| Re: how to calculate a running total of a vector [message #3213] |
Mon, 05 December 1994 02:33 |
sjt
Messages: 72
Registered: November 1993
|
Member |
|
|
Chris Chase S1A (chase@retro.jhuapl.edu) wrote:
: In article <3bnmri$e91@news.doit.wisc.edu> VUKOVIC@uwmfe.neep.wisc.edu (Mirko Vukovic) writes:
: Mirko> Suppose I have a vector v. I want to generate a vector vv
: Mirko> whose i-th element is a sum of the first i elements of v. How
: Mirko> to do it fast -- without loops?
: Mirko> I thought of multiplying a matrix with 1's on and bellow its
: Mirko> diagonal by v. That would do it, but seems like a bit of
: Mirko> overkill.
: This will work, but it is overkill also. I probably does the same
: amount of computations as your matrix multiply solution.
: running_sum = (convol([0,v],replicate(1.,n_elements(v)),center=0,/edge_tru n))(1:*)
: Chris Chase
: --
: ===============================
: Bldg 24-E188
: The Applied Physics Laboratory
: The Johns Hopkins University
: Laurel, MD 20723-6099
: (301)953-6000 x8529
: chris.chase@jhuapl.edu
I recall trying various things for this one a while back (on about
IDL2.0.13 or thereabouts, running on a VAX) and found the quickest was a
single loop:
if DA is your data array:
PS = fltarr(n_elements(DA))
ps(0)=da(0)
for j=1l,n_elements(da)-1 do ps(j)=ps(j-1)+da(j)
I know IDL loops are slow but the other methods tend to have so much
overhead of memory allocation etc that it just won. (That said, I haven't
tried Chris's method)
--
+------------------------+---------------------------------- --+---------+
| James Tappin, | School of Physics & Space Research | O__ |
| sjt@star.sr.bham.ac.uk | University of Birmingham | -- \/` |
| "If all else fails--read the instructions!" | |
+----------------------------------------------------------- --+---------+
|
|
|
|
| Re: how to calculate a running total of a vector [message #3215 is a reply to message #3213] |
Sun, 04 December 1994 08:54  |
VUKOVIC
Messages: 6
Registered: December 1994
|
Junior Member |
|
|
In <CHASE.94Dec2163443@retro.jhuapl.edu> chase@retro.jhuapl.edu writes:
>
> Mirko> Suppose I have a vector v. I want to generate a vector vv
> Mirko> whose i-th element is a sum of the first i elements of v. How
> Mirko> to do it fast -- without loops?
>
>
> This will work, but it is overkill also. I probably does the same
> amount of computations as your matrix multiply solution.
>
> running_sum = (convol([0,v],replicate(1.,n_elements(v)),center=0,/edge_tru n))(1|
> :*)
>
Tried your suggestion, and it works. But it is damn slow compared to a simple
loop algorithm (.4 vs 4 sec for a 1024 length vector on an old vaxstation):
FUNCTION RSUMV,V
;+
; Produces a running total of a vector where result(i) is total(v(0:i))
;-
nn=n_elements(v)
res=fltarr(nn)
res(0)=v(0)
for ii=1,nn-1 do res(ii)=res(ii-1)+v(ii)
return,res
end
My final thought is to go maybe fia the FFT. One can maybe approximate the sum
by an integral, and do the integration fia FFT.
FFT(FFT(V,-1)/W,1)
where W is a frequency like vector. I may try that later.
Mirko
Mirko Vukovic | vukovic@uwmfe.neep.wisc.edu
Dept. of Nucl. Eng. and Eng. Phys. | mvukovic@wiscmacc.bitnet
U of Wisconsin -- Madison | phone: (608) 265-4069
1500 Johnson Drive; Madison WI 53706| fax: (608) 265-2364
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
