Multi Dimensional Scaling (MDS) and Principal Coordinate Analysis (PCoA) are two names for the same dimension reduction technique*.

In scikit-learn, MDS is implemented with the SMACOF algorithm while in other Python libraries (such as scikit-bio) and most R packages offering it, it is implemented using singular value decomposition.

This is usually quite confusing for people who try out the sklearn implementation of MDS when comparing it with other MDS implementations (sklearn stands out).

How could one add another implementation of MDS in sklearn ? Or maybe create a new PCoA method ?

cc @adrinjalali

• Issue #579 being discussed in the scikit-bio repository
• StackOverflow thread
• *MDS vs PCoA