I have a 300x178 matrix and I want to find the minimum of each column of that matrix, i.e. resulting in a 1x178 array. Then I want to store the sum of all elements but the minimum in each column in the 300x178 matrix on the location/pixel of the minimum value, leaving all other elements zero. How can I do this using MATLAB?
Example:
1 4 6 3
2 6 7 4
5 1 5 7
becomes:
1 0 0 1
0 0 0 0
0 1 1 0
and eventually:
8 0 0 14
0 0 0 0
0 11 18 0
Your example and title do not correspond to the question text. Your example sums all values in a column and stores them at the location of the minimum, which the title also asks. You can do this by making smart use of sub2ind:
A = [1 4 6 3
2 6 7 4
5 1 5 7];
C = zeros(size(A));
[tmp, idx] = min(A); % find the locations of minima
% one liner to store the sum of columns
C(sub2ind(size(A), idx, 1:size(A,2))) = sum(A,1);
C =
8 0 0 14
0 0 0 0
0 11 18 0
If, on the other hand, you're after what your question text asks about, subsequently subtract A on the minimum locations using the same sub2ind trick:
C(sub2ind(size(A), idx, 1:size(A,2))) = C(sub2ind(size(A), idx, 1:size(A,2))) - A(sub2ind(size(A), idx, 1:size(A,2)))
C =
7 0 0 11
0 0 0 0
0 10 13 0
This way you get the sum of all elements but the minimum.
For an in-depth explanation what sub2ind does, read this fantastic Q/A by Luis Mendo keeping in mind that in A(2,3) the 2 and 3 are called subscripts, which, in case of a 3-by-4 matrix, translates to linear index 8.
I cannot test this on my R2007b, but according to the documentation on min you could use [M, I] = min(A, [], 1, 'linear') to get the linear indices into I directly, without going through sub2ind:
C = zeros(size(A));
[tmp, idx] = min(A, [], 1, 'linear');
C(idx) = sum(A, 1);
% Optional, to sum all but the minimum
C(idx) = C(idx) - A(idx);
Small note from the documentation on the occurrence of multiple same-valued minima in your original matrix:
If the smallest element occurs more than once, then I contains the index to the first occurrence of the value.
Related
I'd like to extract one value per column of a matrix using a condition. Multiple values on each column match that condition, but only the last one should be selected. It is safe to assume that each row contains at least one such value.
So given an NxM matrix and an equally-sized boolean, extract M values for which the boolean is true and it is the last true value in a column. For example:
m = magic(4);
i = (m > 10);
% m =
% 16 2 3 13
% 5 11 10 8
% 9 7 6 12
% 4 14 15 1
% i =
% 1 0 0 1
% 0 1 0 0
% 0 0 0 1
% 0 1 1 0
And the expected output:
% i_ =
% 1 0 0 0
% 0 0 0 0
% 0 0 0 1
% 0 1 1 0
% x = [16, 14, 15, 12]
I know this could be easily achieved by looping through the columns and using find, but in my experience there often are better ways of formulating these problems.
This would do it
m(max(i.*reshape([1:numel(m)],size(m))))
Explanation
So we are generating an array of indices
reshape([1:numel(m)],size(m))
ans =
1 5 9 13
2 6 10 14
3 7 11 15
4 8 12 16
That represents the indices for each value. The we multiply that with I to get the values we are interested in
i.*reshape([1:numel(m)],size(m))
ans =
1 0 0 13
0 6 0 0
0 0 0 15
0 8 12 0
Then we do a max on that since max works on columns. This will give us the last index in each column.
max(i.*reshape([1:numel(m)],size(m)))
ans =
1 8 12 15
Then apply those indices on m to get the values
m(max(i.*reshape([1:numel(m)],size(m))))
ans =
16 14 15 12
You can use the second output of max to find the last true element of each column. Before that the logical matrx should be multiplied by an increasing column vector.
[~, idx] = max((1:size(i, 1)).' .* i, [], 1, 'linear') ;
x = m(idx) ;
Here's another way, using accumarray:
[~, col] = find(i); % column indices
lin = find(i); % linear indices
x = accumarray(col, m(lin), [], #(x) x(end));
I have a vector M containing single elements and repeats. I want to delete all the single elements. Turning something like [1 1 2 3 4 5 4 4 5] to [1 1 4 5 4 4 5].
I thought I'd try to get the count of each element then use the index to delete what I don't need, something like this:
uniq = unique(M);
list = [uniq histc(M,uniq)];
Though I'm stuck here and not sure how to go forward. Can anyone help?
