Within the context of writing a certain function, I have the following example matrix:
temp =
1 2 0 0 1 0
1 0 0 0 0 0
0 1 0 0 0 1
I want to obtain an array whose each element indicates the number of the element out of all non-zero elements which starts that column. If a column is empty, the element should correspond to the next non-empty column. For the matrix temp, the result would be:
result = [1 3 5 5 5 6]
Because the first non-zero element starts the first column, the third starts the second column, the fifth starts the fifth column and the sixth starts the sixth column.
How can I do this operation for any general matrix (one which may or may not contain empty columns) in a vectorized way?
Code:
temp = [1 2 0 0 1 0; 1 0 0 0 0 0; 0 1 0 0 0 1]
t10 = temp~=0
l2 = cumsum(t10(end:-1:1))
temp2 = reshape(l2(end)-l2(end:-1:1)+1, size(temp))
result = temp2(1,:)
Output:
temp =
1 2 0 0 1 0
1 0 0 0 0 0
0 1 0 0 0 1
t10 =
1 1 0 0 1 0
1 0 0 0 0 0
0 1 0 0 0 1
l2 =
1 1 1 1 1 2 2 2 2 2 2 2 3 3 4 4 5 6
temp2 =
1 3 5 5 5 6
2 4 5 5 6 6
3 4 5 5 6 6
result =
1 3 5 5 5 6
Printing values of each step may be clearer than my explanation. Basically we use cumsum to get the IDs of the non-zero elements. As you need to know the ID before reaching the element, a reversed cumsum will do. Then the only thing left is to reverse the ID numbers back.
Here's another way:
temp = [1 2 0 0 1 0; 1 0 0 0 0 0; 0 1 0 0 0 1]; % data
[~, c] = find(temp); % col indices of nonzero elements
result = accumarray(c, 1:numel(c), [], #min, NaN).'; % index, among all nonzero
% values, of the first nonzero value of each col; or NaN if none exists
result = cummin(result, 'reverse'); % fill NaN's using backwards cumulative maximum
Related
I have a vector
Y = [1 1 0 0 0 1 1 0 1 0 1 1 1 0 0 0 1 1 0 0 0 0 1 0 1 0 1 0 1 1 1 0 0 0 1 0 0 0]
1 occurs 17 times
0 occurs 21 times
How can I randomly remove 0s so that both values have equal amounts, such as 1 (17 times) and 0 (17 times)?
This should also work on much bigger matrix.
Starting with your example
Y = [1 1 0 0 0 1 1 0 1 0 1 1 1 0 0 0 1 1 0 0 0 0 1 0 1 0 1 0 1 1 1 0 0 0 1 0 0 0]
You can do the following:
% Get the indices of the value which is more common (`0` here)
zeroIdx = find(~Y); % equivalent to find(Y==0)
% Get random indices to remove
remIdx = randperm(nnz(~Y), nnz(~Y) - nnz(Y));
% Remove elements
Y(zeroIdx(remIdx)) = [];
You could combine the last two lines, but I think it would be less clear.
The randperm line is choosing the correct number of elements to remove from random indices between 1 and the number of zeros.
If the data can only have two values
Values are assumed to be 0 and 1. The most common value is randomly removed to equalize their counts:
Y = [1 1 0 0 0 1 1 0 1 0 1 1 1 0 0 0 1 1 0 0 0 0 1 0 1 0 1 0 1 1 1 0 0 0 1 0 0 0]; % data
ind0 = find(Y==0); % indices of zeros
ind1 = find(Y==1); % indices of ones
t(1,1:numel(ind0)) = ind0(randperm(numel(ind0))); % random permutation of indices of zeros
t(2,1:numel(ind1)) = ind1(randperm(numel(ind1))); % same for ones. Pads shorter row with 0
t = t(:, all(t,1)); % keep only columns that don't have padding
result = Y(sort(t(:))); % linearize, sort and use those indices into the data
Generalization for more than two values
Values are arbitrary. All values except the least common one are randomly removed to equalize their counts:
Y = [0 1 2 0 2 1 1 2 0 2 1 2 2 0 0]; % data
vals = [0 1 2]; % or use vals = unique(Y), but absent values will not be detected
t = [];
for k = 1:numel(vals) % loop over values
ind_k = find(Y==vals(k));
t(k, 1:numel(ind_k)) = ind_k(randperm(numel(ind_k)));
end
t = t(:, all(t,1));
result = Y(sort(t(:)));
Say I'm given a symmetric row vector with an odd length where each element is smaller than the next one in the first half of the vector and each element is bigger than the next one in the second half and the middle element is the biggest. (e.g [1 2 3 2 1] or [10 20 50 20 10]).
