pgrew / mbb

🤰🏽 make 🤰🏿 build 🤰🏻 brain
0 stars 1 forks source link

Understand all linear algebra required for PSVM #5

Closed pgrew closed 7 years ago

pgrew commented 7 years ago

The PSVM paper, https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/34638.pdf, may use linear algebra operations or algorithms that are not currently implemented in Chapel. We should list every required operation and algorithm to compare with what Chapel supports now.

pgrew commented 7 years ago

@ben-albrecht We need matrix multiplication, matrix division, matrix addition, matrix inverses, the "vec" transform, diagonal extraction, and scalar addition or subtraction from matrices.

ben-albrecht commented 7 years ago

Check-listified:

Division and inverses are coming with LAPACK support, and are straightforward to add.

Are these operations used to perform the incomplete Cholesky decomposition or will we need a routine for that too?

pgrew commented 7 years ago

Most of these are used in the parallel interior-point method. From what I can tell, the parallel incomplete Cholesky factorization doesn't use much linalg. Oh, add "trace" to your list :)