I have a matrix B and I would like to obtain a new matrix C from B by adding its last w*a rows to the first w*a rows (w and a will be defined afterwards).
My matrix B is generally defined by :
I would like to obtain matrix C defined in a general way by:
The characteristics of matrices B and C are:
L and w are defined real values;
B0,B1,...,Bw are of dimension: a by b;
B is of dimension: [(L+w)×a] by (L×b);
C is of dimension: (L×a) by (L×b).
Example: For L = 4 and w = 2 I obtain the following matrix B:
The w*a = 2*1 = 2 last rows of B are:
The w*a = 2*1 = 2 first rows of B are:
By adding the two matrices we have:
The matrix C thus obtained is then:
For B0 = [1 0], B1 = [0 1] and B2 = [1 1]. We obtain :
B0, B1 and B2 are of dimension a by b i.e. 1 by 2;
B is of dimension: [(L+w )×(a)] by (L×b) i.e. [(4+2)×1] by (4×2) i.e. 6 by 8;
C is of dimension: (L×a) by (L×b) i.e. (4×1) by (4×2) i.e. 4 by 8.
The matrices B and C that I get are as follows:
B =
1 0 0 0 0 0 0 0
0 1 1 0 0 0 0 0
1 1 0 1 1 0 0 0
0 0 1 1 0 1 1 0
0 0 0 0 1 1 0 1
0 0 0 0 0 0 1 1
C =
1 0 0 0 1 1 0 1
0 1 1 0 0 0 1 1
1 1 0 1 1 0 0 0
0 0 1 1 0 1 1 0
I would like to have some suggestions on how to program this construction so that from a given matrix B I can deduce the matrix C.
Matlab's range indexing should help you do this in a few steps. The key things to remember are that ranges are inclusive, i.e. A[1:3] is a three 3x1 matrix, and that you can use the keyword end to automatically index the end of the matrix row or column.
%% Variables from OP example
w = 2;
L = 4;
B0 = [1 0];
B1 = [0 1];
B2 = [1 1];
[a, b] = size(B0);
% Construct B
BX = [B0;B1;B2]
B = zeros((L+w)*a, L*b);
for ii = 0:L-1
B(ii+1:ii+w+1, ii*b+1:ii*b+b) = BX;
end
%% Construct C <- THIS PART IS THE ANSWER TO THE QUESTION
% Grab first rows of B
B_first = B(1:end-w*a, :) % Indexing starts at first row, continues to w*a rows before the end, and gets all columns
% Grab last rows of B
B_last = B(end-w*a+1:end, :); % Indexing starts at w*a rows before the end, continues to end. Plus one is needed to avoid off by one error.
% Initialize C to be the same as B_first
C = B_first;
% Add B_last to the first rows of C
C(1:w*a, :) = C(1:w*a, :) + B_last;
I get the output
C =
1 0 0 0 0 0 1 1 0 1
0 1 1 0 0 0 0 0 1 1
1 1 0 1 1 0 0 0 0 0
0 0 1 1 0 1 1 0 0 0
0 0 0 0 1 1 0 1 1 0
Related
I have a given matrix H and I would like to unfold (expand) it to find a matrix B by following the method below :
Let H be a matrix of dimension m × n. Let x = gcd (m,n)
The matrix H is cut in two parts.
The cutting pattern being such that :
The "diagonal cut" is made by alternately moving c = n/x units to the right (we move c units to the right several times).
We alternately move c-b = m/x units down (i.e. b = (n-m)/x) (we move b units down several times).
After applying this "diagonal cut" of the matrix, we copy and paste the two parts repeatedly to obtain the matrix B.
Exemple : Let the matrix H of dimension m × n = 5 × 10 defined by :
1 0 1 1 1 0 1 1 0 0
0 1 1 0 0 1 1 0 1 1
1 1 0 1 1 1 0 1 0 0
0 1 1 0 1 0 1 0 1 1
1 0 0 1 0 1 0 1 1 1
Let's calculate x = gcd (m,n) = gcd (5,10) = 5,
Alternatively move to the right : c = n/x = 10/5 = 2,
Alternatively move down : b = (n-m)/x = (10-5)/5 = 1.
Diagonal cutting diagram : The matrix H is cut in two parts.
The cutting pattern is such that :
We move c = 2 units to the right several times c = 2 units to the right,
We repeatedly move c - b = 1 unit downwards.
We get :
After applying this "diagonal cut" of the matrix, we copy and paste the two parts repeatedly to obtain the matrix :
Remark : In the matrices X, X1 and X2 the dashes are zeros.
The resulting matrix B is (L is factor) :
Any suggestions?
This can be done by creating a logical mask with the cutting pattern, and then element-wise multiplying the input by the mask and by its negation. Repeating by L can be done with blkdiag.
H = [1 0 1 1 1 0 1 1 0 0
0 1 1 0 0 1 1 0 1 1
1 1 0 1 1 1 0 1 0 0
0 1 1 0 1 0 1 0 1 1
1 0 0 1 0 1 0 1 1 1];
L = 2;
[m, n] = size(H);
x = gcd(m, n);
c = n / x;
b = (n-m)/x;
mask = repelem(tril(true(m/b)), b, c);
A = [H.*mask; H.*~mask];
A = repmat({A}, L, 1);
B = blkdiag(A{:});
I have a matrix
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
and b vector
b =
1 2 3 4 5 5
I want to replace value of each row in a matrix with reference value of b matrix value and finally generate a matrix as follows without using for loop.
a_new =
1 0 0 0 0
0 1 0 0 0
0 0 1 0 0
0 0 0 1 0
0 0 0 0 1
0 0 0 0 1
if first element of b, b(1) = 1 so change take first row of a vector and make first element as 1 because b(1) = 1.
How can I implement this without using for loop?
Sure. You only need to build a linear index from b and use it to fill the values in a:
a = zeros(6,5); % original matrix
b = [1 2 3 4 5 5]; % row or column vector with column indices into a
ind = (1:size(a,1)) + (b(:).'-1)*size(a,1); % build linear index
a(ind) = 1; % fill value at those positions
Same as Luis Mendo's answer, but using the dedicated function sub2ind:
a( sub2ind(size(a),(1:numel(b)).',b(:)) ) = 1
Also via the subscript to indices conversion way,
a = zeros(6,5);
b = [1 2 3 4 5 5];
idx = sub2ind(size(a), [1:6], b); % 1:6 just to create the row index per b entry
a(idx) = 1
any of these methods works in Octave:
bsxfun(#eq, [1:5 5]',(1:5))
[1:5 5].' == (1:5)
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