How to vectorize SOC constraints

[Tomasz Gorecki]

Hello, in an attempt to speed up the Yalmip parsing I was trying to vectorize as much as possible my constraints declarations.

I have a bunch of SOC constraints of the form

for i = 1:n
norm(Z(i,:),2) <= z(i)
end

I tried also the manual version

Z(i,:)*Z(i,:)' <= z(i)

which was quite slower but was wondering if there is a way of getting away without for loops, or what is the best practice for this

Best and thanks in advance,

Tomasz

[Johan Löfberg]

cone([z(:)';Z'])

[Tomasz Gorecki]

Great thx.
Does something similar exist for other cones (inf, 1-norm, sdp ?)

[Tomasz Gorecki]

I spoke too fast,

I tried:
cone(z(:)',Z')

but got the error

Error using subsindex
Function 'subsindex' is not defined for values of class 'sdpvar'.

the input arguments have dimension 1x404 and 30x404 respectively

[Johan Löfberg]

You missed the brackets.

And no, this is only for 2-norm, as inf and 1-norm are trivially modelled manually if max performance is needed.

[Johan Löfberg]

and with dimensions the data is simply [z;Z]

[Johan Löfberg]

and the error message is because you have a constant variable called cone

[Tomasz Gorecki]

I tried different arragement of data but cannot get past this error. Even trying just

z = sdpvar(10,10,'full')
Linear matrix variable 10x10 (full, real, 100 variables)
cone(z)

still gives

Error using subsindex
Function 'subsindex' is not defined for values of class 'sdpvar'.

Sorry if I am doing something idiotic

Best

[Tomasz Gorecki]

ah yes sure that was it. I had a cone variable.