| MIN(), MAX() with NaN values [message #89051] |
Sat, 19 July 2014 07:24  |
wlandsman
Messages: 743
Registered: June 2000
|
Senior Member |
|
|
The documentation for MIN() says that
"Note: If the MIN function is run on an array containing NaN values and the NAN keyword is not set, an invalid result will occur."
However, most of the time MIN() seems to work fine with NaN values.
IDL> print,!version
{ x86_64 darwin unix Mac OS X 8.3 Nov 15 2013 64 64}
IDL> a = randomn(seed,500,500)
IDL> a[1,1] = replicate(!values.f_nan,4)
IDL> print,min(a)
-4.76949
The invalid results seem to occur when the NaN values are at the edge of the image (or in the first or last value of a vector)
IDL> a = findgen(10)
IDL> a[0] = !values.f_nan
IDL> print,min(a)
NaN
I can't fault the design or documentation of MIN() -- it says to use /NAN when NaN values are present. But we had difficulty debugging our code since our images have a few randomly located NaN values, and so using MIN() without /NaN was working 99% of the time. We have now added /NaN to the code -- and taken the factor of ~3 speed penalty -- so it works 100% of the time.
|
|
|
|
| Re: MIN(), MAX() with NaN values [message #89053 is a reply to message #89051] |
Sat, 19 July 2014 20:00  |
Matthew Argall
Messages: 286
Registered: October 2011
|
Senior Member |
|
|
I noticed the same thing earlier today!
IDL> print, min([1.0, !values.f_nan])
1.00000
IDL> print, max([1.0, !values.f_nan])
1.00000
IDL> print, min([1.0, !values.f_nan], NAN=0)
1.00000
IDL> print, max([1.0, !values.f_nan], NAN=0)
1.00000
IDL> help, !version
** Structure !VERSION, 8 tags, length=104, data length=100:
ARCH STRING 'x86_64'
OS STRING 'darwin'
OS_FAMILY STRING 'unix'
OS_NAME STRING 'Mac OS X'
RELEASE STRING '8.2'
BUILD_DATE STRING 'Apr 10 2012'
MEMORY_BITS INT 64
FILE_OFFSET_BITS
INT 64
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