Here is a solution using unique, histcounts and ismember:
tmp=unique(M) ; %finding unique elements of M
%Now keeping only those elements in tmp which appear only once in M
tmp = tmp(histcounts(M,[tmp tmp(end)])==1); %Thanks to rahnema for his insight on this
[~,ind] = ismember(tmp,M); %finding the indexes of these elements in M
M(ind)=[];
histcounts was introduced in R2014b. For earlier versions, hist can be used by replacing that line with this:
tmp=tmp(hist(M,tmp)==1);
You can get the result with the following code:
A = [a.', ones(length(a),1)];
[C,~,ic] = unique(A(:,1));
result = [C, accumarray(ic,A(:,2))];
a = A(~ismember(A(:,1),result(result(:,2) == 1))).';
The idea is, add ones to the second column of a', then accumarray base on the first column (elements of a). After that, found the elements in first column which have accum sum in the second column. Therefore, these elements repeated once in a. Finally, removing them from the first column of A.
Here is a cheaper alternative:
[s ii] = sort(a);
x = [false s(2:end)==s(1:end-1)]
y = [x(2:end)|x(1:end-1) x(end)]
z(ii) = y;
result = a(z);
Assuming the input is
a =
1 1 8 8 3 1 4 5 4 6 4 5
we sort the list s and get index of the sorted list ii
s=
1 1 1 3 4 4 4 5 5 6 8 8
we can find index of repeated elements and for it we check if an element is equal to the previous element
x =
0 1 1 0 0 1 1 0 1 0 0 1
however in x the first elements of each block is omitted to find it we can apply [or] between each element with the previous element
y =
1 1 1 0 1 1 1 1 1 0 1 1
we now have sorted logical index of repeated elements. It should be reordered to its original order. For it we use index of sorted elements ii :
z =
1 1 1 1 0 1 1 1 1 0 1 1
finally use z to extract only the repeated elements.
result =
1 1 8 8 1 4 5 4 4 5
Here is a result of a test in Octave* for the following input:
a = randi([1 100000],1,10000000);
-------HIST--------
Elapsed time is 5.38654 seconds.
----ACCUMARRAY------
Elapsed time is 2.62602 seconds.
-------SORT--------
Elapsed time is 1.83391 seconds.
-------LOOP--------
Doesn't complete in 15 seconds.
*Since in Octave histcounts hasn't been implemented so instead of histcounts I used hist.
You can test it Online
X = [1 1 2 3 4 5 4 4 5];
Y = X;
A = unique(X);
for i = 1:length(A)
idx = find(X==A(i));
if length(idx) == 1
Y(idx) = NaN;
end
end
Y(isnan(Y)) = [];
Then, Y would be [1 1 4 5 4 4 5]. It detects all single elements, and makes them as NaN, and then remove all NaN elements from the vector.
I think the question is pretty basic, but still keeps me busy since some time.
Lets assume we have a vector containing 4 integers randomly repetetive, like:
v = [ 1 3 3 3 4 2 1 2 3 4 3 2 1 4 3 3 4 2 2]
I am searching for the vector of all positions of each integer, e.g. for 1 it should be a vector like:
position_one = [1 7 13]
Since I want to search every row of a 100x10000 matrix I was not able to deal with linear indeces.
Thanks in advance!
Rows and columns
Since your output for every integer changes, a cell array will fit the whole task. For the whole matrix, you can do something like:
A = randi(4,10,30); % some data
Row = repmat((1:size(A,1)).',1,size(A,2)); % index of the row
Col = repmat((1:size(A,2)),size(A,1),1); % index of the column
pos = #(n) [Row(A==n) Col(A==n)]; % Anonymous function to find the indices of 'n'
than for every n you can write:
>> pos(3)
ans =
1 1
2 1
5 1
6 1
9 1
8 2
3 3
. .
. .
. .
where the first column is the row, and the second is the column for every instance of n in A.
And for all ns you can use an arrayfun:
positions = arrayfun(pos,1:max(A(:)),'UniformOutput',false) % a loop that goes over all n's
or a simple for loop (faster):
positions = cell(1,max(A(:)));
for n = 1:max(A(:))
positions(n) = {pos(n)};
end
The output in both cases would be a cell array:
positions =
[70x2 double] [78x2 double] [76x2 double] [76x2 double]
and for every n you can write positions{n}, to get for example:
>> positions{1}
ans =
10 1
2 3
5 3
3 4
5 4
1 5
4 5
. .
. .
. .
Only rows
If all you want in the column index per a given row and n, you can write this:
A = randi(4,10,30);
row_pos = #(k,n) A(k,:)==n;
positions = false(size(A,1),max(A(:)),size(A,2));
for n = 1:max(A(:))
positions(:,n,:) = row_pos(1:size(A,1),n);
end
now, positions is a logical 3-D array, that every row corresponds to a row in A, every column corresponds to a value of n, and the third dimension is the presence vector for the combination of row and n. this way, we can define R to be the column index:
R = 1:size(A,2);
and then find the relevant positions for a given row and n. For instance, the column indices of n=3 in row 9 is:
>> R(positions(9,3,:))
ans =
2 6 18 19 23 24 26 27
this would be just like calling find(A(9,:)==3), but if you need to perform this many times, the finding all indices and store them in positions (which is logical so it is not so big) would be faster.