I want to create a square matrix where this row vector is its middle row and the equivalent column vector (v') is its middle column and each other row or column is a reduced version of the given vector according to the middle element in this row or column. And when there are no more "original elements" we put 0.
Examples:
if v = [1 2 3 2 1] we get
0 0 1 0 0
0 1 2 1 0
1 2 3 2 1
0 1 2 1 0
0 0 1 0 0
if v = [3 5 3] we get
0 3 0
3 5 3
0 3 0
What I did so far: I managed to create a matrix with v as the middle row and v' as the middle column with this code I wrote:
s = length(vector);
matrix= zeros(s);
matrix(round(s/2),:) = vector;
matrix(:, round(s/2)) = vector';
but got stuck with assigning the other values.
A more hands-on approach is to produce your matrix as a mosaic, starting from a hankel matrix. For performance comparison, here's a version using the same format as #Divakar's solution:
function out=pyramid_hankel(v)
%I suggest checking v here
%it should be odd in length and a palindrome
i0=ceil(length(v)/2);
v2=v(i0:end);
Mtmp=hankel(v2);
out=zeros(length(v));
out(i0:end,i0:end)=Mtmp;
out(1:i0-1,i0:end)=flipud(Mtmp(2:end,:));
out(:,1:i0-1)=fliplr(out(:,i0+1:end));
>> pyramid_hankel([1 2 3 2 1])
ans =
0 0 1 0 0
0 1 2 1 0
1 2 3 2 1
0 1 2 1 0
0 0 1 0 0
For v=[1 2 3 2 1] the starting block is hankel([3 2 1]), which is
ans =
3 2 1
2 1 0
1 0 0
From here it should be clear what's happening.
Here's one approach -
function out = pyramid(v)
hlen = (numel(v)+1)/2;
updown_vec = [1:(numel(v)+1)/2 (numel(v)-1)/2:-1:1];
upper_part = cumsum(bsxfun(#le,(hlen:-1:1)',updown_vec)); %//'
out = [upper_part ; flipud(upper_part(1:end-1,:))];
out = changem(out,v,updown_vec);
Here's another approach, sort of simpler maybe -
function out = pyramid_v2(v)
hlen = (numel(v)+1)/2;
updown_vec = [1:(numel(v)+1)/2 (numel(v)-1)/2:-1:1];
mask = bsxfun(#le,([hlen:-1:1 2:hlen])',updown_vec); %//'
M = double(mask);
M(hlen+1:end,:) = -1;
out = changem(cumsum(M).*mask,v,updown_vec);
Sample runs -
>> v = [1 2 3 2 1];
>> pyramid(v)
ans =
0 0 1 0 0
0 1 2 1 0
1 2 3 2 1
0 1 2 1 0
0 0 1 0 0
>> v = [3 5 3];
>> pyramid(v)
ans =
0 3 0
3 5 3
0 3 0
>> v = [99,3,78,55,78,3,99];
>> pyramid(v)
ans =
0 0 0 99 0 0 0
0 0 99 3 99 0 0
0 99 3 78 3 99 0
99 3 78 55 78 3 99
0 99 3 78 3 99 0
0 0 99 3 99 0 0
0 0 0 99 0 0 0
Here's another approach:
v = [1 2 3 2 1]; %// symmetric, odd size
m = (numel(v)-1)/2;
w = [0 v(1:m+1)];
t = abs(-m:m);
result = w(max(m+2-bsxfun(#plus, t, t.'),1));
Apologies in advance if this question is a duplicate, or if the solution to this question is very straightforward in Matlab. I have a M x N matrix A, a 1 x M vector ind, and another vector val. For example,
A = zeros(6,5);
ind = [3 4 2 4 2 3];
val = [1 2 3];
I would like to vectorize the following code:
for i = 1 : size(A,1)
A(i, ind(i)-1 : ind(i)+1) = val;
end
>> A
A =
0 1 2 3 0
0 0 1 2 3
1 2 3 0 0
0 0 1 2 3
1 2 3 0 0
0 1 2 3 0
That is, for row i of A, I want to insert the vector val in a certain location, as specificied by the i'th entry of ind. What's the best way to do this in Matlab without a for loop?
