MATLAB: Does “mapstd” returns unexpected dimensions when I apply it to a new sample data

Deep Learning Toolboxmapstd

Why does "mapstd" returns unexpected dimensions when I apply it to a new sample data?
I have 4 sample data, each containing 2 predictor variables:
 
>> X = [ 2 1;
         5 0;
         3 0;
         4 2];
I standardize this using "mapstd" as follow:
 
>> [Xnew, PS] = mapstd(X);
However, when I try standardizing a single new sample data "Xtest", it produces a 4×2 array instead of 1×2 array:
>> Xtest = [2 3];
>> XtestNew = mapstd('apply', Xtest, PS)
XtestNew =
                     0.7071    2.1213
                    -0.1414    0.1414
                     0.2357    0.7071
                    -0.7071         0

Best Answer

The "mapstd" function normalizes the input data row-wise (horizontally). Therefore, with your current implementation, you are actually normalizing each sample individually since you are putting your data one above the other in the following format (4 rows x 2 columns):
   X =
                [ sample1
                  sample2
                  sample3
                  sample4 ]
 In order to use "mapstd" function to normalize each of your 2 predictor variables, you would need to store your data in the following format (2 row x 4 columns):
   X = [ sample1  sample2 sample3 sample4]
 Then, you can use "mapstd" function and get 2 means and 2 standard deviations (one for each predictor variable).
 Making the necessary modification to the original data:
 
>> X = X';   % store sample column by column instead
>> [Xnew, PS] = mapstd(X);   
>> XtestNew = mapstd('apply', Xtest, PS);
Related Question