Is there a function for argmax?

I am looking for a function that selects the argmax value from a vector.

    aux = argmax(P);      %P is a normal vector

Supposedly there should a function directly called argmax, but is not working (error = Undefined function or variable 'argmax'). Is there any other function? (I am working with the 2017a version).

[Geoff Hayes]

Javier - have you tried using max?

[Stephen23]

"Supposedly there should a function directly called argmax"

I can't find any reference to a function called argmax in the MATLAB documentation:

https://www.mathworks.com/help/search.html?qdoc=argmax

Can you please provide a reference for argmax.

[Javier Biera]

Yes, max works, I only need to add a find afterwards. max gives me the maximum value of the vector P. And with a find function I can get the position of the maximum value. Thank you.

[Anton Semechko]

Use syntax

[P_max,id_max]=max(P);

Here, id_max is the index of P_max in P, so that P_max == P(id_max)

[Stephen23]

@Javier Biera: did you read the max documentation? It's second output is an index:

[~,idx] = max(...)

Answer 1

[Image Analyst]

It's probably some function that someone wrote, like the person who gave you the code. Ask them for it. Otherwise, can you tell us how aux is used? That may give us some clue as to what argmax gives. Like it really should be nargin (the number of input arguments) or the max value of some matrix inputted into the argument list or who knows what.

Answer 2

[Alan]

function y = argmax(x)
  [~,y] = max(x);
end

[Walter Roberson]

function y = argmax(x)
    y = find(x == max(x));
end

is closer. But really the meaning needs to be pinned down for the case where x is multidimensional and might contain duplicates.

[Alan]

That is a step more comprehensive, but to cover all the edge cases would need a lot more work. In my opinion, there should be a standard MATLAB function for this.

[Walter Roberson]

Easier I suspect to return

y == max(x, [], dim);

which would be a logical the same size as the original. That removes the need to worry about different sizes of output for different columns.