Suppose I have an array a of bounded integers (in this case bounded by 5):
a = [3 4 4 2 1 5 5];
I want to convert this array of integers to a length(a) x 5 matrix A where each row is a bit array with a 1 in the column indexed by the integer from a:
A = [0 0 1 0 0;
0 0 0 1 0;
0 0 0 1 0;
0 1 0 0 0;
1 0 0 0 0;
0 0 0 0 1;
0 0 0 0 1];
This is easily accomplished with a for loop:
n = length(a)
A = zeros(n, max(a(:)));
for k = 1 : n
A(k, a(k)) = 1;
end
I am looking for a vectorized implementation that does not use a for loop.
Two possible methods:
use sparse:
A = sparse( 1:n, a, 1, n, max(a(:)) );
if you want a non-sparse result
full(A);
Using sun2ind:
A = zeros( n, max(a(:)) );
A( sub2ind(size(A), 1:n, a ) ) = 1;
Related
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.
I want to obtain all the possible permutations of one vector elements by another vector elements. For example one vector is A=[0 0 0 0] and another is B=[1 1]. I want to replace the elements of A by B to obtain all the permutations in a matrix like this [1 1 0 0; 1 0 1 0; 1 0 0 1; 0 1 1 0; 0 1 0 1; 0 0 1 1]. The length of real A is big and I should be able to choose the length of B_max and to obtain all the permutations of A with B=[1], [1 1], [1 1 1],..., B_max.
Thanks a lot
Actually, since A and B are always defined, respectively, as a vector of zeros and a vector of ones, this computation is much easier than you may think. The only constraints you should respect concerns B, which shoud not be empty and it's elements cannot be greater than or equal to the number of elements in A... because after that threshold A will become a vector of ones and calculating its permutations will be just a waste of CPU cycles.
Here is the core function of the script, which undertakes the creation of the unique permutations of 0 and 1 given the target vector X:
function p = uperms(X)
n = numel(X);
k = sum(X);
c = nchoosek(1:n,k);
m = size(c,1);
p = zeros(m,n);
p(repmat((1-m:0)',1,k) + m*c) = 1;
end
And here is the full code:
clear();
clc();
% Define the main parameter: the number of elements in A...
A_len = 4;
% Compute the elements of B accordingly...
B_len = A_len - 1;
B_seq = 1:B_len;
% Compute the possible mixtures of A and B...
X = tril(ones(A_len));
X = X(B_seq,:);
% Compute the unique permutations...
p = [];
for i = B_seq
p = [p; uperms(X(i,:).')];
end
Output for A_len = 4:
p =
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
1 1 0 0
1 0 1 0
1 0 0 1
0 1 1 0
0 1 0 1
0 0 1 1
1 1 1 0
1 1 0 1
1 0 1 1
0 1 1 1
I need a function that creates a checker board matrix with M rows and N columns of P*Q rectangles. I modified the third solution from here to get that:
function [I] = mycheckerboard(M, N, P, Q)
nr = M*P;
nc = N*Q;
i = floor(mod((0:(nc-1))/Q, 2));
j = floor(mod((0:(nr-1))/P, 2))';
r = repmat(i, [nr 1]);
c = repmat(j, [1 nc]);
I = xor(r, c);
it works with no problem:
I=mycheckerboard(2, 3, 4, 3)
I =
0 0 0 1 1 1 0 0 0
0 0 0 1 1 1 0 0 0
0 0 0 1 1 1 0 0 0
0 0 0 1 1 1 0 0 0
1 1 1 0 0 0 1 1 1
1 1 1 0 0 0 1 1 1
1 1 1 0 0 0 1 1 1
1 1 1 0 0 0 1 1 1
But it's not fast enough since there are lots of calls of this function in a single run. Is there a faster way to get the result? How can I remove floating point divisions and/or calls of the floor function?
