xinyangz/ltrn

Code for WWW 2021 paper "Minimally Supervised Structure Rich Text Categorization by Learning on Text-Rich Networks"

ltrn's Issues

debug_train.csv file not found

Hi Xinyang,

I am getting error that seems to point to a local file in your environment.

run_amazon.sh
FileNotFoundError: [Errno 2] No such file or directory: '/data1/xinyang/debug_train.csv'

This is referenced in:
bert/dataset.py: df_train.to_csv("/data1/xinyang/debug_train.csv")

Should this be replaced with train_docs.csv or is there another file needed?

Best regards,
Raja

When I ran the LTRN, the accuracy is low.

Hi Xinyang,

When I ran the LTRN, I am getting some errors.

File "miniconda3/lib/python3.8/site-packages/transformers/hf_argparser.py", line 159

ValueError: Some specified arguments are not used by the HfArgumentParser: [remaining_args]

After printing the value of remaining_args, I found remaining_args is false. So I deleted the ValyeError and return the namespace. Then the code ran successfully.

I revised joint_traing.py train_bert line 202. I revised "torch.distributed.launch" with "torch.distributed.run". If I could not revise, it had an error.

RuntimeError:Invalid device string: '0'

My running the experiments is python3.8+torch1.9.0+cuda11.1. The GPU is RTX30902, memory is 24G2.

Can I modify the code like this?

When the number of iterations is 3, the accuracy of amazon is:

Iter 0,Mi-F1 0.0682 Ma-F1 0.0519

Iter 1,Mi-F1 0.2091 Ma-F1 0.1069

Iter 2,Mi-F1 0.2323 Ma-F1 0.1053

The accuracy of books is:

Iter 0,Mi-F1 0.4348 Ma-F1 0.4300

Iter 1,Mi-F1 0.5535 Ma-F1 0.5694

Iter 2,Mi-F1 0.5431 Ma-F1 0.5569

They are less than 92%.

What do you think are the possible reasons for the low accuracy?

Best regards,

Li Yun