MATLAB Central Newsreader - Need to make code more Efficient

Need to make code more Efficient

Fraser Dickson — Tue, 09 Feb 2010 10:36:04 -0500

Hi i have a piece of code that looks through a data set and works out the average value for each cell based on the values sourrounding that cell ( 10 cells above and 10 cells below averaged). The data set is extremely large ( roughly 500 x 40) so i was wondering if anyone could reccomend how i can speed up my code. I have a lot of for loops and im sure there is nothing efficient about that !

Thanks

average = zeros(500,40);
r1 = zeros(kd,1);
r2 = zeros(kd,1);

for x = 1:40
        for y = 1 : 500

            d_under_test = x+kd+cd;
            r_under_test = y +kr+cr;

            for counter = 1:kd
                r1(counter) = temp_array((d_under_test + cd +counter ),(r_under_test));
            end

            for counter = 1:kd
                r2(counter) = temp_array((d_under_test - cd -counter ),(r_under_test));
            end

            r2mean = mean(r2);
            r1mean = mean(r1);
            average(x,y) = 0.5*(r1mean+r2mean);
     end
    end

Re: Need to make code more Efficient

Oleg Komarov — Tue, 09 Feb 2010 12:49:02 -0500

"Fraser Dickson"
> Hi i have a piece of code that looks through a data set and works out the average value for each cell based on the values sourrounding that cell ( 10 cells above and 10 cells below averaged). The data set is extremely large ( roughly 500 x 40) so i was wondering if anyone could reccomend how i can speed up my code. I have a lot of for loops and im sure there is nothing efficient about that !
>
> Thanks
>
> average = zeros(500,40);
> r1 = zeros(kd,1);
> r2 = zeros(kd,1);
>
> for x = 1:40
> for y = 1 : 500
>
> d_under_test = x+kd+cd;
> r_under_test = y +kr+cr;
>
>
>
> for counter = 1:kd
> r1(counter) = temp_array((d_under_test + cd +counter ),(r_under_test));
> end
>
> for counter = 1:kd
> r2(counter) = temp_array((d_under_test - cd -counter ),(r_under_test));
> end
>
> r2mean = mean(r2);
> r1mean = mean(r1);
> average(x,y) = 0.5*(r1mean+r2mean);
> end
> end

"The data set is extremely large ( roughly 500 x 40) ."
Huge is when you have a 100e6 by 20 matrix...

Plz post the result of:
whos kd
whos cd

You should definitely forget your approach. Read this useful doc:
http://www.mathworks.com/support/tech-notes/1100/1109.html

Oleg

Re: Need to make code more Efficient

Fraser Dickson — Tue, 09 Feb 2010 13:08:02 -0500

Hi

cd = 2
kd = 16

Basically the code looks at 16 values either side of the cell under test and returns the average ( ignoring the 2 cells ( kd) imeediate to the cell) so that will be (500*40) of doing this averaging routine

the code is taking at least 15 seconds to run through which is way to long i need to speed it up. I thought i was using vectorized loops could you explain what i am doing wrong?

Thanks

Re: Need to make code more Efficient

Walter Roberson — Tue, 09 Feb 2010 15:42:36 -0500

Fraser Dickson wrote:
> Hi i have a piece of code that looks through a data set and works out
> the average value for each cell based on the values sourrounding that
> cell ( 10 cells above and 10 cells below averaged).

Sounds like a job for blkproc()

Re: Need to make code more Efficient

Oleg Komarov — Tue, 09 Feb 2010 15:52:04 -0500

> I thought i was using vectorized loops could you explain what i am doing wrong?
>
> Thanks
Have you given a look at the link I posted?

Vectorized ~= Loop in matlab approach.

Oleg

Re: Need to make code more Efficient

Fraser Dickson — Tue, 09 Feb 2010 20:12:18 -0500

I dont realy understand how i could do what i need to do without using a loop???

Ive never heard of this blcproc() func before i shall look into that

Re: Need to make code more Efficient

Walter Roberson — Tue, 09 Feb 2010 22:29:27 -0500

Fraser Dickson wrote:

> for counter = 1:kd
> r1(counter) = temp_array((d_under_test + cd +counter
> ),(r_under_test));
> end
>
> for counter = 1:kd
> r2(counter) = temp_array((d_under_test - cd -counter
> ),(r_under_test));
> end

r2_r1 = temp_array(d_under_test - cd - kd : ...
d_under_test + cd + kd, ...
r_under_test);
r2mean = mean(r2_r1(1:kd));
r1mean = mean(r2_rw(end-kd-1:end));
average(x,y) = (r1mean + r2mean) / 2;

Note: because kr and cr do not vary according to y, you can vectorize to do an
entire column at one time.

Also note that you could move the initialization of d_under_test to outside of
the 'for y' loop.

> r2mean = mean(r2);
> r1mean = mean(r1);
> average(x,y) = 0.5*(r1mean+r2mean);
> end
> end