INSERTROWS (v2.0, may 2008)

Insert rows into a matrix at specific locations

Description

INSERTROWS - Insert rows into a matrix at specific locations

C = INSERTROWS(A,B,IND) inserts the rows of matrix B into the matrix A at the positions IND. Row k of matrix B will be inserted after position IND(k) in the matrix A. If A is a N-by-X matrix and B is a M-by-X matrix, C will be a (N+M)-by-X matrix. IND can contain non-integers.

If B is a 1-by-N matrix, B will be inserted for each insertion position specified by IND. If IND is a single value, the whole matrix B will be inserted at that position. If B is a single value, B is expanded to a row vector. In all other cases, the number of elements in IND should be equal to the number of rows in B, and the number of columns, planes etc should be the same for both matrices A and B.

Values of IND smaller than one will cause the corresponding rows to be inserted in front of A. C = INSERTROWS(A,B) will simply append B to A.

If any of the inputs are empty, C will return A. If A is sparse, C will be sparse as well.

[C, RA, RB] = INSERTROWS(...) will return the row indices RA and RB for which C corresponds to the rows of either A and B.

Examples:
  % the size of A,B, and IND all match
    C = insertrows(rand(5,2),zeros(2,2),[1.5 3])
  % the row vector B is inserted twice
    C = insertrows(ones(4,3),1:3,[1 Inf])
  % matrix B is expanded to a row vector and inserted twice (as in 2)
    C = insertrows(ones(5,3),999,[2 4])
  % the whole matrix B is inserted once
    C = insertrows(ones(5,3),zeros(2,3),2)
  % additional output arguments
    [c,ra,rb] = insertrows([1:4].',99,[0 3])
    c.' % -> [99 1 2 3 99 4]
    c(ra).' % -> [1 2 3 4]
    c(rb).' % -> [99 99]
 
Using permute (or transpose) INSERTROWS can easily function to insert columns, planes, etc:
  % inserting columns, by using the transpose operator:
    A = zeros(2,3) ; B = ones(2,4) ;
    c = insertrows(A.', B.',[0 2 3 3]).' % insert columns
  % inserting other dimensions, by using permute:
    A = ones(4,3,3) ; B = zeros(4,3,1) ;
    % set the dimension on which to operate in front
    C = insertrows(permute(A,[3 1 2]), permute(B,[3 1 2]),1) ;
    C = ipermute(C,[3 1 2])
 
See also horzcat, reshape, cat

(version 2.0, may 2008)

• Acknowledgements: This submission has inspired the following:; Place one 2D matrix inside another, PWLINT: Piecewise Linear Integration
• MATLAB release: MATLAB 6.5 (R13)

[John D'Errico]

This should have been part of matlab long ago. Yes, use of [] can do this too, but this is a better solution.
Nice help, good error checking, its friendly and expands a scalar into a row. Many users will thank you for making what should be a trivial task exactly that.

[Jiro Doke]

Very nicely done. It is very easy to use because the syntax intuitively makes sense. Thanks.

[Jos van der Geest]

If you are looking for a ND insert function, take a look at INTERLACE ("insert subarrays into array") by Duane Hanselman

[Scott Burnside]

Helpful. Nice code. I've stopped writing these and just use this.

[Jenn Jackson]

Glad I checked here first, totally saved me the time.

[Pratik Shah]

Saved a lot of time. Great job!!!

[Tim Davis]

Cool! Very fast for large sparse matrices. It even keeps C sparse if A is sparse and B is full, as it should.

A few comments about the code. Why do you have the statement [abi,abi] = sort ( ... ) ? Which abi does MATLAB return? The first or 2nd argument? Looks like the 2nd. It would be more clear to write this as:

[ignore,abi] = sort ( ... ) ;
clear ignore ;

The "nargout" test is a little odd. It should read "if (nargout > 1)". Right now,
C = insertrows( ... ) is always computing RA and RB and then discarding them.

For the nargin test, I would suggest allowing C = insertrows(A,B), which just does [A;B] and returns - but optionally returning RA and RB too. It would expand B if B was a scalar, too.

[Jos (the author)]

Version 2.0 incorporates the improvements and suggestions mentioned by Tim Davis.

[Adnan Abdulally]

Hello, i came across this program and i am seeing an increase in time vs matlabs appending function. I am looping 500 times and constantly appending AS to A, where A is originally declared as a sparse matrix of zeros size(1,n). AS is full, but is approximately 30% non-zero.

Here are the results from profiler when i have them running one after the other:

156.69s 500 66 A =insertrows(A,AS);
74.28s 500 67 B = [A;AS];

Any reason why insertrows takes longer when it claims to be faster? Am i doing something wrong to make it inefficient?

[Jos (the author)]

Dear Adnan Abdulally, your call "insertrows(A,AS)" is lacking a third input argument ... but anyways, it shouldn't be suprising that [A ; B] is faster than insertrows(A,B,size(A,1)).

Please need keep in mind that insertrows was not designed to simply concatenate two matrices. Let vertcat handle that!

[Rubi]

excellent file n.n thanks a lot, i had a terrible headache trying to use a for with this t=[t(1,:);[1 1];t(2:end,:)] !!

[Rody Oldenhuis]

I agree with John; this should be part of standard matlab. Well documented, short & simple, vectorized and ample documentation & comments; excelent.

[Dips Bhatia]

Hi Jos,
Really good. Thanks

[happydude]

maybe i am running this wrong, but if I want to replace two adjact rows I seem to have problems

For example lets say I want [1, 2, 99, 99, 3, 4,]
But my original arry is [1:4]

I would have assumed you would do:
insertrows([1:4].',99,[2 3])

In face this gives the wrong answer, you would need to do:
insertrows([1:4].',99,[2 2])

This seems counterinutive? would be obliged for you help if I am missing something.

Thanks!

[Jos (10584)]

@happydude: the indices in the 3r argument refer to the positions in the original array, not in the final array, as you did assume. Since the end-user is always king, I'll have to make this more clear in an update.

[Martin]