Shuhao Cui

Well, line 203 tries to create the directory of 'office-home/san', and line 204 tries to create and write to the file. I wonder whether you have succesfully create the directory...

The dataset could be found in "https://github.com/cuishuhao/GVB/tree/master/data/office-home". While I don't think it is the problem. The code could automaticly create the directory, without manually created. I think this is the...

1. What environment are you using? python3? linux? 2.'office-home/san\log.txt', I think it unusual for '\' and '/' 3. I think you might have problems in 'mkdir', you might run the...

I ran the code in linux, and there are some differences on the file path between windows and linux. Since less codes are completed on windows, I recommend you to...

I think it is caused by the environment. I ran the code in Pytorch1.0 and I did not try it on other environments.

1. negative loss value is right (I want to maximize the loss, the same with minimizing the opposite value of the loss). 2. Do you get 100% acc with"source is...

Good Try! I think you are better.

I admit I am a little careless about the weight. In the equation, batch B is divided. In the code, I achieve it with min(B, C). L_bnm can be combined...

I ran visda2017 and obtained results about 72%. To train visda2017, simply changing the learning rate and class number is ok. You can refer to https://github.com/thuml/CDAN/blob/master/pytorch/train_image.py#L266

@valencebond Results of visda-2017 are also supported in https://github.com/tim-learn/ATDOC/blob/main/demo_uda.py#L228