Is there a way to use the RMS function with two options, namely 'omitnan' and an option for the dimension?

Take the following matrix as example.

A=repmat([NaN 1 2 3],3,1);
rms(A)

ans =

NaN 1 2 3

rms(A,2)

ans =

NaN

NaN

NaN

There is an 'omitnan' option for rms. E.g.

rms(A','omitnan') % A' is the flipped matrix

ans =

2.1602 2.1602 2.1602

But I can not do both options/inputs at the same time, i.e.

rms(A,2,'omitnan')
%or
rms(A,'omitnan',2)

Both throw the error:

Error using rms

Too many input arguments.

Now, of course I could use a workaround like this:

my_rms=rms(A','omitnan')'; % A' is the flipped matrix; 
% note that the outcome of rms() is also flipped

But that seems ugly and also confusing (for my future self) to me. Another workaround (even uglier) would be to for-loop over every single row in A with

for i = 1:3
my_rms(i)=rms(A(i,:),'omitnan');
end

I'm aware that I'm kind of answering my own question by giving work arounds, but as I said, they do not seem nice and clean to me, so I hope someone might have a better idea.

Answer 1

[Jan]

There is no documented 'omitnan' argument for rms in current Matlab versions, see: https://www.mathworks.com/help/signal/ref/rms.html (link). I get:

A=repmat([NaN 1 2 3],3,1
rms(A, 'omitnan')
>> [NaN, 1, 2, 3]

But I observe this:

rms(A')
[NaN, 1, 2, 3]
rms(A', 'omitnan')
[2.1602, 2.1602, 2.1602]

This happens, because the dim argument is forwarded to mean without a check inside rms.m:

y = sqrt(mean(x .* x), dim);

If you want the 'omitnan' and the specified dimension, simply use an expanded rms version:

function y = rms2(x, varargin)
if isreal(x)
  y = sqrt(mean(x .* x, varargin{:}));
else
  absx = abs(x);
  y    = sqrt(mean(absx .* absx, varargin{:}));
end
end

Add documentation, tests of inputs and number of inputs in addition, as well as the more efficient method to get the absolute value for imaginary input.

If you want to, send an enhancement request to MathWorks, to catch the 'omitnan' flag explicitly.

[JD4]

Hi Jan,

I know it is not in the documentation.

Of course,

rms(A,'omitnan')

does yield the same as

rms(A)

but if you flip A, i.e.

B=A'

ans =

NaN NaN NaN

1 1 1

2 2 2

3 3 3

and then try

rms(B)

you get

ans =

NaN NaN NaN

but with

rms(B,'omitnan')

you get

ans =

2.1602 2.1602 2.1602

[JD4]

Your edit was faster than my answer :D

Thanks!

Answer 2

[Matt J]

Here's a better workaround. In this implementation, all the same input syntax options as for mean() are available.

function out=rms(A,varargin)

    out=sqrt(mean(A.^2,varargin{:}));

end