I need some help in this problem
I have this matrix in MATLAB:
A = [ 25 1.2 1
28 1.2 2
17 2.6 1
18 2.6 2
23 1.2 1
29 1.2 2
19 15 1
22 15 2
24 2.6 1
26 2.6 2];
1st column is some measured values for temperature
2nd column is an index code representing the color (1.2:red,.....etc)
3rd column is the hour of taking the sample. Only at hours from 1 to 2
I want the matrix to be controlled by 2nd column as follows:
if it is 1.2, the program will find the average of all temperatures at hour 1 that
corresponds to 1.2
So, here ( 25 + 23 )/2 = 24
and also finds the average of all temperatures at hour 2 and that corresponds
to 1.2, ( 28 + 29 ) /2 = 28.5
and this average values:
[24
28.5]
will replace all temperature values at hours 1 and 2
that corresponds to 1.2 .
Then, it does the same thing for indices 2.6 and 15
So, the desired output will be:
B = [ 24
28.5
15.5
22
24
28.5
19
22
15.5
22]
My problem is in using the loop. I could do it for only one index at one run.
for example,
T=[];
index=1.2;
for i=1:length(A)
if A(i,2)==index
T=[T A(i,1)];
else
T=[T 0];
end
end
So, T is the extracted T that corresponds to 1.2 and other entries are zeros
Then, I wrote long code to find the average and at the end I could find the matrix
that corresponds to ONLY the index 1.2 :
B = [24
28.5
0
0
24
28.5
0
0
0
0]
But this is only for one index and it assigns zeros for the other indices. I can do this for all
indices in separate runs and then add the B's but this will take very long time since my real
matrix is 8760 by 5 .
I am sure that there is a shorter way to do that.
Thanks
Regards
Try this:
B = zeros(size(A, 1), 1);
C = unique(A(:, 2))';
T = [1 2];
for c = C,
for t = T,
I1 = find((A(:, 2) == c) & (A(:, 3) == t));
B(I1) = mean(A(I1, 1));
end
end
Edit
I think your expected answer is wrong for c = 2.6 and t = 1... Shouldn't it be (17 + 24)/2 = 20.5?
This can be done, perhaps more neatly, with accumarray:
[~, ~, ii] = unique(A(:,2)); %// indices corresponding to second col values
ind = [ii A(:,3)]; %// build 2D-indices for accumarray
averages = accumarray(ind, A(:,1), [], #mean); %// desired averages of first col
result = averages(sub2ind(max(ind), ind(:,1), ind(:,2))); %// repeat averages
Related
I have 3d matrix A that has my data. At multiple locations defined by row and column indcies as shown by matrix row_col_idx I want to extract all data along the third dimension as shown below:
A = cat(3,[1:3;4:6], [7:9;10:12],[13:15;16:18],[19:21;22:24]) %matrix(2,3,4)
row_col_idx=[1 1;1 2; 2 3];
idx = sub2ind(size(A(:,:,1)), row_col_idx(:,1),row_col_idx(:,2));
out=nan(size(A,3),size(row_col_idx,1));
for k=1:size(A,3)
temp=A(:,:,k);
out(k,:)=temp(idx);
end
out
The output of this code is as follows:
A(:,:,1) =
1 2 3
4 5 6
A(:,:,2) =
7 8 9
10 11 12
A(:,:,3) =
13 14 15
16 17 18
A(:,:,4) =
19 20 21
22 23 24
out =
1 2 6
7 8 12
13 14 18
19 20 24
The output is as expected. However, the actual A and row_col_idx are huge, so this code is computationally expensive. Is there away to vertorize this code to avoid the loop and the temp matrix?
This can be vectorized using linear indexing and implicit expansion:
out = A( row_col_idx(:,1) + ...
(row_col_idx(:,2)-1)*size(A,1) + ...
