[GitHub] incubator-spark pull request: Principal Component Analysis

[rezazadeh]

Github user rezazadeh commented on the pull request:

    https://github.com/apache/incubator-spark/pull/564#issuecomment-34585044
  
    Yes I agree, densifying the matrix for sake of centering is a problem for 
sparse matrices. I can solve it, but with appropriate sparse vector support. 
Once we have the sparse vector primitives in mllib I will change this to use 
those and avoid densifying. It can be done by constructing A^TA and centering A 
all in one step (A^TA is n x n and small, assuming n is small).
    
    Mentioned before, sparse vector primitives are tracked by 
https://spark-project.atlassian.net/browse/MLLIB-18