| Re: Optimizing code for faster calculation [message #88033 is a reply to message #88032] |
Thu, 13 March 2014 00:55   |
Helder Marchetto
Messages: 520
Registered: November 2011
|
Senior Member |
|
|
On Thursday, March 13, 2014 7:33:31 AM UTC+1, Kenneth D wrote:
> I've been looking at this block of code now for... ever.
>
>
>
> I've been editing a program created by my Adviser to reduce run time wherever possible. So far I've reduced the run time by nearly half, and I'm trying to juice any performance I can get from absolutely anywhere. My final project will use an array roughly 17,000 by 17,000. And I have to iterate through the program at least 17,000*10 times. If I'm lucky it won't take a month to process my data-sets now. This code is about all I have left to work with:
>
>
>
> exceed_subs = where(min_rmse GT rmse_threshold, counter)
>
> if counter GT 0 then modeled_class(exceed_subs) = "unmodeled"
>
>
>
> min_rmse is an array Float[200], such as [0.347272, 0.312437, 0.360164,...]
>
> rmse_threshold = 0.025
>
> modeled_class is an array String[200], such as ["soil","quag","soil","grass",...]
>
>
>
> The code find the locations where min_rmse is greater than a threshold value, and replaces those index locations in the string array (modeled_class) with "unmodeled".
>
>
>
> This may well be the most efficient way to do this (this code will run a minimum of 17,000 times) but a look at the histograms page at Exelis:
>
> http://www.exelisvis.com/docs/HISTOGRAM.html
>
>
>
> shows:
>
> For example, make the histogram of array A:
>
> H = HISTOGRAM(A, REVERSE_INDICES = R)
>
> ;Set all elements of A that are in the ith bin of H to 0.
>
> IF R[i] NE R[i+1] THEN A[R[R[I] : R[i+1]-1]] = 0
>
>
>
> ;The above is usually more efficient than the following:
>
> bini = WHERE(A EQ i, count)
>
> IF count NE 0 THEN A[bini] = 0
>
>
>
> Which looks so similar to what I'm trying to do. I tried to implement this with no luck (maybe because strings?). Is there anything else I can do? That is, besides taking out iterations, they simply must be there to do what I need.
Hi,
I think you should have a look at
http://www.idlcoyote.com/tips/histogram_tutorial.html
You will find the information you need in there.
That said, my guess is that you will need to set the proper binsize in your histogram command. Depending on the type of values you have, you might try using binsize = 0.025, but I don't have time to check if that is a good option.
Hope it helps.
Cheers,
Helder
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