(0:size(A,1)*size(A,2):numel(A)-1) ).';
The above builds an indexing matrix as large as the output. If this is unacceptable due to memory limiations, it can be avoided by reshaping A:
sz = size(A); % store size A
A = reshape(A, [], sz(3)); % collapse first two dimensions
out = A(row_col_idx(:,1) + (row_col_idx(:,2)-1)*sz(1),:).'; % linear indexing along
% first two dims of A
A = reshape(A, sz); % reshape back A, if needed
A more efficient method is using the entries of the row_col_idx vector for selecting the elements from A. I have compared the two methods for a large matrix, and as you can see the calculation is much faster.
For the A given in the question, it gives the same output
A = rand([2,3,10000000]);
row_col_idx=[1 1;1 2; 2 3];
idx = sub2ind(size(A(:,:,1)), row_col_idx(:,1),row_col_idx(:,2));
out=nan(size(A,3),size(row_col_idx,1));
tic;
for k=1:size(A,3)
temp=A(:,:,k);
out(k,:)=temp(idx);
end
time1 = toc;
%% More efficient method:
out2 = nan(size(A,3),size(row_col_idx,1));
tic;
for jj = 1:size(row_col_idx,1)
out2(:,jj) = [A(row_col_idx(jj,1),row_col_idx(jj,2),:)];
end
time2 = toc;
fprintf('Time calculation 1: %d\n',time1);
fprintf('Time calculation 2: %d\n',time2);
Gives as output:
Time calculation 1: 1.954714e+01
Time calculation 2: 2.998120e-01
I have a n x m array (could be any size array but it will not be a 1 x m) and I want to rotate / shift each square loop individually no matter the array size.
How can I alternate the rotation / shift each square loop no matter the size of the array.
Please note: I'm not trying to calculate the values in the array but shift the values.
My thought process was to get the values of each "square loop" and place them into one row and do a circshift then place them back into another array.
I ran into problems trying to get the values back into the original n x m array size and I wasn't sure how I could loop through the process for different n x m arrays.
The pink highlighted section, left of the arrows is the starting position of the array and it's "loops" and the green highlighted section, right of the arrows is the type of rotation / shift of the values that I'm trying to create. The array could have more than 3 "loops" this is just an example.
Code below:
I=[1:5;6:10;11:15;16:20;21:25;26:30]
[rw,col] = size(I);
outer_1=[I(1,:),I(2:end-1,end).',I(end,end:-1:1),I(end-1:-1:2,1).'] %get values in one row (so I can shift values)
outer_1_shift=circshift(outer_1,[0 1]) %shift values
new_array=zeros(rw,col);
Ps: I'm using Octave 4.2.2 Ubuntu 18.04
Edit: The circshift function was changed for Octave 5.0, the last edit made it compatible with previous versions
1;
function r = rndtrip (n, m, v)
rv = #(x) x - 2 * (v - 1);
r = [v * ones(1,rv(m)-1) v:n-v+1 (n-v+1)*ones(1,rv(m)-2)];
if (rv(m) > 1)
r = [r n-v+1:-1:v+1];
endif
endfunction
function idx = ring (n, m , v)
if (2*(v-1) > min (n, m))
r = [];
else
r = rndtrip (n, m, v);
c = circshift (rndtrip (m, n, v)(:), - n + 2 * v - 1).';
idx = sub2ind ([n m], r, c);
endif
endfunction
# your I
I = reshape (1:30, 5, 6).';
# positive is clockwise, negative ccw
r = [1 -1 1];
for k = 1:numel(r)
idx = ring (rows(I), columns(I), k);
I(idx) = I(circshift(idx(:), r(k)));
endfor
I
gives
I =
6 1 2 3 4
11 8 9 14 5
16 7 18 19 10
21 12 13 24 15
26 17 22 23 20
27 28 29 30 25
run it on tio
So, I had the same idea as in Andy's comment. Nevertheless, since I was already preparing some code, here is my suggestion:
% Input.
I = reshape(1:30, 5, 6).'
