replace elements of a vector

Hi,

Is there a possibility to replace the zeros in A by the last non zero observation such as this case:

A=[1 0 2 2 8 0 0 0 3 5 7]

B=[1 1 2 2 8 8 8 8 3 5 7]

Thank you

[rithy vorn]

hell0, could it be possible to do like this A=[1,2,3,4,5] to B=[5,4,3,2,1}

Answer 1

[Sean de Wolski]

A=[1 0 2 2 8 0 0 0 3 5 7];
B = A; %copy

idx = logical(A); %locations to keep
Anr = A(idx); %locations to keep
idr = cumsum(idx); %where to replace?
idx = ~idx; %only do this once
B(idx) = Anr(idr(idx)) %replace

This will only run into issues if the first element is zero.

[Jan]

Let me repeat again, that this needs twice as long as Fangjun's loop.

Answer 2

[Fangjun Jiang]

What's wrong with using a simple, old, straightforward for-loop? Jan will help me prove that it's faster than other approaches.

A=[1 0 2 2 8 0 0 0 3 5 7];
if A(1)==0
    error('invalid data');
end
for k=2:length(A)
    if A(k)==0;
        A(k)=A(k-1);
    end
end

Or an even better solution

A=[1 0 2 2 8 0 0 0 3 5 7];
if A(1)==0
    error('invalid data');
end
Ind=find(A==0);
for k=Ind
    A(k)=A(k-1);
end

[Jan]

+1: Fastest Matlab method. You find the measurements in my answer.

[Tiina]

Thank you works fine.

[Fangjun Jiang]

@Jan, Thank you for backing me up.

I can think of another improvement. See update.

Answer 3

[Sven]

Hi Tiina, try this. It uses find and arrayfun to replace the zero-values of A with the nearest preceding non-zero value:

zeroInds = find(A==0);
repVals = arrayfun(@(ind)A(find(A(1:ind),1,'last')), zeroInds);
A(zeroInds) = repVals;

Note that if the first element of A is a zero, then the question itself hits a problem, but the above works fine in other cases.

[Walter Roberson]

You seem to have missed the closing bracket on the arrayfun, and you missed the parameter indicating the array that is to be processed. I suspect the line should be

repVals = arrayfun(@(ind)A(find(A(1:ind),1,'last'), zeroInds);

[Sven]

Yep, thanks Walter... the laws of MATLAB Answers:

Rule 1. IF you type an answer without running the code in MATLAB, you WILL make a typo.

Rule 2. IF you start your answer in the browser, remember Rule 1, then check your answer in MATLAB, you WILL forget to copy it back into the browser :)

[Sean de Wolski]

Sven, that should be in the tutorial!

[Jan]

Rule 3. All the small dots and crumbs, quotes and the tails of curly braces looks like flyspecks when you observe the monitor from a certain distance.

Rule 4. If you are in doubt, they are flyspecks.

[Sven]

Rule 5. If you're in a hurry and just copy Walter's comment code, HE WILL make a typo :)

(now *that* was unexpected)

Answer 4

[Jan]

A = rand(1, 1e6);
A(rand(size(A)) < 0.1) = 0;   % 10% zeros

tic; for i = 1:20; B = F1(A); end; toc

Some timings (Matlab 2009b/64, Win7, Core2Duo):

function A = F1(A)  % ======================================
n = A~=0;
A(~n) = cell2mat(arrayfun(@(x,y)A(x)*ones(1,y-x),...
                 strfind(n,[1, 0]),strfind([n, 1],[0, 1]),'un',0))

20.528223 sec

function A = F2(A)   % ======================================
nonZero = (A ~= 0);
B       = A(nonZero);
A       = B(cumsum(nonZero));

1.198049 sec

function A = F3(A)   % ======================================
idx = logical(A); %locations to keep
Anr = A(idx); %locations to keep
idr = cumsum(idx); %where to replace?
idx = ~idx; %only do this once
A(idx) = Anr(idr(idx)); %replace

1.151062 sec

function A=F4(A)     % ======================================
if A(1)==0
  error('invalid data');
end
for k=2:length(A)
  if A(k)==0
    A(k)=A(k-1);
  end
end

0.615451 sec

#include "mex.h"     // ====================================
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
   double *A;
   mwSize i, n;
   //
   n       = mxGetNumberOfElements(prhs[0]);
   plhs[0] = mxDuplicateArray(prhs[0]);
   A       = mxGetPr(plhs[0]);
   //
   if (A[0] == 0.0) {
      mexErrMsgIdAndTxt("JSimon:FillZeros:BadData", "Input starts with 0.");
   }
   for (i = 0; i < n; i++) {
      if (A[i] == 0.0) {
         A[i] = A[i - 1];
      }
   }
   return;
}

0.210786 sec

The loop ist the fastest Matlab method, because it does not need temporary memory. Another hint to follow the KISS rule: Keep it simply stupid.

But still C is faster in loops.

[Fangjun Jiang]

+1, I'd better learn C!