| SMOOTH or TS_SMOOTH [message #82100] |
Tue, 20 November 2012 15:36  |
markjamie
Messages: 30
Registered: November 2012
|
Member |
|
|
I'm looking to calculate moving / windowed averages for a given time series using the last N values to calculate the mean.
Am I right in thinking ts_smooth with the /backward keyword is the right way of doing this? The documentation for ts_smooth doesn't give any formulae in the same way it does for smooth.
Any advice would be greatly appreciated. Many thanks
|
|
|
|
| Re: SMOOTH or TS_SMOOTH [message #82179 is a reply to message #82100] |
Wed, 21 November 2012 06:28  |
lecacheux.alain
Messages: 325
Registered: January 2008
|
Senior Member |
|
|
Le mercredi 21 novembre 2012 14:09:46 UTC+1, mark...@gmail.com a écrit :
> Typing code too quick....corrected version
>
>
>
> X = 50
>
> N = 3
>
> arrayData = dindgen(X)
>
> arrayBackwardMovingMean = dblarr(X)
>
> for i = 0, X - 1 do arrayBackwardMovingMean[i] = mean(arrayData[min([0, i - N] : i)])
A more "IDL" way (untested) might be :
pad = !Values.F_NAN + dindgen(N)
print, mean(reform([pad,reform(rebin([arrayData,pad],X+N,N),(X*N)*N )],X+N+1,N),DIM=2,/NAN)
or, more explicitely:
print,mean( $
reform( $
[pad,reform(rebin([arrayData,pad],X+N,N),(X*N)*N)] $ ;replicate and pad your array N times
,X+N+1,N), $ ;do the shift
DIM=2, $ ;do the average
/NAN) ;handle the bounds
alx.
|
|
|
|
| Re: SMOOTH or TS_SMOOTH [message #82182 is a reply to message #82100] |
Wed, 21 November 2012 05:09 |
markjamie
Messages: 30
Registered: November 2012
|
Member |
|
|
Typing code too quick....corrected version
X = 50
N = 3
arrayData = dindgen(X)
arrayBackwardMovingMean = dblarr(X)
for i = 0, X - 1 do arrayBackwardMovingMean[i] = mean(arrayData[min([0, i - N] : i)])
|
|
|
|
| Re: SMOOTH or TS_SMOOTH [message #82183 is a reply to message #82100] |
Wed, 21 November 2012 05:01 |
markjamie
Messages: 30
Registered: November 2012
|
Member |
|
|
Thanks for your reply.
I've been playing with TS_SMOOTH today and I don't think it's quite what I'm after. As you mention it calculates the average based on a forward or backward forecast of the data. This gives a similar answer to using a mean of the last N-values but not exactly the same.
I was hoping there would be an IDL function to give the average of the last N points in a data series. In essence, a boxcar smoothing where the window is not centered.
For example:
N = 3
arrayData = dindgen(X)
arrayBackwardMovingMean = dblarr(X)
for i = 0, X do arrayBackwardMovingMean[i] = mean(arrayData[min([0, i - N] : i)])
The min function is to avoid using negative indices and incurring an array out of bounds error.
The code above is ok for what I'm after but it didn't feel very 'IDL'.
Mark
|
|
|
|
| Re: SMOOTH or TS_SMOOTH [message #82185 is a reply to message #82100] |
Wed, 21 November 2012 02:38 |
lecacheux.alain
Messages: 325
Registered: January 2008
|
Senior Member |
|
|
Le mercredi 21 novembre 2012 00:36:39 UTC+1, mark...@gmail.com a écrit :
> I'm looking to calculate moving / windowed averages for a given time series using the last N values to calculate the mean.
>
>
>
> Am I right in thinking ts_smooth with the /backward keyword is the right way of doing this? The documentation for ts_smooth doesn't give any formulae in the same way it does for smooth.
>
>
>
> Any advice would be greatly appreciated. Many thanks
TS_SMOOTH(/BACKWARD) uses a N-th order autoregressive model for calculating the (moving) average of the past N points, i.e. doing a weighted average in which weights are computed such that they minimize the difference with the present point. This is usually different from the simple average of the past N-points. It depends on what kind of "moving average" you actually want.
alx.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