[m, n] = size(I);
% Determine number of loops.
nLoops = min(ceil([m, n] / 2));
% Iterate loops.
for iLoop = 1:nLoops
% Determine number of repetitions per row / column.
row = n - 2 * (iLoop - 1);
col = m - 2 * (iLoop - 1);
% Initialize indices.
idx = [];
% Add top row indices.
idx = [idx, [repelem(iLoop, row).']; iLoop:(n-(iLoop-1))];
% Add right column indices.
idx = [idx, [[iLoop+1:(m-(iLoop-1))]; repelem(n-(iLoop-1), col-1).']];
if (iLoop != m-(iLoop-1))
% Add bottom row indices.
idx = [idx, [repelem(m-(iLoop-1), row-1).'; (n-(iLoop-1)-1:-1:iLoop)]]
end
if (iLoop != n-(iLoop-1))
% Add left column indices.
idx = [idx, [[(m-(iLoop-1))-1:-1:iLoop+1]; repelem(iLoop, col-2).']]
end
% Convert subscript indices to linear indices.
idx = sub2ind(size(I), idx(1, :), idx(2, :));
% Determine direction for circular shift operation.
if (mod(iLoop, 2) == 1)
direction = [0 1];
else
direction = [0 -1];
end
% Replace values in I.
I(idx) = circshift(I(idx), direction);
end
% Output.
I
Unfortunately, I couldn't think of a smarter way to generate the indices, since you need to maintain the right order and avoid double indices. As you can see, I obtain subscript indices with respect to I, since this can be done quite easy using the matrix dimensions and number of loops. Nevertheless, for the circshift operation and later replacing of the values in I, linear indices are more handy, so that's why the sub2ind operation.
Input and output look like this:
I =
1 2 3 4 5
6 7 8 9 10
11 12 13 14 15
16 17 18 19 20
21 22 23 24 25
26 27 28 29 30
I =
6 1 2 3 4
11 8 9 14 5
16 7 18 19 10
21 12 13 24 15
26 17 22 23 20
27 28 29 30 25
I was right, that the "shift direction" changes with every loop?
Hope that helps!
Caution: I haven't tested for generality, yet. So, please report any errors you might come across.
I have an array :
Z = [1 24 3 4 52 66 77 8 21 100 101 120 155];
I have another array:
deletevaluesatindex=[1 3; 6 7;10 12]
I want to delete the values in array Z at indices (1 to 3, 6 to 7, 10 to 12) represented in the array deletevaluesatindex
So the result of Z is:
Z=[4 52 8 21 155];
I tried to use the expression below, but it does not work:
X([deletevaluesatindex])=[]
Another solution using bsxfun and cumsum:
%// create index matrix
idx = bsxfun(#plus , deletevaluesatindex.', [0; 1])
%// create mask
mask = zeros(numel(Z),1);
mask(idx(:)) = (-1).^(0:numel(idx)-1)
%// extract unmasked elements
out = Z(~cumsum(mask))
out = 4 52 8 21 155
This will do it:
rdvi= size(deletevaluesatindex,1); %finding rows of 'deletevaluesatindex'
temp = cell(1,rdvi); %Pre-allocation
for i=1:rdvi
%making a cell array of elements to be removed
temp(i)={deletevaluesatindex(i,1):deletevaluesatindex(i,2)};
end
temp = cell2mat(temp); %Now temp array contains the elements to be removed
Z(temp)=[] % Removing the elements
If you control how deletevaluesatindex is generated, you can instead directly generate the ranges using MATLAB's colon operator and concatenate them together using
deletevaluesatindex=[1:3 6:7 10:12]
then use the expression you suggested
Z([deletevaluesatindex])=[]
If you have to use deletevaluesatindex as it is given, you can generate the concatenated range using a loop or something like this
lo = deletevaluseatindex(:,1)
up = deletevaluseatindex(:,2)
x = cumsum(accumarray(cumsum([1;up(:)-lo(:)+1]),[lo(:);0]-[0;up(:)]-1)+1);
deleteat = x(1:end-1)
Edit: as in comments noted this solution only works in GNU Octave
with bsxfun this is possible:
Z=[1 24 3 4 52 66 77 8 21 100 101 120 155];
deletevaluesatindex = [1 3; 6 7;10 12];
idx = 1:size(deletevaluesatindex ,1);
idx_rm=bsxfun(#(A,B) (A(B):deletevaluesatindex (B,2))',deletevaluesatindex (:,1),idx);
Z(idx_rm(idx_rm ~= 0))=[]
I have a 3D-cell array designated as A{s,i,h}, serving as a store for large amounts of numerical data during a nested-loop portion of my script. Some of the cell entries will be blank [ ], whilst the rest consist of numbers - either singular or in arrays (1 x 10 double etc.):
I want to convert this cell array to a set of 2D matrices.
