SVD implementation map reduces

Question

Hi, I need to do a decomposition of a singular value on large dense square matrices using Map Reduce.

I already checked the Mahout project, but what they provide is the TSQR algorithm http://arbenson.imtqy.com/portfolio/Math221/AustinBenson-math221-report.pdf .

The problem is that I want a full rank, and this method does not work in this case. The implementation of the distributed Lanczos SVD, which they used before it did not fit my case.

I found that TWO-SIDED JACOBI SCHEME can be used for this purpose, but I could not find any available implementation.

Does anyone know where and where I can find the reference code?

Answer 1

If this can help, look at the spark spark lib (mlib). It was an implementation. You can use it or look at it, you can make your own. https://spark.apache.org/docs/latest/mllib-dimensionality-reduction.html