I have the following matrix:
M =[ 0 0 0 0 0 0 0 0 ;
0 0 0 0 0 0 6 0 ;
0 0 0 0 3 0 3 3 ;
0 0 0 9 9 6 9 6 ;
0 0 0 9 6 9 9 0 ;
0 0 0 0 6 3 0 0 ;
0 0 0 0 0 0 0 0 ;
0 0 0 0 0 0 0 0 ];
and I want to add 1 in ten randomly chosen elements of M which are greater than 0.
How to do this?
Find the indices of the elements greater than zero and generate n random samples from those indices which will be incremented. Then simply add 1 to the elements at those n random indexes.
n=10; %No. of elements of M greater than zero which will be incremented by 1
t1 = find(M>0); %Finding indices of elements of M which are greater than zero
t2 = randperm(length(t1)); %Generating random indices from which first n will be selected
M(t1(t2(1:n))) = M(t1(t2(1:n)))+1 %Incrementing elements of random n indices
If you have Statistics and Machine Learning Toolbox, you can use the following simpler
solution using randsample:
n=10;
temp = randsample(find(M>0),n);
M(temp) = M(temp)+1
You can select which elements to increment by creating a random permutation of the positions of the non-zero elements.
m = 10; % we want to increment 10 elements
elements = find(M); % positions of the nonzero elements
rand_order = randperm(numel(elements), m); % generate random permutation
M(elements(rand_order)) = M(elements(rand_order)) + 1;
randperm ensures that each nonzero element will only be selected once, in a random order.
Related
I need to create all possible permutation matrices for a matrix where every permutation matrix contains only one 1 in each column and each row, and 0 in all other places.
For example, below example in (1) is all possible permutation matrices for 2x2 matrix and in (2) is a all possible permutation matrices for 3x3 matrix and so on
So how can I get these matrices of a matrix NxN in MATLAB and store them into one three-dimensional matrix?
Here's my solution, using implicit expansion (tested with Octave 5.2.0 and MATLAB Online):
n = 3;
% Get all permutations of length n
p = perms(1:n);
% Number of permutations
n_p = size(p, 1);
% Set up indices, where to set elements to 1
p = p + (0:n:n^2-1) + (0:n^2:n^2*n_p-1).';
% Set up indices, where to set elements to 1 (for MATLAB R2016a and before)
%p = bsxfun(#plus, bsxfun(#plus, p, (0:n:n^2-1)), (0:n^2:n^2*n_p-1).');
% Initialize 3-dimensional matrix
a = zeros(n, n, n_p);
% Set proper elements to 1
a(p) = 1
The output for n = 3:
a =
ans(:,:,1) =
0 0 1
0 1 0
1 0 0
ans(:,:,2) =
0 1 0
0 0 1
1 0 0
ans(:,:,3) =
0 0 1
1 0 0
0 1 0
ans(:,:,4) =
0 1 0
1 0 0
0 0 1
ans(:,:,5) =
1 0 0
0 0 1
0 1 0
ans(:,:,6) =
1 0 0
0 1 0
0 0 1
Using repelem, perms and reshape:
n = 3; % matrix size
f = factorial(n); % number of permutation
rep = repelem(eye(n),1,1,f) % repeat n! time the diagonal matrix
res = reshape(rep(:,perms(1:n).'),n,n,f) % indexing and reshaping
Where res is:
res =
ans(:,:,1) =
0 0 1
0 1 0
1 0 0
ans(:,:,2) =
0 1 0
0 0 1
1 0 0
ans(:,:,3) =
0 0 1
1 0 0
0 1 0
ans(:,:,4) =
0 1 0
1 0 0
0 0 1
ans(:,:,5) =
1 0 0
0 0 1
0 1 0
ans(:,:,6) =
1 0 0
0 1 0
0 0 1
And according to your comment:
What I need to do is to multiply a matrix i.e Z with all possible
permutation matrices and choose that permutation matrix which
resulting a tr(Y) minimum; where Y is the results of multiplication of
Z with the permutation matrix. I Think I don't need to generate all
permutation matrices and store them in such variable, I can generate
them one by one and get the result of multiplication. Is that possible
?
You're trying to solve the assignment problem, you can use the well known hungarian algorithm to solve this task in polynomial time. No needs to generate a googleplex of permutation matrix.
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 have a 64-by-1 vector which contains 27 non-zero values. I want to create N copies from that vector such that each copy contains only 4 non-zero values (in that case the first 6 copies will have 4 non-zero values and the last copy will contain only 3 non-zero values) using MATLAB.
For example:
orig_vector = [0 0 0 0 1 0 0 0 0 5 0 0 0 2 0 1 0 2 3 1 1 ];
first_copy = [0 0 0 0 1 0 0 0 0 5 0 0 0 2 0 1 0 0 0 0 0 ];
second_copy = [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 3 1 1 ];
How can this be done?
