how to velocize it (vectorizing)

a=magic(10)
b=[4;5;9;2;3;4;2;7;4;9]  %width(a) element
x=ones(size(a));
for i=1:numel(b)
    x(1:max(b(i)-1,1),i)=0
end

Answer 1

[Bruno Luong]

I don't know why a is matter beside that the first dimension is 10

a=magic(10);
b=[4;5;9;2;3;4;2;7;4;9];  %width(a) element

h = size(a,1);
x = double(ndgrid(1:h,b)>=b(:)')

x = 10×10

0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 1 1 0 1 0 0 0 1 0 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 1 1 0 1 1 0 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

;

[aldo]

i used magic(10) to build a matrix in a short time but the matrix that I will have to deal with has height<>width.. Do you think it will work the same? I used a matrix that has n<>m but the result is different

I probably set the problem wrong

this is code correct..excuse me

a=20;
b=[4;5;9;2;3;4;2;7;4;9]  %width(a) element

x=ones(a,numel(b))


for i=1:numel(b)
   x(1:max(b(i)-1,1),i)=0
end

[Bruno Luong]

The only thing my code requires is

numel(b) == width(a)

which obviously what you also have

b=[4;5;9;2;3;4;2;7;4;9]  %width(a) element

[Bruno Luong]

a=200;%it's an example
x=ones(a,numel(b))

now you change a to be a size NOT a matrix as in your original question.

In my code I call it "h" and not "a".

How hard to be self consistent?

[aldo]

I redid the example with different variables.. It is often not easy to convey your problem to others because you have to transform the numerical arrays you use with simpler examples now i saw how my code differs from yours: with max(b)<h is equal.. with max(b)>=h it is different

h=10;
b=[4;5;9;2;3;4;2;7;4;9];

x=ones(h,numel(b))
for i=1:numel(b)
   x(1:max(b(i)-1,1),i)=0;
end

x1=ones(h,numel(b))
%h = size(a,1);
x1 = double(ndgrid(1:h,b)>=b(:)');

isequal(x,x1)

ans =

  logical

   1
   
 
   with 
   h=7;
b=[4;5;9;2;3;4;2;7;4;9];

ans =

  logical

   0


 second problem:
 i try to test speed code and i see it: ( with MATRIX 5710 x 82)
     
Elapsed time is 0.001608 seconds. (MY CODE)
Elapsed time is 0.003299 seconds. (Your Code)

size(x)
ans =
        5710          82
ans =


   
   
   

[Bruno Luong]

" with max(b)<h is equal.. with max(b)>=h it is different"

because in that case your code grows x beyond the initial dimension of height h. Your final x will have more than h rows. Do you really want to do that?

h=7

h = 7

b=[4;5;9;2;3;4;2;7;4;9];
x=ones(h,numel(b));
for i=1:numel(b)
    x(1:max(b(i)-1,1),i)=0;
end
x

x = 8×10

0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 1 1 0 1 0 0 0 1 0 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0

x1 = double(ndgrid(1:h,b)>=b(:)')

x1 = 7×10

0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 1 1 0 1 0 0 0 1 0 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 0 1 0 1 1 0 1 1 1 1 1 1 0

isequal(x(1:h,:), x1)

ans = logical
   1

As for speed I could possibly speed up, but who tells you that vectorized code must be faster?

[aldo]

"because in that case your code grows x beyond the initial dimension of height h. Your final x will have more than h rows. Do you really want to do that?"

you are right...it is correct to keep the original size of the matrix

"As for speed I could possibly speed up, but who tells you that vectorized code must be faster?"

I was convinced that such a simple code could save time compared to a loop Maybe because "ndgrid" is a resource intensive instruction?

I omitted to write at the beginning that the matrix "x" will be used of type Logical

[Bruno Luong]

Here is the timing on bigger arrays (I speed up the methods based on what you said about logical).

Yes for-loop is faster on online server (on my compter I get the opposite). The persons who claim for-loop is slow can come here and look or try to beat it.

h = 5710;
b=randi([0 h+1], 1, 82);

isequal(forloop(h, b), cmpb(h, b))

ans = logical
   1

t_forloop = timeit(@()forloop(h, b), 1)

t_forloop = 1.6679e-04

t_cmpb = timeit(@()cmpb(h, b), 1)

t_cmpb = 2.5579e-04

function x = forloop(h, b)
x=true(h,numel(b));
for i=1:numel(b)
    x(1:max(b(i)-1,1),i)=false;
end
end

function x = cmpb(h, b)
x = (1:h)'>=b(:)';
end

[aldo]

"Yes for-loop is faster on onlive server (on my compter I get the opposite)." i don't use online server but my pc "Processore AMD Ryzen 9 5950X 16-Core Processor, 3401 Mhz, 16 core, 32 processori logici"

I think is better to use loop in my pc because it work with parallel processor

I've noticed dramatic improvements when it comes to vectorizing code where there is a loop inside another

thanks for taking the time

[Bruno Luong]

"I've noticed dramatic improvements when it comes to vectorizing code where there is a loop inside another"

Not really. The speed depens what you do in the body of the loop(s), not loop are nested or not.

I know what I'm talking on speeding MATLAB code.