chrisallenming / ltc-msda Goto Github PK

Hello, when I train the full model using "python train.py --use_target --save_model --target "mnistm" --checkpoint_dir "checkpoint"", I get the following error,
Traceback (most recent call last):
File "train.py", line 114, in
main()
File "train.py", line 105, in main
num = solver.train_gcn_adapt(t, record_file=record_train)
File "./LtC-MSDA/solver.py", line 348, in train_gcn_adapt
loss.backward(retain_graph = True)
File "/root/anaconda3/lib/python3.7/site-packages/torch/tensor.py", line 185, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "/root/anaconda3/lib/python3.7/site-packages/torch/autograd/init.py", line 127, in backward
allow_unreachable=True) # allow_unreachable flag
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [2048]] is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

However, though I set all "inplace=True" as "inplace=False", the error also occurs. Can you tell me how to deal with it? Thanks.

Hi, I know that if I want to train with my own dataset, I should use custom dataloader to load the data.
Inside the datasets folder, these *.py file were designed for loading mat file.

However, I think the most relevant to custom dataset setting should be DomainNet, but I don't see any dataloader for DomainNet in this repo, Do you mind provide the example code?

Could you provide dataloader for DomainNet to train it?

Hi,

Could you provide a more detailed training setup of using AlexNet in office-31? E.g., learning rate decay, data augmentation.

Table 6 in the supplementary is helpful but not sufficient to reproduce the results, especially on the Amazon domain where I could only get around 41% accuracy, which is far from the 51.6% accuracy as reported for SourceOnly.

Btw, I used a constant learning rate of 5e-5, and data augmentation of random flip & crop.

Thanks