any specific error reported?

Best,

[NNNEFIT3]

Yes, there were no special errors reported regarding the generation of validation sets, only a warning message:

 RoReg-master/backbone/fcgf/misc.py:96: UserWarning: To copy construct from a tensor, it is recommended to use 
 sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
 coords = torch.tensor(coords, dtype=torch.int32)

Then I modified the code to change torch.tensor( ) to torch.as_tensor( ), and the warning disappeared.

I found it time-consuming to generate a validation set, so I modified the batch_ Size, maybe this is the reason.I will modify it again.thank you very much.

Best wishes.