| Re: Rebin/Reform/Histogram [message #72633] |
Wed, 22 September 2010 02:21 |
Mrunmayee
Messages: 24
Registered: May 2009
|
Junior Member |
|
|
On Sep 22, 1:41 am, chris <rog...@googlemail.com> wrote:
> On 21 Sep., 15:45, Mrunmayee <gaur...@gmail.com> wrote:
>
> sf = rebin(/sample,reform(/over,transpose(dt),1,n2,n1),nv,n1,n2) # $
> (rebin(/sample,reform(w1, nv, 1, n1,/over),nv, n2, n1) * $
> transpose(rebin(/sample,reform(w2, nv, 1, n2,/over),nv, n1, n2),
> [0,2,1]))
>
> However, if you have large matrices then you will run into memory
> problems. Just reduce redundancy and maximise the work within the
> loop. Then you won't "feel" the loop overhead. The 'over' keyword
> transforms the matrices fast in place - so keep it in mind if you like
> to use them later.
Thanks, Chris, I tried this. And I did run into memory problem. I see
what you did above and I was looking for just something like it. I
just needed to tweak nv, n1, n2 in rebin commands to match dimensions
(it doesn't match as you have written). So for nv = 231, n1 = 1547 and
n2 = 1537, I cannot rebin w1 (nv x n1 size) without running into "%
Array has too many elements."
I just read the tip from David about "Memory used to subscript
arrays". I am thinking of keeping the loop, but instead of using "*"
for the subscript, I can just use a pre-created index array. Just to
save some memory. But I probably won't be able to rebin to the desired
dimensions anyway.
If there is a way around, do let me know.
|
|
|
|
| Re: Rebin/Reform/Histogram [message #72641 is a reply to message #72633] |
Tue, 21 September 2010 13:41  |
rogass
Messages: 200
Registered: April 2008
|
Senior Member |
|
|
On 21 Sep., 15:45, Mrunmayee <gaur...@gmail.com> wrote:
> One wrong statement in the loop:
>
>
>
>> for i = 0, n1-1 do begin
>>>> w = rebin(w1[*,i], nv, n2) ; Makes nvxn2 matrix
>> p = w*w2 ; Performs number 3 above.
>> SF[i,*] = dt # p
>> endfor
Hi, maybe you have to do this in 3D. It can look something like the
following (no proof-just an idea):
sf = rebin(/sample,reform(/over,transpose(dt),1,n2,n1),nv,n1,n2) # $
(rebin(/sample,reform(w1, nv, 1, n1,/over),nv, n2, n1) * $
transpose(rebin(/sample,reform(w2, nv, 1, n2,/over),nv, n1, n2),
[0,2,1]))
However, if you have large matrices then you will run into memory
problems. Just reduce redundancy and maximise the work within the
loop. Then you won't "feel" the loop overhead. The 'over' keyword
transforms the matrices fast in place - so keep it in mind if you like
to use them later.
Regards
CR
|
|
|
|
| Re: Rebin/Reform/Histogram [message #72648 is a reply to message #72641] |
Tue, 21 September 2010 06:45 |
Mrunmayee
Messages: 24
Registered: May 2009
|
Junior Member |
|
|
One wrong statement in the loop:
> for i = 0, n1-1 do begin
>>> w = rebin(w1[*,i], nv, n2) ; Makes nvxn2 matrix
> p = w*w2 ; Performs number 3 above.
> SF[i,*] = dt # p
> endfor
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
