Quick way to sum array elements based on flag in another array?

Say I have 2 arrays: one with "clean" data, and one that tells me which data gets lost, as indicated by zeros. If an element is lost, it should be added to the next observed element. Anybody got a quick one-or-two-liner to come up with the summed array?

Example:

clean   = [1 2 3 4 5 6]
lost    = [1 0 3 0 0 6]
desired = [1 5 15]

[Sean de Wolski]

Fun question, well asked with example inputs and expected outputs +1

Answer 1

[Stephen23]

Try this:

tmp = (1:numel(lost))-[0,cumsum(lost(1:end-1)==0)];
desired = accumarray(tmp.',clean).'

Tested:

>> clean   = [1 2 3 4 5 6];
>> lost    = [1 0 3 0 0 6];
>> tmp = (1:numel(lost))-[0,cumsum(lost(1:end-1)==0)];
>> desired = accumarray(tmp.',clean).'
desired =
     1     5    15
>> clean = [40 10 50 30 40 10 50 30 40 10 50 30];
>> lost  = [40  0 50 30  0 10  0  0 40 10 50 30];
>> tmp = (1:numel(lost))-[0,cumsum(lost(1:end-1)==0)];
>> desired = accumarray(tmp.',clean).'
desired =
    40    60    30    50   120    10    50    30

[Andrei Bobrov]

+1

Answer 2

[Andrei Bobrov]

desired = flip(accumarray(cumsum(lost(end:-1:1)~=0)',clean(end:-1:1)),1)';

or

desired = accumarray(cumsum(xor([0,diff(lost > 0)],lost))',clean)';

[Ted]

Thanks for the help. Try:

clean = [40 10 50 30 40 10 50 30 40 10 50 30];
lost  = [40  0 50 30  0 10  0  0 40 10 50 30];

It should give:

desired = [40 60 30 50 120 10 50 30];

Your solution does not.

[Andrei Bobrov]

corrected

Answer 3

[Sean de Wolski]

accumarray(interp1(find(lost),1:nnz(lost),1:numel(lost),'next')',clean)

[Ted]

Thanks for the help. Try:

clean = [40 10 50 30 40 10 50 30 40 10 50 30];
lost  = [40  0 50 30  0 10  0  0 40 10 50 30];

It should give:

desired = [40 60 30 50 120 10 50 30];

Your solution does not.

[Sean de Wolski]

Yes it does! Well it's transposed but the numbers are correct

clean = [40 10 50 30 40 10 50 30 40 10 50 30];
lost  = [40  0 50 30  0 10  0  0 40 10 50 30];
accumarray(interp1(find(lost),1:nnz(lost),1:numel(lost),'next')',clean)
ans =
  40
  60
  30
  50
 120
  10
  50
  30

You could always transpose the answer with '

[Andrei Bobrov]

+1

[Ted]

Sorry Sean, I must have fat-fingered the test!