| Re: Cumulative total [message #12828] |
Fri, 18 September 1998 00:00  |
bowman
Messages: 121
Registered: September 1991
|
Senior Member |
|
|
In article <3602B1F3.210@cdc.noaa.gov>, Andrew Loughe <afl@cdc.noaa.gov> wrote:
> Nice, elegant solution, Eddie.
>> here is the way i do it:
>>
>> IDL> A = findgen(10)
>> IDL> N = n_elements(A)
>> IDL> result = A # (lindgen(N,N) ge transpose(lindgen(N,N)))
>> IDL> print,result
>> 0.00000
>> 1.00000
>> 3.00000
>> 6.00000
>> 10.0000
>> 15.0000
>> 21.0000
>> 28.0000
>> 36.0000
>> 45.0000
In my opinion this is a baroque construction that reveals a shortcoming in
IDL. (Hey, not a major shortcoming. I'm not committing heresy here!)
To compute the cumulative sum of a vector of length N, i.e.,
x_cum[0] = x[0]
FOR i = 1, N-1 DO x_cum[i] = x_cum[i-1] + x[i]
should require N loads, N flops (adds), and N stores. This may not
optimize well, since each result depends on the previous one, but I
suspect most modern Fortran or C compilers would do pretty well.
The IDL approach above requires creating an N^2 matrix filled with
integers, performing an if test on every element of that array and its
transpose, and then performing a matrix-vector multiply (N^2 multiplies
and N^2 adds). What do you do when N = 100,000? It seem a silly way to
do a simple task.
I'm not trying to pick on Eddie. It works for him. I wonder which method
is faster?
Maybe I should write a Fortran function to do it ... yuck -- highly
non-portable.
So, note to RSI: Add CUMULATIVE function to next release of IDL and make
it a good IDL function so that one can specify which dimension of a
possibly multidimensional array to accumulate over, etc. It should be
quite useful with HISTOGRAM.
Ken
--
Kenneth P. Bowman, Professor 409-862-4060
Department of Meteorology 409-862-4466 fax
Texas A&M University bowmanATcsrp.tamu.edu
College Station, TX 77843-3150 Change the AT to @
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
