| Re: percentile with dimension keyword [message #76965] |
Wed, 20 July 2011 13:00 |
JDS
Messages: 94
Registered: March 2009
|
Member |
|
|
SORT_ND, plus Ben's suggestion. Sorting for selection is overkill, but that's what's fast. Reminds me of my favorite numerical recipes quote: "Selection is sorting's austere sister."
JD
|
|
|
|
|
|
| Re: percentile with dimension keyword [message #76977 is a reply to message #76975] |
Wed, 20 July 2011 07:03  |
ben.bighair
Messages: 221
Registered: April 2007
|
Senior Member |
|
|
On 7/20/11 12:53 AM, JP wrote:
> thanks,
>
> What I am after is a function that could be used in an array with 3 dimensions.
> like:
>
> array = Lindgen(1000,1000,100)
> median_array = MEDIAN(array, dimension=3)
>
> the result will be a 2d array of 1000x1000
>
> something like that but for any percentile (the example abobe would give the 50th percentile)
>
> thanks
>
> JP
Hi again,
I think you could use the PERCENTILE function (provided by Kim) or some
variant of it with Craig Markwardt's CMAPPLY function. CMAPPLY accepts
a user defined function name as the operation and the dimension over
which to apply the operation. You can find it here...
http://www.physics.wisc.edu/~craigm/idl/down/cmapply.pro
Something along the lines of (untested) ...
r = CMAPPLY("USER:PERCENTILE", data, 3, functargs = {PERCENT: 95})
Cheers,
Ben
|
|
|
|
| Re: percentile with dimension keyword [message #76979 is a reply to message #76977] |
Tue, 19 July 2011 21:53 |
JP
Messages: 55
Registered: April 2008
|
Member |
|
|
thanks,
What I am after is a function that could be used in an array with 3 dimensions.
like:
array = Lindgen(1000,1000,100)
median_array = MEDIAN(array, dimension=3)
the result will be a 2d array of 1000x1000
something like that but for any percentile (the example abobe would give the 50th percentile)
thanks
JP
|
|
|
|
| Re: percentile with dimension keyword [message #76988 is a reply to message #76979] |
Tue, 19 July 2011 07:34 |
Kim
Messages: 19
Registered: January 2009
|
Junior Member |
|
|
On Jul 19, 9:42 am, Ben Tupper <ben.bigh...@gmail.com> wrote:
>
> Now that I think of it, have you searched for a function called
> percentile.pro? It might be worth looking for that as a starting point.
>
> Cheers,
> Ben
; PERCENTILE: This Function returns the Values of data Corresponding
to input Percentiles
; NOTE: The Values returned are linear interpolates of the
sorted data
; EXAMPLE: DATA=FINDGEN(1001) & PRINT,
PERCENTILE(DATA,PERCENT=[0,0.01,1,10,50,98,99.9,100.0])
FUNCTION PERCENTILE,DATA,PERCENT=PERCENT,MISSING=MISSING,ERROR=error
IF N_ELEMENTS(PERCENT) EQ 0 THEN PERCENT = FINDGEN(101)
INDEX = INTERPOL([0.0,100.0],N_ELEMENTS(DATA)) ; Normalize INDEX
FROM 0 to 100 percent
VALUES = INTERPOL(DATA(SORT(DATA)) ,INDEX , PERCENT)
IF N_ELEMENTS(VALUES) EQ 1 THEN RETURN, VALUES(0) ELSE RETURN,
VALUES
END
|
|
|
|
| Re: percentile with dimension keyword [message #76992 is a reply to message #76988] |
Tue, 19 July 2011 06:42 |
ben.bighair
Messages: 221
Registered: April 2007
|
Senior Member |
|
|
On 7/18/11 9:31 PM, JP wrote:
> Dear IDLers,
>
> The MEDIAN (i.e. 50th percentile) function has a dimension keyword.
> I need to calculate other percentiles (e.g. 5th, 95th) in a large 3D array and don't want to loop.
> Does anyone know of a function that could do that in a similar way to MEDIAN?
Hi,
Since the median operates in a rank like way, I wonder if you could
simple sort the original and determine the quantiles you want by
relative order. Perhaps like this?
x = randomn(s, 3,4,5) ; some array
ix = sort(x) ; it's ascending order indices
sx = x[ix] ; the original sorted
n = size(ix,/N_ELEMENTS) ; the number of items
p = [0.05, 0.5, 0.95] ; your quantiles (5th, 50th, and 95th)
ip = n * p ; quantiles as indices into ix
v = sx[ip] ; the quantiles as values
print, "sorted original ", sx
print, "the median ", median(x)
print, "the quantiles 5th 50th and 95th ", v
I might have fuzzy thinking on the correct way to handle the indices,
but I think it is a good start. The next step is to wrap the above in a
function that takes the array, the quantile you specify and the
dimension to operate on as arguments.
Now that I think of it, have you searched for a function called
percentile.pro? It might be worth looking for that as a starting point.
Cheers,
Ben
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
