| Re: Optimization Question: Sum at each element of array [message #75083] |
Tue, 15 February 2011 12:15 |
![James[2] is currently offline](theme/default/images/xoffline.png.pagespeed.ic.XRkd1fkXye.png "James[2] is currently offline")
James[2]
Messages: 44
Registered: November 2009
|
Member |
|
|
On Feb 14, 9:58 pm, Charles Steinhardt <charles.steinha...@ipmu.jp>
wrote:
> Hello,
>
> I'm trying to optimize the following:
>
> for i=0, 100 do begin
> y = y + myfunc(x, x + sigma * (i-50)/10.0, P[2]*myfunc2(x + sigma
> * (i-50)/10.0), sigma)
> endfor
>
> Here, x, y, and sigma are arrays of the same cardinality. I know the
> for loop is slow in IDL compared to array operations, but I'm having
> problems finding a faster way to do this. Is it really faster to make
> an array of findgen(101) and then do some sort of summation over
> that?
>
> I'm hoping somebody has run into this before - I'd appreciate any
> advice you have!
>
> Thank you,
>
> -Charles
One simple optimization step is to replace:
y = y + (blahblahblah...)
with
y += (blahblahblah...)
This avoids making IDL acquire memory each iteration to store the
result of y + (blahblahblah...).
|
|
|
|
| Re: Optimization Question: Sum at each element of array [message #75100 is a reply to message #75083] |
Tue, 15 February 2011 06:31  |
Jeremy Bailin
Messages: 618
Registered: April 2008
|
Senior Member |
|
|
On Tuesday, February 15, 2011 12:58:02 AM UTC-5, Charles Steinhardt wrote:
> Hello,
>
> I'm trying to optimize the following:
>
> for i=0, 100 do begin
> y = y + myfunc(x, x + sigma * (i-50)/10.0, P[2]*myfunc2(x + sigma
> * (i-50)/10.0), sigma)
> endfor
>
> Here, x, y, and sigma are arrays of the same cardinality. I know the
> for loop is slow in IDL compared to array operations, but I'm having
> problems finding a faster way to do this. Is it really faster to make
> an array of findgen(101) and then do some sort of summation over
> that?
>
> I'm hoping somebody has run into this before - I'd appreciate any
> advice you have!
>
> Thank you,
>
> -Charles
It depends on myfunc and myfunc2. It looks like myfunc currently takes 3 arguments of the same cardinality, and myfunc2 takes one argument? If they can be written in such a way that they can take an extra dimension (of length 101, in this case), then yes, it would be faster to just do a total. So, for example:
xsize = size(x, /dimen)
ndimen = n_elements(xsize)
fullsize = [xsize,101]
xfull = rebin(x, fullsize, /sample)
sigmafull = rebin(sigma, fullsize, /sample)
ifull = rebin( (findgen(101)-50)/10., fullsize, /sample)
y = total( myfunc(xfull, xfull + sigmafull * ifull, $
P[2]*myfunc2(xfull + sigmafull * ifull), sigmafull), ndimen )
-Jeremy.
|
|
|
|
| Re: Optimization Question: Sum at each element of array [message #75102 is a reply to message #75100] |
Tue, 15 February 2011 05:09 |
Brian Daniel
Messages: 80
Registered: July 2009
|
Member |
|
|
You're using the for loop to count over regression iterations.
Without knowing more about myfunc(), it's not clear to me that you'll
be able to do this with only array operations.
Regards,
Brian
On Feb 15, 12:58 am, Charles Steinhardt <charles.steinha...@ipmu.jp>
wrote:
> Hello,
>
> I'm trying to optimize the following:
>
> for i=0, 100 do begin
> y = y + myfunc(x, x + sigma * (i-50)/10.0, P[2]*myfunc2(x + sigma
> * (i-50)/10.0), sigma)
> endfor
>
>
>
> -Charles
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
