sktsherlock / tag-benchmark Goto Github PK

I'm currently trying to reproduce the experimental results from your paper using the train_command file. However, it seems the provided commands are incomplete, making it difficult to run the experiments successfully. Could you please update the train_command file with all necessary parameters and settings for a comprehensive reproduction of the results? Your assistance in this matter would be greatly appreciated.

Hi! After downloading the dataset from the google drive link, I can't open the .pt files using torch.load. An error occurs:
_pickle.UnpicklingError: invalid load key, '?'.
I wonder if this is related to torch version. Thanks!

I am trying to load your Children dataset, and I am not entirely sure what the npy file(Children_roberta_base_512)'s use is. It is not described in the readMe. If you could clarify and give an example, that would be very helpful. Thank you

I am currently in the process of replicating the experiments from your paper, specifically focusing on Table 3. However, I have encountered an unexpected phenomenon, particularly with the Photo_RW dataset.

When using the default split method (split=time), the observed accuracies are as follows:

PLM-Based (tiny): ≈66%
GNN-based (T-SAGE): ≈83%
Co-Training Based (SAGE(T)): ≈82%
However, when employing the random split method (split=random), the accuracies change to:

PLM-Based (tiny): ≈73%
GNN-based (T-SAGE): ≈88%
Co-Training Based (SAGE(T)): ≈87%
In both split methods, it seems that GNN-based accuracy is consistently higher than Co-Training Based accuracy. This contrasts with the values reported in Table 3 of your paper, where PLM-Based (tiny) ≈73%, GNN-based (T-SAGE) ≈83%, and Co-Training Based (SAGE(T)) ≈86%. According to the paper, Co-Training Based should outperform GNN-based.

I have provided my parameter configurations for your reference. Do you have any suggestions or insights into this inconsistency? Alternatively, could you please share specific running commands or the wandb experiment records for further clarification?

Hi, thanks for you repo. I meet a problem when run the train_CLM.py, the error information is that "ModuleNotFoundError: No module named 'datasets'". In the "LMs" package, I don't find the 'datasets' package as well as the "load_dataset" function.

Traceback (most recent call last):
File "/home/pycharm_project/TAG-Benchmark/LMs/train_CLM.py", line 32, in
import datasets
ModuleNotFoundError: No module named 'datasets'

Hello,

I've noticed that for most datasets, there are both CSV and PT files available. However, for the Arxiv dataset, it seems only the CSV file is provided. Could you please guide me on where to find the PT file for the Arxiv dataset? I've looked through the repository but haven't had any luck.

Thank you for your assistance.

Sincerely,
Syzseisus