Closed bendemeo closed 1 day ago
+1 but this should go hand-in-hand with a metric that assesses how well the density is preserved in the embedding. This can be taken directly from the denSNE paper -- Eq (10) in https://www.nature.com/articles/s41587-020-00801-7.
This issue has been automatically closed because it has not had recent activity.
What is the method? denSNE (and the related densMAP, separate issue) address the issue of density preservation in two-dimensional visualization. While t-SNE assigns larger populations a greater area in the projection, denSNE aims to assign area based not on population size but on the amount of space it takes up in the input multi-dimensional space. denSNE is published here.
Where is the code located? https://github.com/hhcho/densvis
Which task(s) could it be used for? Dimensionality reduction, label projection