mapcontinuity-ne-reliability
A map-continuity based method to assess and improve the reliability of neighbor embedding methods.
Science Score: 49.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
○CITATION.cff file
-
✓codemeta.json file
Found codemeta.json file -
✓.zenodo.json file
Found .zenodo.json file -
✓DOI references
Found 2 DOI reference(s) in README -
✓Academic publication links
Links to: zenodo.org -
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (8.2%) to scientific vocabulary
Repository
A map-continuity based method to assess and improve the reliability of neighbor embedding methods.
Basic Info
Statistics
- Stars: 3
- Watchers: 1
- Forks: 0
- Open Issues: 0
- Releases: 2
Metadata Files
README.md
A map continuity view of assessing and improving the reliability of neighbor embedding methods
We provide two diagnostic scores to assess and improve the reliability of neighbor embedding methods based on a map-continuity view. The perturbation score is to diagnose points around the location of overconfidence-inducing discontinuity and the singularity score is to diagnose points of fracture-inducing discontinuity. Our approach is flexible and works as a wrapper around many neighbor embedding algorithms. The R package is now available for t-SNE.
The method is based on the paper:
Liu, Z., Ma, R., Zhong, Y. Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective. Nature Communications (2025) 16:5037.
Contents
The directory neMDBD contains the R package neMDBD for implementation of perturbation score and singularity score.
The directory Code and data contains the R scripts for reproducing the data analysis and examples in the manuscript.
The directory RtsneWithP contains the modified R package Rtsne with the similarity matrix as an extra output.
Get started
For the quick guide to the two diagnostic scores, please see https://github.com/zhexuandliu/MapContinuity-NE-Reliability/blob/main/neMDBD/README.md.
For other implementation details including how to draw the LOO loss landscape, please see the folder
Code and data.To check how to extract the similarity matrix computed in
Rtsne, please check out https://github.com/zhexuandliu/MapContinuity-NE-Reliability/blob/main/RtsneWithP/README.md.
For further questions and inquiries, please contact Zhexuan Liu (zhexuan.liu2@wisc.edu).
Owner
- Login: zhexuandliu
- Kind: user
- Repositories: 1
- Profile: https://github.com/zhexuandliu
GitHub Events
Total
- Release event: 6
- Watch event: 3
- Push event: 178
- Create event: 5
Last Year
- Release event: 6
- Watch event: 3
- Push event: 178
- Create event: 5