I'm not sure if this actually gets out to the rest of the world, as
I did suggest this method during the general discussion two weeks ago,
so could somebody outside, say, Scandinavia, send me a mail to confirm
that they are seeing this? Bill, do you read me?
Anyway, the method I suggested was to use reform and then rebin with the
/sample keyword to replicate the data along the new dimension.
For the original problem:
> small_array = reform(small_array,1,8)
> newarray = rebin(cos(small_array),300,8,/sample) * sin(large_array)
Or, in the test program supplied from David Landers, it would read:
> arr2 = rebin(reform(arr1,1,Size2,/overwrite),Size1,Size2,/sample)
> arr1 = reform(arr1,Size2,/overwrite)
I simply inserted the above two lines in the loop where the previous lindgen
timing was done, and, surprise! On a DECstation 5000/240 it comes out as:
(The BYTE(i)/INT(i)/LONG(i)/FLOAT(i)/DOUBLE(i) lines now refer to the
REBIN(REFORM(..)..) method, also marked with a *)
Converting Array(135) to Array(213,135) ( 107 reps)
BYTE(i) 0.7 *
BYTE(*) = 1B # BYTE 18.3
BYTE(*) = 1.0 # BYTE 17.6
LONG = 1B # BYTE 4.1
FLOAT = 1.0 # BYTE 2.7
INT(i) 0.7 *
INT(*) = 1 # INT 19.8
INT(*) = 1.0 # INT 18.7
LONG = 1 # INT 4.3
FLOAT = 1.0 # INT 2.6
LONG(i) 0.7 *
LONG(*) = 1L # LONG 19.8
LONG(*) = 1.0 # LONG 25.5
LONG = 1L # LONG 4.2
FLOAT = 1.0 # LONG 2.6
FLOAT(i) 0.7 *
FLOAT(*) = 1.0 # FLOAT 18.6
FLOAT(*) = 1.0 # FLOAT 23.6
FLOAT = 1.0 # FLOAT 2.5
FLOAT = 1.0 # FLOAT 2.5
DOUBLE(i) 0.9 *
DOUBLE(*) = 1.0D # DOUBLE 20.4
DOUBLE(*) = 1.0 # DOUBLE 19.9
DOUBLE = 1.0D # DOUBLE 4.1
DOUBLE = 1.0 # DOUBLE 3.9
% REBIN: Complex expression not allowed in this context: ARR1.
% Execution halted at TEST_ARR </mn/leda/u1/steinhh/IDL/UTIL/idltiming.pro(
35)> (REBIN).
% Called from $MAIN$ .
In other words, it's from 3.5 to 4.3 times faster than using replicate/#.
The drawback, of course, is that REBIN doesn't take complex values.
Circumnavigation by doing:
arr1 = reform(arr1,1,Size2)
arr2 = complex(rebin(float(arr1),Size1,Size2,/sample),$
rebin(imaginary(arr1),Size1,Size2,/sample))
arr1 = reform(arr1,Size2,/overwrite)
which gives:
COMPLEX(i) 10.5
COMPLEX(*) = (1.0, 0.0) # COMPLEX 20.0
COMPLEX(*) = 1.0 # COMPLEX 19.9
COMPLEX = (1.0, 0.0) # COMPLEX 4.4
COMPLEX = 1.0 # COMPLEX 4.4
Mind you, on a DEC Alpha it looks like:
Converting Array(135) to Array(213,135) ( 107 reps)
BYTE(i) 0.3 *
BYTE(*) = 1B # BYTE 4.0
BYTE(*) = 1.0 # BYTE 4.1
LONG = 1B # BYTE 0.7
FLOAT = 1.0 # BYTE 0.5
INT(i) 0.3 *
INT(*) = 1 # INT 4.4
INT(*) = 1.0 # INT 4.4
LONG = 1 # INT 0.8
FLOAT = 1.0 # INT 0.5
LONG(i) 0.3 *
LONG(*) = 1L # LONG 3.7
LONG(*) = 1.0 # LONG 4.3
LONG = 1L # LONG 0.6
FLOAT = 1.0 # LONG 0.4
FLOAT(i) 0.1 *
FLOAT(*) = 1.0 # FLOAT 3.9
FLOAT(*) = 1.0 # FLOAT 4.2
FLOAT = 1.0 # FLOAT 0.4
FLOAT = 1.0 # FLOAT 0.4
DOUBLE(i) 0.4 *
DOUBLE(*) = 1.0D # DOUBLE 4.0
DOUBLE(*) = 1.0 # DOUBLE 4.1
DOUBLE = 1.0D # DOUBLE 0.6
DOUBLE = 1.0 # DOUBLE 0.4
COMPLEX(i) 0.6 *With fix
COMPLEX(*) = (1.0, 0.0) # COMPLEX 3.6
COMPLEX(*) = 1.0 # COMPLEX 4.5
COMPLEX = (1.0, 0.0) # COMPLEX 0.6
COMPLEX = 1.0 # COMPLEX 0.6
It's a pity that REBIN doesn't take complex values, at least
combined with the /sample keyword. Something for future versions, RSI?
Anyhow, isn't there a way to fool IDL into looking (temporarily) at
a COMPLEX as if it were a DOUBLE? This would make the REBIN(REFORM(..)..)
method better on all platforms/cases.
This method should also work on 2D -> 3D data, whereas the matrix
multiplication method only works for 1D -> 2D data.
Unless a general fix for rebin(/sample) comes from IDL, the indexed method
is the only way for strings, however...
Stein Vidar Haugan
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
