Re: [Gnuplot-info] Getting Gnuplot to ignore outliers

[Amir S. <ami...@gm...>]

I see it as a probability problem:
If you suppose a normal distribution, you calculate the mean, then the
standard deviation sigma. If a value is over 2 sigma you have 95%
confidence that the value is wrong. Then you can build a rule that excludes
that value with N/A

Hope this help,
Amir
Il 31/ott/2014 20:31 "Dave Horsfall" <da...@ho...> ha scritto:

> Consider the following made-up sequence:
>
> 1 2 2 1 3 3 1 10 2 4 1
>
> I have a simple script which mashes out some stats, and amongst other
> things it reports the standard deviation as 2.49 (because of that outlier
> of 10).
>
> If I ignore it, I get a standard deviation of 1, which is a lot closer to
> reality.
>
> Is there some way to get Gnuplot to ignore outliers and substitute
> "missing", or do I have to write some sort of a filter for it?  I can
> understand that the definition of "outlier" is a loose concept, but I'm
> thinking of defining an expected range for that particular column.
>
> The background here is that I've started analysing my daily spam load;
> there's a very-much-prototype graph over on www.horsfall.org/plotspam.pdf
> (80k), and I would very much like to auto-ignore the spike for "Banner"
> (the green one; it's those morons at Security Focus not being able to
> handle my SMTP banner and retrying immediately) and the spike for "Reject"
> (the red one; it's the idiots who treat it as a challenge).
>
> As an aside, I found I had to use a log scale because of the magnitudes
> involved (many attempts but few successes), but it actually turned out
> quite well.
>
> --
> Dave Horsfall (VK2KFU)  "Bliss is a MacBook with a FreeBSD server."
> http://www.horsfall.org/spam.html (and check the home page whilst you're
> there)
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> gnuplot-info mailing list
> gnu...@li...
> Membership management via:
> https://lists.sourceforge.net/lists/listinfo/gnuplot-info
>