Your code is fairly fast for small matrices, but becomes less so as the dimensions get larger. Here's a one-liner using bsxfun and imresize (requires Image Processing toolbox that most have):
m = 2;
n = 3;
p = 4;
q = 3;
I = imresize(bsxfun(#xor, mod(1:m, 2).', mod(1:n, 2)), [p*m q*n], 'nearest')
Or, inspired by #AndrasDeak's use of kron, this is faster with R2015b:
I = kron(bsxfun(#xor, mod(1:m, 2).', mod(1:n, 2)), ones(p, q))
For a small bit more speed, the underlying code for kron can be simplified by taking advantage of the structure of the problem:
A = bsxfun(#xor, mod(1:m, 2).', mod(1:n, 2));
A = permute(A, [3 1 4 2]);
B = ones(q, 1, p);
I = reshape(bsxfun(#times, A, B), [m*n p*q]);
or as one (long) line:
I = reshape(bsxfun(#times, permute(bsxfun(#xor, mod(1:m, 2).', mod(1:n, 2)), [3 1 4 2]), ones(q, 1, p)), [m*n p*q]);
I suggest first creating a binary matrix for the checkerboard's fields, then using the built-in kron to blow it up to the necessary size:
M = 2;
N = 3;
P = 4;
Q = 3;
[iM,iN] = meshgrid(1:M,1:N);
A = zeros(M,N);
A(mod(iM.'+iN.',2)==1) = 1;
board = kron(A,ones(P,Q))
In each iteration I want to add 1 randomly to binary vector,
Let say
iteration = 1,
k = [0 0 0 0 0 0 0 0 0 0]
iteration = 2,
k = [0 0 0 0 1 0 0 0 0 0]
iteration = 3,
k = [0 0 1 0 0 0 0 1 0 0]
, that goes up to length(find(k)) = 5;
Am thinking of for loop but I don't have an idea how to start.
If it's important to have the intermediate vectors (those with 1, 2, ... 4 ones) as well as the final one, you can generate a random permutation and, in your example, use the first 5 indices one at a time:
n = 9; %// number of elements in vector
m = 5; %// max number of 1's in vector
k = zeros(1, n);
disp(k); %// output vector of all 0's
idx = randperm(n);
for p = 1:m
k(idx(p)) = 1;
disp(k);
end
Here's a sample run:
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 1
1 0 0 1 0 0 0 0 1
1 0 0 1 1 0 0 0 1
1 1 0 1 1 0 0 0 1
I wouldn't even use a loop. I would generate a random permutation of indices that sample from a vector going from 1 up to the length of k without replacement then just set these locations to 1. randperm suits the task well:
N = 10; %// Length 10 vector
num_vals = 5; %// 5 values to populate
ind = randperm(N, num_vals); %// Generate a vector from 1 to N and sample num_vals values from this vector
k = zeros(1, N); %// Initialize output vector k to zero
k(ind) = 1; %// Set the right values to 1
Here are some sample runs when I run this code a few times:
k =
0 0 1 1 0 1 1 0 0 1
k =
1 0 0 0 1 0 1 1 0 1
k =
1 0 0 0 1 0 1 1 0 1
k =
0 1 1 1 0 0 1 0 0 1
However, if you insist on using a loop, you can generate a vector from 1 up to the desired length, randomly choose an index in this vector then remove this value from the vector. You'd then use this index to set the location of the output:
N = 10; %// Length 10 vector
num_vals = 5; %// 5 values to populate
vec = 1 : N; %// Generate vector from 1 up to N
k = zeros(1, N); %// Initialize output k
%// Repeat the following for as many times as num_vals
for idx = 1 : num_vals
%// Obtain a value from the vector
ind = vec(randi(numel(vec), 1));
%// Remove from the vector
vec(ind) = [];
%// Set location in output to 1
k(ind) = 1;
end
The above code should still give you the desired effect, but I would argue that it's less efficient.
i have a vector a = [1; 6; 8]
and want to create a matrix with n columns and size(a,1) rows.
Each i'th row is all zeros but on the a(i) index is one.
>> make_the_matrix(a, 10)
ans =
1 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0
use sparse
numCol = 10; % number of colums in output matrix, should not be greater than max(a)
mat = sparse( 1:numel(a), a, 1, numel(a), numCol );
if you want a full matrix just use
full(mat)
Here is my first thought:
a = [1;6;8];
nCols = 10;
nRows = length(a);
M = zeros(nRows,nCols);
M(:,a) = eye(nRows)
Basically the eye is assigned to the right columns of the matrix.