Specifically, one separate matrix for each value of h (h is always equal 1:3) and one column in each matrix for every value of s. Each column will contain all the numerical data combined - it does not need to be separated by i.
How can I go about this? I ordinarily deal with 3D-cell arrays in this form to produce separate matrices (one for each value of h) using something like this:
lens = sum(cellfun('length',reshape(A,[],size(A,3))),1);
max_length = max(lens);
mat = zeros(max_length,numel(lens));
mask = bsxfun(#le,[1:max_length]',lens);
mat(mask) = [A{:}];
mat(mat==0) = NaN;
mat = sort(mat*100);
Matrix1 = mat(~isnan(mat(:,1)),1);
Matrix2 = mat(~isnan(mat(:,2)),2);
Matrix3 = mat(~isnan(mat(:,3)),3);
However in this instance, each matrix had only a single column. I'm have trouble adding multiple columns to each output matrix.
1. Result in the form of a cell array of matrices (as requested)
Here's one possible approach. I had to use one for loop. However, the loop can be easily avoided if you accept a 3D-array result instead of a cell array of 2D-arrays. See second part of the answer.
If you follow the comments in the code and inspect the result of each step, it's straightforward to see how it works.
%// Example data
A(:,:,1) = { 1:2, 3:5, 6:9; 10 11:12 13:15 };
A(:,:,2) = { 16:18, 19:22, 23; 24:28, [], 29:30 };
%// Let's go
[S, I, H] = size(A);
B = permute(A, [2 1 3]); %// permute rows and columns
B = squeeze(mat2cell(B, I, ones(1, S), ones(1, H))); %// group each col of B into a cell...
B = cellfun(#(x) [x{:}], B, 'uniformoutput', false); %// ...containing a single vector
t = cellfun(#numel, B); %// lengths of all columns of result
result = cell(1,H); %// preallocate
for h = 1:H
mask = bsxfun(#le, (1:max(t(:,h))), t(:,h)).'; %'// values of result{h} to be used
result{h} = NaN(size(mask)); %// unused values will be NaN
result{h}(mask) = [B{:,h}]; %// fill values for matrix result{h}
end
Result in this example:
A{1,1,1} =
1 2
A{2,1,1} =
10
A{1,2,1} =
3 4 5
A{2,2,1} =
11 12
A{1,3,1} =
6 7 8 9
A{2,3,1} =
13 14 15
A{1,1,2} =
16 17 18
A{2,1,2} =
24 25 26 27 28
A{1,2,2} =
19 20 21 22
A{2,2,2} =
[]
A{1,3,2} =
23
A{2,3,2} =
29 30
result{1} =
1 10
2 11
3 12
4 13
5 14
6 15
7 NaN
8 NaN
9 NaN
result{2} =
16 24
17 25
18 26
19 27
20 28
21 29
22 30
23 NaN
2. Result in the form of 3D array
As indicated above, using a 3D array to store the result permits avoiding loops. In the code below, the last three lines replace the loop used in the first part of the answer. The rest of the code is the same.
%// Example data
A(:,:,1) = { 1:2, 3:5, 6:9; 10 11:12 13:15 };
A(:,:,2) = { 16:18, 19:22, 23; 24:28, [], 29:30 };
%// Let's go
[S, I, H] = size(A);
B = permute(A, [2 1 3]); %// permute rows and columns
B = squeeze(mat2cell(B, I, ones(1, S), ones(1, H))); %// group each col of B into a cell...
