| Re: Histogram & Cumulative Distribution Functions [message #40736] |
Mon, 30 August 2004 06:52 |
sdj
Messages: 20
Registered: November 2003
|
Junior Member |
|
|
Dear Justin,
Thanks for your help, your tip has indeed solved my problem.
FYI, I also found an alternative function for "value_locate" written
by Martin Schultz.
Regards,
Pepe
*************************************************
Pepe S. D. Juevara
- Risspekt de man and de nature - Ahi -
*************************************************
;----------------------------------------------------------- --
; Name: SEARCH (function)
;
; Purpose: Perform a binary search for the data point closest
; to a given value. Data must be sorted.
;
; Calling Sequence: index = SEARCH(data, value)
;
; Inputs:
; data -> a sorted data vector
; value -> the value to look for
;
; Outputs: The function returns the index of the nearest data point.
;
; Notes: This routine is much faster than WHERE or MIN for
; large arrays. It was written in response to a
newsgroup
; request by K.P. Bowman.
;
; Example:
; test = findgen(10000)
; print, search(test, 532.3)
; ; prints 532
;
; Modification History: mgs, 21 Sep 1998: VERSION 1.00
;
;-
; Copyright (C) 1998, Martin Schultz, Harvard University
; This software is provided as is without any warranty
; whatsoever. It may be freely used, copied or distributed
; for non-commercial purposes. This copyright notice must be
; kept with any copy of this software. If this software shall
; be used commercially or sold as part of a larger package,
; please contact the author to arrange payment.
; Bugs and comments should be directed to mgs@io.harvard.edu
; with subject "IDL routine search"
;----------------------------------------------------------- --
FUNCTION search, data, value
; search first occurence of value in data set
; data must be sorted
; simple error checking on data and value
if (n_elements(value) eq 0) then begin
message,'Must supply sorted data array and value)',/CONT
return, -1
endif
ndat = n_elements(data)
try = fix(0.5*ndat)
step = 0.5*try
; find index of nearest points
while (step gt 1) do begin
if (data[try] gt value) then $
try = try-step $
else $
try = try+step
step = fix(0.5*(step+1))
endwhile
; now get the data point closest to value
; can only be one out of three (try-1, try, try+1)
dummy = min( abs(value-data[try-1:try+1]), location )
return,try+location-1
end
Justin <kf1zr0y02@sneakemail.com> wrote in message news:<Xns9552C28517E5kf1zr0y02sneakemail@18.181.0.25>...
> Ooops. Late on a Friday. I was meaning cdf in several places I wrote pdf.
> Still would have worked mind you. Soz.
>
> So if h is the output of HISTO then:
> cumul = TOTAL(h, /CUMULATIVE)
> tot = TOTAL(FLOAT(h))
> cdf = cumul/tot
>
> To find the 95th percentile use VALUE_LOCATE on the cdf to get the
> index of the array element closest to 0.95
>
> index = VALUE_LOCATE(cdf, 0.95)
>
> If 'l' contains the histo locations then your 95th percentile is at:
> l[index]
>
> Justin <kf1zr0y02@sneakemail.com> wrote in
> news:Xns9552C1E35BA22kf1zr0y02sneakemail@18.181.0.25:
>
>> To get the CDF from a (discrete) PDF use the TOTAL function with the
>> CUMULATIVE keyword:
>>
>> So if h is the output of HISTO then:
>> cumul = TOTAL(h, /CUMULATIVE)
>> tot = TOTAL(FLOAT(h))
>> pdf = cumul/tot
>>
>> To find the 95th percentile use VALUE_LOCATE on the pdf to get the
>> index of the array element closest to 0.95
>>
>> index = VALUE_LOCATE(pdf, 0.95)
>>
>> If 'l' contains the histo locations then your 95th percentile is at:
>> l[index]
>>
>> Make sure you have enough bins in the histogram otherwise the
>> percentile value can be coarse. You could even create a new histogram
>> (just for the cdf calculation) with nbins >= number of data points to
>> give an accuarate percentile value.
>>
>> Hope this helps,
>>
>> Justin
|
|
|
|
| Re: Histogram & Cumulative Distribution Functions [message #40743 is a reply to message #40736] |
Fri, 27 August 2004 16:08  |
Justin[3]
Messages: 8
Registered: November 2003
|
Junior Member |
|
|
Ooops. Late on a Friday. I was meaning cdf in several places I wrote pdf.
Still would have worked mind you. Soz.
So if h is the output of HISTO then:
cumul = TOTAL(h, /CUMULATIVE)
tot = TOTAL(FLOAT(h))
cdf = cumul/tot
To find the 95th percentile use VALUE_LOCATE on the cdf to get the
index of the array element closest to 0.95
index = VALUE_LOCATE(cdf, 0.95)
If 'l' contains the histo locations then your 95th percentile is at:
l[index]
Justin <kf1zr0y02@sneakemail.com> wrote in
news:Xns9552C1E35BA22kf1zr0y02sneakemail@18.181.0.25:
> To get the CDF from a (discrete) PDF use the TOTAL function with the
> CUMULATIVE keyword:
>
> So if h is the output of HISTO then:
> cumul = TOTAL(h, /CUMULATIVE)
> tot = TOTAL(FLOAT(h))
> pdf = cumul/tot
>
> To find the 95th percentile use VALUE_LOCATE on the pdf to get the
> index of the array element closest to 0.95
>
> index = VALUE_LOCATE(pdf, 0.95)
>
> If 'l' contains the histo locations then your 95th percentile is at:
> l[index]
>
> Make sure you have enough bins in the histogram otherwise the
> percentile value can be coarse. You could even create a new histogram
> (just for the cdf calculation) with nbins >= number of data points to
> give an accuarate percentile value.
>
> Hope this helps,
>
> Justin
>
>
>
|
|
|
|
| Re: Histogram & Cumulative Distribution Functions [message #40744 is a reply to message #40743] |
Fri, 27 August 2004 16:04 |
Justin[3]
Messages: 8
Registered: November 2003
|
Junior Member |
|
|
To get the CDF from a (discrete) PDF use the TOTAL function with the
CUMULATIVE keyword:
So if h is the output of HISTO then:
cumul = TOTAL(h, /CUMULATIVE)
tot = TOTAL(FLOAT(h))
pdf = cumul/tot
To find the 95th percentile use VALUE_LOCATE on the pdf to get the index of
the array element closest to 0.95
index = VALUE_LOCATE(pdf, 0.95)
If 'l' contains the histo locations then your 95th percentile is at:
l[index]
Make sure you have enough bins in the histogram otherwise the percentile
value can be coarse. You could even create a new histogram (just for the
cdf calculation) with nbins >= number of data points to give an accuarate
percentile value.
Hope this helps,
Justin
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
