Choosing algorithm for SVD

General Usage
,
1

The svd function in LinearAlgebra package has an option to select the algorithm for factorization method. But could not find proper way to use it.

using LinearAlgebra
A= rand(Float32, 10, 20);
F = svd(A, full=true) # works fine 
F2 = svd(A, full=true, alg=QRIteration()) # does not work

Asks me to use LinearAlgebra.Algorithm and when I use it :

F2 = svd(A, full=true, alg=LinearAlgebra.QRIteration())
#throws and error saying

ERROR: TypeError: in keyword argument alg, expected LinearAlgebra.Algorithm, got Type{LinearAlgebra.QRIteration}

What is the proper use of another algorithm for svd ?
Yet another question is can svd be used with CUDA.jl ?

B.R.

2

This works for me with Julia 1.5.3. What version of Julia are you using?

2 Likes
3

versioninfo() shows 1.5.2 . Probably, I didn’t write it as a function call QRIteration().

Thaks

4

Is there a reason we have to write the function with () at the end? This seems inconsistent with the way other functions are passed as arguments, e.g. sum(exp, rand(5)).

1 Like
5

Is there a reason we have to write the function with () at the end? This seems inconsistent with the way other functions are passed as arguments, e.g. sum(exp, rand(5)) .

nice point

6

LinearAlgebra.QRIteration() is a call to a constructor (for the QRIteration type), it’s not a function like exp.

4 Likes