Scaling the natrix in specific range

I have a 300x1 matrix

where the minmum value is

   -2.1188e+003

maximum is

    460.3171

i want to scale in region from -1 to 1

Answer 1

[Nasser M. Abbasi]

Since I just wrote the code, might as well post it. A quick function to scale it

function C = nma_rescale(A,new_min,new_max)
%Nasser M. Abbasi 011212
%NO ERROR CHECKING DONE ON INPUT. Rescale a matrix or a vector A

current_max = max(A(:));
current_min = min(A(:));

C =((A-current_min)*(new_max-new_min))/(current_max-current_min) + new_min;

end

to use:

EDU>> C=rand(3);
EDU>> B=nma_rescale(C,-1,1)

B =

     -0.7625    0.8960   -0.2033
     -0.3188   -0.8938   -1.0000
     -0.8317    1.0000   -0.3316

[kash]

I have some values of 300x1 matrix A= -2.1188e+003 to 460.3171 and other values B= -1 to 1 ,i want to scale A in range betweem -1 to 1 corresponding to values in B,please help

[Nasser M. Abbasi]

I am not sure I understand. To scale a matrix A to be between -1 and 1, just type

C = nma_rescale(A,-1,1)

I do not understand what B role is here. May be you can update your question and make it more clear what is needed

[Walter Roberson]

The code in my Answer does that, at least when given vectors (the problems statement involved vectors.) But before you go further you should decide whether the negative range is to be scaled independently of the positive range (as in my first solution), or if the entire range is to be treated as a linear block to be re-centered at the average of the max and min (as in my second solution.)

[Dario Dematties]

Hello, with respect to the solution proposed by Nasser M. Abbasi. What should I do when current_max=current_min? What is the conventional way to solve this problem?

[Jan]

@Dario: There is no convention. It depends on your problem, if the result of a scaled scalar is -1, 0, +1 or Inf, perhaps NaN.

[Dario Dematties]

Thanks Jan, Does anyone know which is the reaction of libsvm software on matlab when it is fed with vectors with missing data (NaN)?

Answer 2

[Walter Roberson]

Does 0 need to remain 0, or should it become the midpoint of your [-2118.8 to 460.317] range?

0 remains 0:

matmin = min(mat);
matmax = max(mat);
newmat = mat;
newmat(mat < 0) = mat(mat < 0) ./ abs(minmat);
newmat(mat > 0) = mat(mat > 0) ./ maxmat;

Midpoint becomes 0:

matmin = min(mat);
matmax = max(mat);
newmat = 2 * (mat - matmin) ./ (matmax - matmin) - 1;