R package "tsne" and "Rtsne" give different cell clustering results

Question

[Elaine.7]

I was doing cell clustering for single cell analysis and found these two R packages to do t-SNE clustering. According to the description, package "tsne" is a "pure R" implementation of the t-SNE algorithm, while package "Rtsne" is an R wrapper around the fast t-SNE implementation by Van der Maaten. In addition, when I run "Rtsne", I set parameter theta=0.0 which uses a standard t-SNE implementation. I applied these two functions to the same data set with same parameters, but the cell clustering results were not the same. In detail, "Rtsne" clustered two cell types together, while "tsne" completely separated these two cell types. Why do these two functions give such different results?

Answer 1

[igor]

Apparently, there may have been a bug in the tsne package, which may or may not be related to what you are seeing. Check the discussion here to see if it applies: https://gist.github.com/mikelove/74bbf5c41010ae1dc94281cface90d32