[Kst] [Kst-plot/kst] 5fdad6: Fix a bug in saving despike filters.

[Barth Netterfield]

Cool.

Is this a statistic which you invented?

cbn

On November 20, 2013 08:07:55 PM Nicolas Brisset wrote:

> Hi,
> 
> The "activity level" of a signal is an empiric value computed as the product
> of the standard deviation of the signal with the number of reversals within
> a sliding window.
> 
> I used it a few years ago to do some pilot workload evaluations and it
> worked quite well. It correlates well with human intuition, even if the
> absolute value does not mean much in itself but relative variations are
> interesting. And it has the advantage of being a mathematical, qualitative
> value. You have to tune the parameters (window width and noise level, as
> noise can create many false reversals which pollute the signal) according
> to your signal, but in the end it can produce interesting results.
> 
> Attached to this mail is a plot where you will hopefully get a feeling of
> what it means. At the top you have the signal, then its activity then the
> number of reversals and finally the standard deviation. Activity is the
> product of the latter two.
> 
> The idea is the following:
> - if you have mostly "noise" the amplitude is small and the resulting
> activity is small due to the standard deviation term (200 to 300 s area in
> the plot) - if you have large changes with slow dynamics, the product is
> small due to a low number of reversals (see smaller peaks around 1550 and
> 1580 s in the plot, compared to the larger peak at 1505 s) - if the signal
> has large amplitudes and frequent direction changes, it means that there is
> a lot of activity (see around 1320 s, 1350 and 1505 s)
> 
> Recently a colleague asked me for something similar so I thought I should
> revive it :-) I hope it can be useful to someone else too. I'll commit it
> in the coming minutes, when I've checked that the code is clean.
> 
> Nicolas
> 
> P.S.: 1) if you have a better name, we can change it. I originally
> considered "workload" but I found it even worse. 2) I have tried 2
> approaches for the denoising, one is commented out in the code but feel
> free to test it and change the implementation if results are better.
> 
> ----- Mail original -----
> 
> > De: "Barth Netterfield" <barth.netterfield at utoronto.ca>
> > À: kst at kde.org
> > Envoyé: Mardi 19 Novembre 2013 22:57:24
> > Objet: Re: [Kst] [Kst-plot/kst] 5fdad6: Fix a bug in saving
> > despike	filters.
> > 
> > What is the "activity level of a signal"?

-- 
C. Barth Netterfield
416-845-0946