Hi, I'm using your excellent repo for a course project on graph-based fake news detection, and I noticed the consistent use of an unusual train/validation/test split ratio of 2:1:7.
num_training = int(len(dataset) * 0.2)
num_val = int(len(dataset) * 0.1)
num_test = len(dataset) - (num_training + num_val)
This is quite different from common splits (e.g., 6:2:2). I'm very curious about the reasoning behind it. Could you share the rationale behind this ratio?
Understanding the rationale behind this split would be incredibly helpful for my project. I'll appreciate any insights you can share!
Hi, I'm using your excellent repo for a course project on graph-based fake news detection, and I noticed the consistent use of an unusual train/validation/test split ratio of 2:1:7.
This is quite different from common splits (e.g., 6:2:2). I'm very curious about the reasoning behind it. Could you share the rationale behind this ratio?
Understanding the rationale behind this split would be incredibly helpful for my project. I'll appreciate any insights you can share!