It can be done using bsxfun's masking capability: build a mask telling where the values will be placed, and then fill those values in. In doing this, it's easier to work with columns instead of rows (because of Matlab's column major order), and transpose at the end.
The code below determines the minimum number of columns in the final A so that all values fit at the specified positions.
Your example applies a displacement of -1 with respect to ind. The code includes a generic displacement, which can be modified.
%// Data
ind = [3 4 2 4 2 3]; %// indices
val = [1 2 3]; %// values
d = -1; %// displacement for indices. -1 in your example
%// Let's go
n = numel(val);
m = numel(ind);
N = max(ind-1) + n + d; %// number of rows in A (rows before transposition)
mask = bsxfun(#ge, (1:N).', ind+d) & bsxfun(#le, (1:N).', ind+n-1+d); %// build mask
A = zeros(size(mask)); %/// define A with zeros
A(mask) = repmat(val(:), m, 1); %// fill in values as indicated by mask
A = A.'; %// transpose
Result in your example:
A =
0 1 2 3 0
0 0 1 2 3
1 2 3 0 0
0 0 1 2 3
1 2 3 0 0
0 1 2 3 0
Result with d = 0 (no displacement):
A =
0 0 1 2 3 0
0 0 0 1 2 3
0 1 2 3 0 0
0 0 0 1 2 3
0 1 2 3 0 0
0 0 1 2 3 0
If you can handle a bit of bsxfun overdose, here's one with bsxfun's adding capability -
N = numel(ind);
A(bsxfun(#plus,N*[-1:1]',(ind-1)*N + [1:N])) = repmat(val(:),1,N)
Sample run -
>> ind
ind =
3 4 2 4 2 3
>> val
val =
1 2 3
>> A = zeros(6,5);
>> N = numel(ind);
>> A(bsxfun(#plus,N*[-1:1]',(ind-1)*N + [1:N])) = repmat(val(:),1,N)
A =
0 1 2 3 0
0 0 1 2 3
1 2 3 0 0
0 0 1 2 3
1 2 3 0 0
0 1 2 3 0
I have a logical matrix A, and I would like to select all the elements to the left of each of my 1s values given a fixed distant. Let's say my distance is 4, I would like to (for instance) replace with a fixed value (saying 2) all the 4 cells at the left of each 1 in A.
A= [0 0 0 0 0 1 0
0 1 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 1 0 1]
B= [0 2 2 2 2 1 0
2 1 0 0 0 0 0
0 0 0 0 0 0 0
2 2 2 2 2 2 1]
In B is what I would like to have, considering also overwrting (last row in B), and cases where there is only 1 value at the left of my 1 and not 4 as the fixed searching distance (second row).
