Subtract column values where leading columns have values flipped (i.e., (A,B,C1) - (B,A,C2))

Suppose I have a table, T:

A B C

1 6 10

2 5 20

3 4 30

4 3 45

5 2 6

6 1 10

I want to subtract each entry in column C where the A and B values are flipped, and put this value into column D

For example, the first row (1, 6, 10) - the last row (6, 1, 10); (2, 5, 20) - (5, 2, 6) and so on

So D should be:

0

14

-15

15

-14

0

Thanks a lot!

Edit: (i) Flipped rows could occur anywhere in the table; (ii) there may be repeated values in Α and B

[John D'Errico]

Define "flipped". The exmple you show does not suggest what it might mean.

[Torsten]

M = [1	6	10
    2	5	20
    3	4	30
    4	3	45
    5	2	6
    6	1	10];
D = M(:,3)-flipud(M(:,3))

D = 6x1

0 14 -15 15 -14 0

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

[Neel Kulkarni]

The first row has (A, B, C) as (1, 6, 10). The last row has the first two values flipped, i.e., (6, 1, 10). Similarly between the second and penultimate row the first two values are flipped: 2, 5, 20 and 5, 2, 6. I would like the code to identify such rows and subtract the C column values from each other (so 10-10 = 0; 20-6 = 14)

[Image Analyst]

Is the flipped row always at numRows-row+1, like 1 and 6, 2 and 5, 3 and 4, or might the "flipped row" occur anywhere, at any row?

[Torsten]

@Image Analyst

Seems it can occur anywhere - otherwise my suggested solution would work.

[Neel Kulkarni]

@Torsten thanks for your solution; indeed it works if the example is stuck to, but my actual, larger dataset is not as orderly unfortunately. The flipped rows occur anywhere, and there are repeated values in A and B columns as well (forgot to add this in the original question, have added it as an edit just now). The (A,B) couple itself is flipped, not merely that B is A flipped upside down. Apologies that the original question was not clear about this.

[Torsten]

Please give an example for

there are repeated values in A and B columns as well

and how this case is handled.

[Neel Kulkarni]

@Torsten

A B C D

1 6 4 -6

2 6 9 3

2 5 10 -3

3 6 10 No flipped row (6,3) so preserve C value (=10)

4 3 45 No flipped row (3,4), preserve C value (=45)

5 2 13 3

6 2 6 -3

6 1 10 6

[Image Analyst]

Do you have a table variable, or a double variable? Can you attach your variable in a mat file?

save('answers.mat', 'T');

When you say "flipped" do you mean vertically or horizontally? Your original example seemed to have it both ways. Now it seems just that you mean horizontally, so that columnA is in columnB and ColumnB is in columnA, so in essence for a particular row, colA and colB are swapped.

If you have any more questions, then attach your data and code to read it in with the paperclip icon after you read this:

TUTORIAL: How to ask a question (on Answers) and get a fast answer

[Neel Kulkarni]

@Image Analyst Hi, it is a table variable. I have attached the mat file with my example above. It indeed is a horizontal flip.

Answer 1

[dpb]

load answers T
T

T = 8x3 table
    A    B    C 
    _    _    __

    1    6     4
    2    6     9
    2    5    10
    3    6    10
    4    3    45
    5    2    13
    6    2     6
    6    1    10

T.D=nan(height(T),1);
for i=1:height(T)
  ix=find(all(fliplr(T{:,1:2})==T{i,1:2},2));
  try
    T.D(i)=T.C(i)-T.C(ix);
  catch
  end
end
T

T = 8x4 table
    A    B    C      D 
    _    _    __    ___

    1    6     4     -6
    2    6     9      3
    2    5    10     -3
    3    6    10    NaN
    4    3    45    NaN
    5    2    13      3
    6    2     6     -3
    6    1    10      6

The above is the "dead ahead" solution that computes the difference for every record in the file; one could be a little more clever and set both i and ix (to the negative difference) in the one loop iteration. Then one would need to make and update the list of records yet to be handled to know when to quit. The list would start initialized to 1:height(T) and each i, ix would be removed each iteration. When empty, quit.

The above does nothing when there is no match, thereby leaving the initial default NaN value; your choice as to what to do here if this isn't the desired result. Your initial example had no unpaired records and your followup didn't define what is the desired result for those records. In this case, the catch clause would have to remove those i entries. Whether shortening the number of iterations would "win" over the extra logic probably would depend upon how large T is.