Create a special diagonal matrix - matlab

I want to create a special diagonal matrix in MATLAB that
uses [1 1] or any other array as main diagonal elements.
Something like the following:
[1 1 0 0 0 0
0 0 1 1 0 0
0 0 0 0 1 1]
How can I do that without using any loop structures?

Let
v = [1 1];
n = 3;
Using kron:
result = kron(eye(n), v);
Using blkdiag:
vv = repmat({v}, 1, n);
result = blkdiag(vv{:});

Related

Writing a loop that works for any value of N

I have a code for a 1D heat equation. Im trying to format a for loop so that the A matrix will follow a certain pattern of 1 -2 1 down the entire diagonal of a matrix that could be infinite. The pattern starts to take shape when I mess around with the initialized count at the beginning of the for loop but this changes the size of the matrix which fails the rest of the code.
My current code is below. The commented A matrix edits are what it should be.
N = 5;
%A(2,1:3) = [1 -2 1];
%A(3,2:4) = [1 -2 1];
%A(4,3:5) = [1 -2 1];
%A(5,4:6) = [1 -2 1];
A = zeros(N+1,N+1);
A(1,1) = 1;
for count=N:N+1
A(count+1,count:count+2) = [1 -2 1];
end
A(N+1,N+1) = 1;
In Matlab you can often avoid loops. In this case you can get the desired result with 2D convolution:
>> N = 6;
>> A = [1 zeros(1,N-1); conv2(eye(N-2), [1 -2 1]); zeros(1,N-1) 1]
A =
1 0 0 0 0 0
1 -2 1 0 0 0
0 1 -2 1 0 0
0 0 1 -2 1 0
0 0 0 1 -2 1
0 0 0 0 0 1
Or, depending on what you want,
>> A = conv2(eye(N), [1 -2 1], 'same')
A =
-2 1 0 0 0 0
1 -2 1 0 0 0
0 1 -2 1 0 0
0 0 1 -2 1 0
0 0 0 1 -2 1
0 0 0 0 1 -2
There are many simple ways of creating this matrix.
Your loop can be amended as follows:
N = 5;
A = zeros(N+1,N+1);
A(1,1) = 1;
for row = 2:N
A(row, row-1:row+1) = [1 -2 1];
end
A(N+1,N+1) = 1;
I've renamed count to row, we're indexing each row (from 2 to N, skipping the first and last rows), then finding with row-1:row+1 the three indices for that row that you want to address.
Directly indexing the diagonal and off-diagonal elements. Diagonal elements for an NxN matrix are 1:N+1:end. This is obviously more complex, I'd prefer the loop:
N = 6;
A = zeros(N,N);
A(1:N+1:end) = -2;
A(2:N+1:end-2*N) = 1; % skip last row
A(2*N+2:N+1:end) = 1; % skip first row
A(1,1) = 1;
A(N,N) = 1;
Using diag. We need to special-case the first and last rows:
N = 6;
A = diag(-2*ones(N,1),0) + diag(ones(N-1,1),1) + diag(ones(N-1,1),-1);
A(1,1:2) = [1,0];
A(end,end-1:end) = [0,1];

