Dataset splits and evaluation procedure

#1
by gings - opened

Dear authors,

Thanks for providing this great dataset. I was wondering:

  1. How to split data into train/val/test as you did in your paper?

  2. Can you give more information as to exactly how to evaluate a model on your benchmark, in order to produce comparable numbers to your paper?

Best,

Simon Ging

Owner

Thank you for your interest and for reaching out. I appreciate your attention to our work.

The code still needs some organization and refinement. I anticipate releasing it in September or October, which will certainly be before the conference.

Regarding the dataset split, we used the following approach:

Training set: The first 7 datasets
Validation set: The 8th dataset
Test set: The last 2 datasets
This split can be easily identified through our naming convention in the dataset.

Once I have everything properly arranged, including the code and detailed information about the dataset split, I will upload a link to the GitHub repository on Hugging Face. Please stay tuned for this update.

Sign up or log in to comment