how to use function handles

Instead of the following, I want to use function handles and move if-statments out of the loop:

for n=1:10
    if a==0
        A = function2(n,a,b);
    else
        A = function1(n,a,b,c,d);
    end
end

function A = function1(n,a,b,c,d)
A = n+a+b+c+d; % here it can be any expression
end

function A = function2(n,a,b)
A = n*a*b; % here it can be any expression
end

moving if-statement outside of the loop and using handles:

if a==0
    func = @function2;
else
    func = @function1;
end

and then I can solve my problem in two ways as

for n=1:10
    A = func(n,a,b,c,d);   
end
    
function A = function1(n,a,b,c,d)
A = n+a+b+c+d;
end

function A = function2(n,a,b,~,~)
A = n*a*b;
end

or as

for n=1:10
    A = func(n,a,b,c,d)   
end
    
function A = function1(n,a,varargin)
b = varargin{1};
c = varargin{2};
d = varargin{3};
A = n+a+b+c+d;
end

function A = function2(n,a,varargin)
b = varargin{1};
A = n*a*b;
end

Is there any more elegant way to do this?

[G A]

may be this way could be better:

function A = function1(n,S)
A = n + S.a + S.b + S.c + S.d;
end
function A = function2(n,S)
A = n*S.a*S.b;
end

Answer 1

[Yongjian Feng]

You don't need varargin

function A = function1(n,a,b, c, d)
A = n+a+b+c+d;
end
function A = function2(n,a,b, ~, ~)
A = n*a*b;
end

[G A]

That is the one way - without varargin. My question was - is any way better than this, i.e. without matching the number of arguments?

[Stephen23]

This is the best approach. In contrast when you use VARARGIN you lose the clarity of this function signature, which:

• shows the variable names in MATLAB Editor (e.g. tab completion / pop-up help)
• clearly indicates which variables are not used within the function

[G A]

Stephen, what about the approach with using struct (see my other comment)?

[Stephen23]

@G A: using a structure is also good approach. Which approach to use depends on many things, you will have to find the right balance/optimum for your project and needs, e.g.:

structure:

• useful when you have many arrays (e.g. more than 6 or 7) which easily lead to confusion when supplied as positional input arguments (we regularly get questions on this topic).
• might be easier to process when those arrays are optional

input arguments:

• (partially) self-documentating names
• pop-up help in the Editor, etc.

I am sure that other users on this forum can add other benefits or considerations.

[G A]

Thanks, Stephen! One question: what is the main difficulty to make (or reason not to make) it possible in the Editor to higlight structure name with its field throughout a function like it occurs with simple variables when you click on it or select it? Is this question to go to "What frustrates you about Matlab"?

[Stephen23]

"what is the main difficulty to make (or reason not to make) it possible in the Editor to higlight structure name with its field throughout a function like it occurs with simple variables when you click on it or select it?"

You have to consider that variables in MATLAB are dynamically typed. In general it is difficult to determine the type of a variable until runtime (e.g. changes to function scoping, variables created dynamically, inputs are not typed, etc.)

[G A]

Thanks! - and I will accept this answer together with comments.

Answer 2

[the cyclist]

I think the basic idea you need is

f1 = @(x) x;
f2 = @(x) x.^2;
f = @(a,x) (a~=0)*f1(x) + (a==0)*f2(x);
f(0,2)

ans = 4

f(1,2)

ans = 2

[G A]

How would you apply your idea to the non-anonimous functions with different number of input arguments which must be passed within for-loop? The other question is: when a = 0, will Matlab execute function f2 to multiply its return by 0 or, seeing that one term is 0, it will return 0 without execution of f2?

[the cyclist]

Hm. Not finding a good way to handle this that does not potentially result in empty, Inf, or NaN when one of the functions does not have all the arguments. For example, you could pass zeros into the function when they are not needed, but it is potentially hazardous:

f1 = @(x) x;
f2 = @(x,y) x.^2 + y.^2;
f = @(a,x,y) (a~=0)*f1(x) + (a==0)*f2(x,y);
f(0,2,3)

ans = 13

f(1,2,0)

ans = 2

[G A]

Also, it is time consuming if two functions are executed instead of one. Better to put if-statement into a loop. I have half of answer to my question in my comment above: 0*inf = NaN, 0*A = 0, so to give the answer 0 or NaN, Matlab must execute f2 before multipliing its return by 0.

[the cyclist]

Searching keywords such as conditional anonymous function MATLAB turns up a lot of these same ideas. I did not find a great solution.