Create a neighbourhood graph from a given binary matrix using matlab

I am trying to create a neighbourhood graph from a given binary matrix B. Neighbourhood graph (A) is defined as an adjacency matrix such that
(A(i,j) = A(j,i) = 1)
if the original matrix B(i) = B(j) = 1 and i and j are adjacent to each (left, right, up, down or diagonal). Here I used the linear subscript to access the original matrix B. For example, consider the below matrix
B = [ 0 1 0;
0 1 1;
0 0 0 ];
My A will be a 9 * 9 graph as given below
A = [ 0 0 0 0 0 0 0 0 0;
0 0 0 0 0 0 0 0 0;
0 0 0 0 0 0 0 0 0;
0 0 0 0 1 0 0 1 0;
0 0 0 1 0 0 0 1 0;
0 0 0 0 0 0 0 0 0;
0 0 0 0 0 0 0 0 0;
0 0 0 1 1 0 0 0 0;
0 0 0 0 0 0 0 0 0 ];
Since in the original B matrix, B(4), B(5) and B(8) are adjacent with corresponding entries 1, the adjacency matrix A has 1 at A(4,5), A(5,4), A(4,8), A(8,4), A(5,8) and A(8,5).
How can I create such an adjacency matrix A given the matrix B in an efficient way?
This doesn't require any toolbox, and works for square or rectangular matrices. It uses array operations with complex numbers.
Consider a binary matrix B of size M×N.
Create an M×N matrix, t, that contains the complex coordinates of each nonzero entry of B. That is, entry t(r,c) contains r+1j*c if B(r,c) is nonzero, and NaN otherwise.
Compute an M*N×M*N matrix, d, containing the absolute difference for each pair of entries of B. Pairs of entries of B that are nonzero and adjacent will produce 1 or sqrt(2) in matrix d.
Build the result matrix, A, such that it contains 1 iff the corresponding entry in d equals 1 or sqrt(2). Equivalently, and more robust to numerical errors, iff the corresponding entry in d is between 0 and 1.5.
Code:
B = [0 1 0; 0 1 1; 0 0 0]; % input
t = bsxfun(#times, B, (1:size(B,1)).') + bsxfun(#times, B, 1j*(1:size(B,2)));
t(t==0) = NaN; % step 1
d = abs(bsxfun(#minus, t(:), t(:).')); % step 2
A = d>0 & d<1.5; % step 3
To get B back from A:
B2 = zeros(sqrt(size(A,1)));
B2(any(A,1)) = 1;
Here is a solution using image processing toolbox* that creates sparse matrix representation of the adjacency matrix:
B = [ 0 1 0;
0 1 1;
0 0 0 ]
n = numel(B);
C = zeros(size(B));
f = find(B);
C(f) = f;
D = padarray(C,[1 1]);
%If you don't have image processing toolbox
%D = zeros(size(C)+2);
%D(2:end-1,2:end-1)=C;
E = bsxfun(#times, im2col(D,[3 3]) , reshape(B, 1,[]));
[~ ,y] = find(E);
result = sparse(nonzeros(E),y,1,n,n);
result(1:n+1:end) = 0;
*More efficient implementation of im2col can be found here.

Perform "outer product" of 2-D matrix and return a 3-D array in MATLAB

I would like to do an operation on a 2-D matrix which somehow looks like the outer product of a vector. I already have written some codes for this task, but it is pretty slow, so I would like to know if there is anything I can do to accelerate it.
I would like to show the code I wrote first, followed by an example to illustrate the task I wanted to do.
My code, version row-by-row
function B = outer2D(A)
B = zeros(size(A,1),size(A,2),size(A,2)); %Pre-allocate the output array
for J = 1 : size(A,1)
B(J,:,:) = transpose(A(J,:))*A(J,:); %Perform outer product on each row of A and assign to the J-th layer of B
end
end
Using the matrix A = randn(30000,20) as the input for testing, it spends 0.317 sec.
My code, version page-by-page
function B = outer2D(A)
B = zeros(size(A,1),size(A,2),size(A,2)); %Pre-allocate the output array
for J = 1 : size(A,2)
B(:,:,J) = repmat(A(:,J),1,size(A,2)).*A; %Evaluate B page-by-page
end
end
Using the matrix A = randn(30000,20) as the input for testing, it spends 0.146 sec.
Example 1
A = [3 0; 1 1; 1 0; -1 1; 0 -2]; %A is the input matrix.
B = outer2D(A);
disp(B)
Then I would expect
(:,:,1) =
9 0
1 1
1 0
1 -1
0 0
(:,:,2) =
0 0
1 1
0 0
-1 1
0 4
The first row of B, [9 0; 0 0], is the outer product of [3 0],
i.e. [3; 0]*[3 0] = [9 0; 0 0].
The second row of B, [1 1; 1 1], is the outer product of [1 1],
i.e. [1; 1]*[1 1] = [1 1; 1 1].
The third row of B, [1 0; 0 0], is the outer product of [1 0],
i.e. [1; 0]*[1 0] = [1 0; 0 0].
And the same for the remaining rows.
Example 2
A =
0 -1 -2
0 1 0
-3 0 2
0 0 0
1 0 0
B = outer2D(A)
disp(B)
Then, similar to the example 1, the expected output is
(:,:,1) =
0 0 0
0 0 0
9 0 -6
0 0 0
1 0 0
(:,:,2) =
0 1 2
0 1 0
0 0 0
0 0 0
0 0 0
(:,:,3) =
0 2 4
0 0 0
-6 0 4
0 0 0
0 0 0
Because the real input in my project is like in the size of 30000 × 2000 and this task is to be performed for many times. So the acceleration of this task is quite essential for me.
I am thinking of eliminating the for-loop in the function. May I have some opinions on this problem?
With auto expansion:
function B = outer2D(A)
B=permute(permute(A,[3 1 2]).*A',[2 3 1]);
end
Without auto expansion:
function B = outer2Dold(A)
B=permute(bsxfun(#times,permute(A,[3 1 2]),A'),[2 3 1]);
end
Outer products are not possible in the matlab language.

