Bug Lab dataset

[sumit-agrwl]

I saw that there were experiments where you compared against buglab. I am having difficulties in generating data using the docker method proposed by buglab. It would be really helpful if you can share the buglab data which you used for your experiments.

[LostBenjamin]

Hi,

We did not use the BugLab data. Instead, we ran BugLab with our dataset converted to the BugLab format.

Best,
Jingxuan

[LostBenjamin]

Why not use our dataset?

[sumit-agrwl]

Hi Jingxuan,

Thank you for your prompt reply. I am currently planning to use your dataset. I have a small confusion there. The dataset folder contains three splits "synthetic", "constrastive" and "real". I believe the contrastive data was used in the first iteration with the contrastive loss and the real data was used in the second iteration. What is the synthetic split ?

[LostBenjamin]

Your understanding about "contrastive" and "real" is correct. "synthetic" is for a version of the first iteration without the contrastive loss.

[LostBenjamin]

I received a notification on a commnet asking where the tokinizer is but somehow the comment is not shown now.

The tokenizer vocabulary is included in the pretrained model (README updated for this).