How is the code implemented on the fully labeled source data?

[zhouyaqian666]

Thank you very much for sharing the code! I have a problem that how is the code implemented by using full source labels?

[zhengzangw]

You can use all source labels by using the data split file without the suffix '_labeled_xx'. For example, there is a txt file 'amazon.txt' containing all label information.

[zhouyaqian666]

You can use all source labels by using the data split file without the suffix '_labeled_xx'. For example, there is a txt file 'amazon.txt' containing all label information.

Sorry, I didn't quite understand. If I use this file“amazon.txt”, it's obviously incorrect to execute this command "CUDA_VISIBLE_DEVICES=0,1 python pcs/run.py --config config/office/D-A-1.json".
Besides, if it is necessary to modify the code in the program？

[zhengzangw]

I read the code again and find https://github.com/zhengzangw/PCS-FUDA/blob/master/pcs/agents/CDSAgent.py#L168, if the fewshot is None, then all should be used. Maybe you can try setting fewshot=None. (You can modify the config file to achieve this).

If this does not work, you can rename the 'amazon.txt' to 'amazon_labeled_999' (999 is a nonsense word). And pass the 999 to the fewshot number. This hard-coded way can work, too, I think.

[zhouyaqian666]

我又看了一遍代码，发现https://github.com/zhengzangw/PCS-FUDA/blob/master/pcs/agents/CDSAgent.py#L168，如果fewshot是None，那么应该全部使用。也许你可以尝试设置fewshot=None。（您可以修改配置文件来实现此目的）。

如果这不起作用，您可以将“amazon.txt”重命名为“amazon_labeled_999”（999 是一个无意义的词）。并将 999 传递给少数人。我认为这种硬编码方式也可以工作。

I have tried the above ways. If the fewshot is None, it does not work. So we set the fewshot number to 500, but it shows the following error" RuntimeError: cannot perform reduction function max on tensor with no elements because the operation does not have an identity." This is very distressing to me.

[zhengzangw]

This is strange. Do you rename the file in data/split/office? As I remembered, it is not hard to conduct full label training based on the released code. If you still cannot run the full label setting, I can come back to you after the CVPR deadline.

[zhouyaqian666]

This is strange. Do you rename the file in data/split/office? As I remembered, it is not hard to conduct full label training based on the released code. If you still cannot run the full label setting, I can come back to you after the CVPR deadline.

Thanks, I'll think again about how to solve.