Conditionally replace neighbouring cells

Lets say I have a matrix A:
A =
0 1 0 0
0 0 0 0
0 0 0 1
0 0 0 0
And I want to create a new matrix B of the same dimension where all ones and accompanying neighbours are replaced by the following matrix:
X =
1 1 1
1 2 1
1 1 1
The 2 in matrix X should be placed 'on top' of the 1 values as to get:
B =
1 2 1 0
1 1 2 1
0 0 1 2
0 0 1 1
Values should be added up where elements overlap and matrix X should be 'cut off' in places where it extends the dimensions of matrix A/B The idea is to eventually replace X by a 2d gaussian distribution and matrix A will be large containing many more ones. So it's essential that the code is efficient and fast. This is the code i came up with:
A = [0 1 0 0;0 0 0 0;0 0 0 1;0 0 0 0]
X = [1 1 1;1 2 1;1 1 1]
B = zeros(4,4);
t=1;
indA = find(A==1);
indX = find(X==2);
all = find(X>0);
[iall jall] = ind2sub(size(X),all);
[ia ja] = ind2sub(size(A),indA)
[ix jx] = ind2sub(size(X),indX)
iv = ia-ix
jv = ja-jx
for t=1:numel(iv),
ib = iall+iv(t);
jb = jall+jv(t);
ibjb = [ib(:), jb(:)]
c1 = (ibjb(:,1)>4)|(ibjb(:,1)<1); c2 = (ibjb(:,2)>4)|(ibjb(:,1)<1);
ibjb((c1|c2),:)=[]
isel = ibjb(:,1)-iv(t)
jsel = ibjb(:,2)-jv(t)
B(ibjb(:,1), ibjb(:,2)) = B(ibjb(:,1), ibjb(:,2))+ X(isel, jsel)
t=t+1;
end
Is there a more efficient/faster way (minimizing the loops) to code this function?
What you want is a (2D) convolution. So use conv2:
B = conv2(A, X, 'same');

Permute elements within rows of matrix

I have a matrix A
A = [0 0 0 0 1; 0 0 0 0 2; 0 1 2 3 4];
and I would like to randomly permute the elements within each row. For example, matrix A2
A2 = [1 0 0 0 0; 0 0 0 2 0; 4 1 3 2 0]; % example of desired output
I can do this with a vector:
Av = [0 1 2 3 4];
Bv = Av(randperm(5));
But I am unsure how to do this a row at time for a matrix and to only permute the elements within a given row. Is this possible to do? I could construct a matrix from many permuted vectors, but I would prefer not to do it this way.
Thanks.
You can use sort on a random array of any size (which is what randperm does). After that, all you need to do is some index-trickery to properly reshuffle the array
A = [0 0 0 0 1; 0 0 0 0 2; 0 1 2 3 4];
[nRows,nCols] = size(A);
[~,idx] = sort(rand(nRows,nCols),2);
%# convert column indices into linear indices
idx = (idx-1)*nRows + ndgrid(1:nRows,1:nCols);
%# rearrange A
B = A;
B(:) = B(idx)
B =
0 0 1 0 0
0 2 0 0 0
2 1 3 4 0