28-10-2026, 10:19 PM
(This post was last modified: 28-10-2026, 10:23 PM by npcompl33t.)
Here are the 2 Q13 samples on your github, compared to the actual Bio in MDS.
Your samples substantially misestimate several common Bio transitions, even with Q13 included in training. Reproducing a broad arc needs to be distinguished from reproducing the manuscript’s detailed distance structure.
If you can share the data or source code though, we could probably get to the bottom of what is happening! I think a generator could be really useful to analyze, if it could replicate the MS statistics convincingly. I'd be happy to help!
Your samples substantially misestimate several common Bio transitions, even with Q13 included in training. Reproducing a broad arc needs to be distinguished from reproducing the manuscript’s detailed distance structure.
If you can share the data or source code though, we could probably get to the bottom of what is happening! I think a generator could be really useful to analyze, if it could replicate the MS statistics convincingly. I'd be happy to help!