Find linear indexes in a matrix: I = find(A == 1).
Find two dimensional indexes in matrix A: [row, col] = find(A == 1).
%Create sample matrix, with elements equal one:
A = zeros(5, 4);
A([2 10 14]) = 1
A =
0 0 0 0
1 0 0 0
0 0 0 0
0 0 1 0
0 1 0 0
Find ones as linear indexes in A:
find(A == 1)
ans =
2
10
14
%This is the same as reshaping A to a vector and find ones in the vector:
B = A(:);
find(B == 1);
B' =
0 1 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0
Find two dimensional indexes:
[row, col] = find(A == 1)
row =
2
5
4
col =
1
2
3
You can do that with accumarray using an anonymous function as follows:
positions = accumarray(v(:), 1:numel(v), [], #(x) {sort(x.')});
For
v = [ 1 3 3 3 4 2 1 2 3 4 3 2 1 4 3 3 4 2 2];
this gives
positions{1} =
1 7 13
positions{2} =
6 8 12 18 19
positions{3} =
2 3 4 9 11 15 16
positions{4} =
5 10 14 17
I want to create a variable that finds a pattern (let's say [1 1]) in different rows of a matrix (A). Of course there aren't an equal number of occurrences of this string in each row.
A = [ 0 0 0 1 1
1 1 1 0 0
0 1 0 1 1
1 1 1 0 0
0 1 0 0 1
1 0 1 1 1
0 1 0 1 0
1 1 1 0 1];
I could do:
for i = 1:n
var(i,:) = strfind(A(i,:),[1 1]);
end
but then both sides of the equation won't be equal.
ERROR: ??? Subscripted assignment dimension mismatch.
I try to preallocate. I create a matrix with what I think would be the maximum number of occurrences of this string in each row of matrix A (let's say 50).
for i = 1:n
var(i, :) = NaN(1,50)
end
That's followed by the previous bit of code and it's no good either.
I've also tried:
for i = 1:n
var(i,1:numel(strfind(A(i,:),[1 1])) = strfind(A(i,:),[1 1])
end
Error: The expression to the left of the equals sign is not a valid
target for an assignment.
How should I go about doing this?
The output I expect is a matrix var(i,:) that gives me the position in the matrix where each of these patterns occur. It works fine for just one row.
For example:
var(1,:) = [1 2 5 8 10 22 48]
var(2,:) = [2 3 4 7 34 45 NaN]
var(3,:) = [4 5 21 32 33 NaN]
Thanks!
In your first try: you tried to build a matrix with different length of rows.
In your second try: you pre-allocated, but then run it over by re-definning var(i,:), while you tried to put there your desired result.
In your third try: unfortunately you just missed one brackets- ) at the end of left expression.
This code suppose to work (what you did at 2nd and 3rd attempts, with pre-allocate and fixed brackets):
var=NaN(1,50);
for i = 1:n
var(i,1:numel(strfind(A(i,:),[1 1]))) = strfind(A(i,:),[1 1])
end
I'm looking for an elegant solution to this very simple problem in MATLAB. Suppose I have a matrix
>> M = magic(5)
M =
17 24 1 8 15
23 5 7 14 16
4 6 13 20 22
10 12 19 21 3
11 18 25 2 9
and a logical variable of the form
I =
0 0 0 0 0
0 1 1 0 0
0 1 1 0 0
0 0 0 0 0
0 0 0 0 0
If I try to retrieve the elements of M associated to 1 values in I, I get a column vector
>> M(I)
ans =
5
6
7
13
What would be the simplest way to obtain the matrix [5 7 ; 6 13] from this logical indexing?
If I know the shape of the non-zero elements of I, I can use a reshape after the indexing, but that's not a general case.
Also, I'm aware that the default behavior for this type of indexing in MATLAB enforces consistency with respect to the case in which non-zero values in I do not form a matrix, but I wonder if there is a simple solution for this particular case.
This is a one way to do this. It is assumed that all rows of I have same number of ones. It is also assumed that all columns of I have same number have ones, because Submatrix must be rectangular.
%# Define the example data.
M = magic(5);
I = zeros(5);
I(2:3, 2:3) = 1;
%# Create the Submatrix.
Submatrix = reshape(M(find(I)), max(sum(I)), max(sum(I')));
Here is a very simple solution:
T = I(any(I'),any(I));
T(:) = M(I);
M = magic(5);
I = [ ... ];
ind = find(I); %# find indices of ones in I
[y1, x1] = ind2sub(size(M), ind(1)); %# get top-left position
[y2, x2] = ind2sub(size(M), ind(end)); %# get bottom-right position
O = M(y1:y2, x1:x2); %# copy submatrix