Perhaps something like:
non_zero_indices = find(orig_vector); % get array indices of non-zero elements
n_non_zero = length(non_zero_indices);
n_copies = ceil(n_non_zero / 4); % eg. with 6 non-zero elements we will have 2 copies
new_vectors = zeros(n_copies, length(orig_vector)); % matrix of new vectors where vectors go in rows
for i=0:n_copies - 2
idx = non_zero_indices(1+i*4:4+i*4);
new_vectors(i+1, idx) = orig_vector(idx);
end
idx = non_zero_indices(1+(n_copies-1)*4:end); % handle end which may have fewer than 4 elements
new_vectors(n_copies, idx) = orig_vector(idx);
I'd like to create a Matrix in MATLAB where:
The First row consists of a random arrangement of 0s and 1s, split evenly (i.e. 50-50).
The Second row randomly assigns zeros to 50% of the 0s and 1s in the first row, and ones to the remaining 50%.
The Third row randomly assigns zeros to 50% of the 0s and 1s in the second row, and ones to the remaining 50%.
Non-randomized Example:
0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1
0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1
0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1
Any suggestions?
A solution based on checking whether numbers are bigger or smaller than the median. As long as the number of columns tested is even, exactly half of a set of random doubles will be bigger than the median, and half will be smaller. This guarantees that there's exactly 50% of bits get flipped.
nRows = 3;
nCols = 16; %# divisible by 4
%# seed the array
%# assume that the numbers in each row are unique (very, very likely)
array = rand(nRows,nCols);
out = false(nRows,nCols);
%# first row is special
out(1,:) = array(1,:) > median(array(1,:));
%# for the rest of the row, check median for the zeros/ones in the previous row
for iRow = 2:nRows
zeroIdx = out(iRow-1,:) == 0;
%# > or < do not matter, both will replace zeros/ones
%# and replace with exactly half zeros and half ones
out(iRow,zeroIdx) = array(iRow,zeroIdx) > median(array(iRow,zeroIdx));
out(iRow,~zeroIdx) = array(iRow,~zeroIdx) > median(array(iRow,~zeroIdx));
end
I'd offer a short bsxfun solution:
%// number of divisions
n = 4;
%// unshuffled matrix like in your example
unshuffled = bsxfun(#(a,b) mod(a,2*b) > b-1, meshgrid(1:n^2,1:n) - 1, (2.^((n-1):-1:0)).') %'
%// shuffle columns
shuffled = unshuffled(:,randperm(n^2))
unshuffled =
0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1
0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1
0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
shuffled =
1 0 1 1 0 1 0 1 1 1 1 0 0 0 0 0
1 1 1 0 0 1 1 0 1 0 0 0 1 0 1 0
1 0 0 1 0 0 0 0 1 1 0 1 1 0 1 1
1 1 1 1 0 0 0 0 0 0 1 0 0 1 1 1
First you need to create the unshuffled matrix, which can be done by comparing the matrix generated by meshgrid(1:n^2,1:n) with a row dependent modulus. Finally you just need to shuffle the columns.
If you have the Statistics Toolbox, you can do it very easily with randsample:
M = 3; %// number of rows
N = 16; %// number of columns. Should be multiple of 4, according to problem definition
result = zeros(M,N); %// preallocate and initiallize to zeros
result(1, randsample(1:N,N/2)) = 1; %// first row: half values set to one, half to zero
for m = 2:M
result(m, :) = result(m-1, :); %// initiallize row m equal to row m-1
result(m, randsample(find(result(m-1,:)), N/4)) = 0; %// change half of ones
result(m, randsample(find(~result(m-1,:)), N/4)) = 1; %// change half of zeros
end
Example result:
result =
0 1 0 1 1 0 0 0 0 1 0 1 1 0 1 1
1 1 0 0 0 1 1 1 0 1 0 1 0 0 0 1
1 0 0 0 1 0 0 1 0 1 1 0 1 1 0 1
A solution using randperm:
nrows = 3;
ncols = 16;
M = zeros(nrows,ncols);
%// seed the first row
M(1,1:ncols/2) = 1;
M(1,:) = M(1,randperm(ncols));
for r = 2:nrows
%// Find ncols/4 random between 1 and ncols/2. These will be used to index half of the previous rows 1 elements and set them to one
idx = randperm(ncols/2);
idx1 = idx(1:ncols/4);
%// Do the same thing again, but this time it will be used for the 0 elements of the previous row
idx = randperm(ncols/2);
idx0 = idx(1:ncols/4);
idx_prev1 = find(M(r-1,:)); %// Find where the 1 elements were in the last row
idx_prev0 = find(~M(r-1,:)); %// Find where the 0 elements were in the last row
M(r,idx_prev1(idx1))=1; %// Set half of the previous rows 1 elements in this row to 1
M(r,idx_prev0(idx0))=1; %// Set half of the previous rows 0 elements in this row to 1
end
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.