How about this lovely one-liner?
n = 3;
const = 5;
A = [0 0 0 0 0 1 0;
0 1 0 0 0 0 0;
0 0 0 0 0 0 0;
0 0 0 0 1 0 1]
A(bsxfun(#ne,fliplr(filter(ones(1,1+n),1,fliplr(A),[],2)),A)) = const
results in:
A =
0 0 5 5 5 1 0
5 1 0 0 0 0 0
0 0 0 0 0 0 0
0 5 5 5 5 5 1
here some explanations:
Am = fliplr(A); %// mirrored input required
Bm = filter(ones(1,1+n),1,Am,[],2); %// moving average filter for 2nd dimension
B = fliplr(Bm); %// back mirrored
mask = bsxfun(#ne,B,A) %// mask for constants
A(mask) = const
Here is a simple solution you could have come up with:
w=4; % Window size
v=2; % Desired value
B = A;
for r=1:size(A,1) % Go over all rows
for c=2:size(A,2) % Go over all columns
if A(r,c)==1 % If we encounter a 1
B(r,max(1,c-w):c-1)=v; % Set the four spots before this point to your value (if possible)
end
end
end
d = 4; %// distance
v = 2; %// value
A = fliplr(A).'; %'// flip matrix, and transpose to work along rows.
ind = logical( cumsum(A) ...
- [ zeros(size(A,1)-d+2,size(A,2)); cumsum(A(1:end-d-1,:)) ] - A );
A(ind) = v;
A = fliplr(A.');
Result:
A =
0 2 2 2 2 1 0
2 1 0 0 0 0 0
0 0 0 0 0 0 0
2 2 2 2 2 2 1
Approach #1 One-liner using imdilate available with Image Processing Toolbox -
A(imdilate(A,[ones(1,4) zeros(1,4+1)])==1)=2
Explanation
Step #1: Create a morphological structuring element to be used with imdilate -
morph_strel = [ones(1,4) zeros(1,4+1)]
This basically represents a window extending n places to the left with ones and n places to the right including the origin with zeros.
Step #2: Use imdilate that will modify A such that we would have 1 at all four places to the left of each 1 in A -
imdilate_result = imdilate(A,morph_strel)
Step #3: Select all four indices for each 1 of A and set them to 2 -
A(imdilate_result==1)=2
Thus, one can write a general form for this approach as -
A(imdilate(A,[ones(1,window_length) zeros(1,window_length+1)])==1)=new_value
where window_length would be 4 and new_value would be 2 for the given data.
Approach #2 Using bsxfun-
%// Paramters
window_length = 4;
new_value = 2;
B = A' %//'
[r,c] = find(B)
extents = bsxfun(#plus,r,-window_length:-1)
valid_ind1 = extents>0
jump_factor = (c-1)*size(B,1)
extents_valid = extents.*valid_ind1
B(nonzeros(bsxfun(#plus,extents_valid,jump_factor).*valid_ind1))=new_value
B = B' %// B is the desired output
Given a 1*N matrix or an array, how do I find the first 4 elements which have the same value and then store the index for those elements?
PS:
I'm just curious. What if we want to find the first 4 elements whose value differences are within a certain range, say below 2? For example, M=[10,15,14.5,9,15.1,8.5,15.5,9.5], the elements I'm looking for will be 15,14.5,15.1,15.5 and the indices will be 2,3,5,7.