B = cellfun(#(x) [x{:}], B, 'uniformoutput', false); %// ...containing a single vector
t = cellfun(#numel, B); %// lengths of all columns of result
mask = bsxfun(#le, (1:max(t(:))).', permute(t, [3 1 2])); %'// values of result to be used
result = NaN(size(mask)); %// unused values will be NaN
result(mask) = [B{:}]; %// fill values
This gives (compare with result of the first part):
>> result
result(:,:,1) =
1 10
2 11
3 12
4 13
5 14
6 15
7 NaN
8 NaN
9 NaN
result(:,:,2) =
16 24
17 25
18 26
19 27
20 28
21 29
22 30
23 NaN
NaN NaN
Brute force approach:
[num_s, num_i, num_h] = size(A);
cellofmat = cell(num_h,1);
for matrix = 1:num_h
sizemat = max(cellfun(#numel, A(:,1,matrix)));
cellofmat{matrix} = nan(sizemat, num_s);
for column = 1:num_s
lengthcol = length(A{column, 1, matrix});
cellofmat{matrix}(1:lengthcol, column) = A{column, 1,matrix};
end
end
Matrix1 = cellofmat{1};
Matrix2 = cellofmat{2};
Matrix3 = cellofmat{3};
I don't know what your actual structure looks like but this works for A that is setup using the following steps.
A = cell(20,1,3);
for x = 1:3
for y = 1:20
len = ceil(rand(1,1) * 10);
A{y,1,x} = rand(len, 1);
end
end
In trying to port an algorithm from C# to Matlab I found that Matlab is inefficient at running for loops. As such I want to vectorize the algorithm.
I have following inputs:
lowrange:
[ 00 10 20 30 40 50 ... ]
highrange:
[ 10 20 30 40 50 60 ... ]
These arrays are equal in length.
I now have a third array Values (which could be any length) and for this array I want to count the occurrences of Values elements between lowerange(i) and highrange(i) (You can see I'm coming from a for loop).
The output should be an array of length lowrange/highrange.
So with the above arrays and input LineData:
[ 1 2 3 4 6 11 12 16 31 34 45 ]
I expect to get:
[ 05 03 00 02 01 00 ... ]
I tried the (for me) obvious thing:
LineData(LineData < PixelEnd & LineData > PixelStart)
But that doesn't work because it just checks LineData on an element by element way. It does not try to apply the comparison over all values in LineData.
Unfortunately, I cannot come up with anything else since I'm not yet used to think in a Matlab 'vector' way, let alone knowing all applicable instructions from memory.
As you are looking to do a basic histogram with given edges, you can use Matlabs built-in function histc:
values = [ 1 2 3 4 6 11 12 16 31 34 45 ];
edges = 0:10:60;
histc(values, edges)
ans =
5 3 0 2 1 0 0
For ranges with identical intervals and starting from 0, here's a bsxfun based counting approach -
LineData = [ 1 2 3 4 6 11 12 16 31 34 45 ] %// Input
interval = 10; %// interval width
num_itervals = 6; %// number of intervals
%// Get matches for each interval and sum them within each interval for the counts
out = sum(bsxfun(#eq,ceil(LineData(:)/interval),1:num_itervals))
Output -
LineData =
1 2 3 4 6 11 12 16 31 34 45
out =
5 3 0 2 1 0
Assuming that the last interval would be the one holding the max of input data, you can try out a diff + indexing based approach too -
LineData = [ 1 2 3 4 6 11 12 16 31 34 45 ] %// Input
interval = 10; %// interval width
labels = ceil(LineData(:)/interval); %// set labels to each input entry
df_labels = diff(labels)~=0; %// mark the change of labels
df_labels_pos = find([df_labels; 1]); %// get the positions of label change
intv_pos= labels([true;df_labels]);%// position of each interval with nonzero counts
%// get counts from interval between label position change and put at right places
out(intv_pos) = [ df_labels_pos(1) ; diff(df_labels_pos)];