If you want the first value present 4 times in the array 'tab' in Matlab, you can use
num_min = 4
val=NaN;
for i = tab
if sum(tab==i) >= num_min
val = i;
break
end
end
ind = find(tab==val, num_min);
By instance with
tab = [2 4 4 5 4 6 4 5 5 4 6 9 5 5]
you get
val =
4
ind =
2 3 5 7
Here is my MATLAB solution:
array = randi(5, [1 10]); %# random array of integers
n = unique(array)'; %'# unique elements
[r,~] = find(cumsum(bsxfun(#eq,array,n),2) == 4, 1, 'first');
if isempty(r)
val = []; ind = []; %# no answer
else
val = n(r); %# the value found
ind = find(array == val, 4); %# indices of elements corresponding to val
end
Example:
array =
1 5 3 3 1 5 4 2 3 3
val =
3
ind =
3 4 9 10
Explanation:
First of all, we extract the list of unique elements. In the example used above, we have:
n =
1
2
3
4
5
Then using the BSXFUN function, we compare each unique value against the entire vector array we have. This is equivalent to the following:
result = zeros(length(n),length(array));
for i=1:length(n)
result(i,:) = (array == n(i)); %# row-by-row
end
Continuing with the same example we get:
result =
1 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0
0 0 1 1 0 0 0 0 1 1
0 0 0 0 0 0 1 0 0 0
0 1 0 0 0 1 0 0 0 0
Next we call CUMSUM on the result matrix to compute the cumulative sum along the rows. Each row will give us how many times the element in question appeared so far:
>> cumsum(result,2)
ans =
1 1 1 1 2 2 2 2 2 2
0 0 0 0 0 0 0 1 1 1
0 0 1 2 2 2 2 2 3 4
0 0 0 0 0 0 1 1 1 1
0 1 1 1 1 2 2 2 2 2
Then we compare that against four cumsum(result,2)==4 (since we want the location where an element appeared for the forth time):
>> cumsum(result,2)==4
ans =
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 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
Finally we call FIND to look for the first appearing 1 according to a column-wise order: if we traverse the matrix from the previous step column-by-column, then the row of the first appearing 1 indicates the index of the element we are looking for. In this case, it was the third row (r=3), thus the third element in the unique vector is the answer val = n(r). Note that if we had multiple elements repeated 4 times or more in the original array, then the one first appearing for the forth time will show up first as a 1 going column-by-column in the above expression.
Finding the indices of the corresponding answer value is a simple call to FIND...
Here is C++ code
std::map<int,std::vector<int> > dict;
std::vector<int> ans(4);//here we will store indexes
bool noanswer=true;
//my_vector is a vector, which we must analize
for(int i=0;i<my_vector.size();++i)
{
std::vector<int> &temp = dict[my_vector[i]];
temp.push_back(i);
if(temp.size()==4)//we find ans
{
std::copy(temp.begin(),temp.end(),ans.begin() );
noanswer = false;
break;
}
}
if(noanswer)
std::cout<<"No Answer!"<<std::endl;
Ignore this and use Amro's mighty solution . . .
Here is how I'd do it in Matlab. The matrix can be any size and contain any range of values and this should work. This solution will automatically find a value and then the indicies of the first 4 elements without being fed the search value a priori.
tab = [2 5 4 5 4 6 4 5 5 4 6 9 5 5]
%this is a loop to find the indicies of groups of 4 identical elements
tot = zeros(size(tab));
for nn = 1:numel(tab)
idxs=find(tab == tab(nn), 4, 'first');
if numel(idxs)<4
tot(nn) = Inf;
else
tot(nn) = sum(idxs);
end
end
%find the first 4 identical
bestTot = find(tot == min(tot), 1, 'first' );
%store the indicies you are interested in.
indiciesOfInterst = find(tab == tab(bestTot), 4, 'first')
Since I couldn't easily understand some of the solutions, I made that one:
l = 10; m = 5; array = randi(m, [1 l])
A = zeros(l,m); % m is the maximum value (may) in array
A(sub2ind([l,m],1:l,array)) = 1;
s = sum(A,1);
b = find(s(array) == 4,1);
% now in b is the index of the first element
if (~isempty(b))
find(array == array(b))
else
disp('nothing found');
end
I find this easier to visualize. It fills '1' in all places of a square matrix, where values in array exist - according to their position (row) and value (column). This is than summed up easily and mapped to the original array. Drawback: if array contains very large values, A may get relative large too.
You're PS question is more complicated. I didn't have time to check each case but the idea is here :
M=[10,15,14.5,9,15.1,8.5,15.5,9.5]
val = NaN;
num_min = 4;
delta = 2;
[Ms, iMs] = sort(M);
dMs = diff(Ms);
ind_min=Inf;
n = 0;
for i = 1:length(dMs)
if dMs(i) <= delta
n=n+1;
else
n=0;
end
if n == (num_min-1)
if (iMs(i) < ind_min)
ind_min = iMs(i);
end
end
end
ind = sort(iMs(ind_min + (0:num_min-1)))
val = M(ind)