Re: [PSAMP] PSAMP-INFO IE realtiveError

["Thomas Dietz" <[e-mail]>]

Hi Tanja, Andrew, Paul,

since the topic seems to be settled now can you please summarize your
discussion in this thread? I would like to know what to put into an updated
version of the draft so that it can get published asap.

Best Regards,

Thomas

-- 
Thomas Dietz                 E-mail: [e-mail]
NEC Europe Ltd.              Phone:  [phone]
NEC Laboratories Europe      Fax:    [phone]
Network Research Division
Kurfuersten-Anlage 36
69115 Heidelberg, Germany    http://www.nw.neclab.eu

NEC Europe Limited           Registered in England 2832014
Registered Office: NEC House, 1 Victoria Road, London W3 6BL

> -----Original Message-----
> From: [e-mail] [mailto:[e-mail]] On Behalf
> Of Zseby, Tanja
> Sent: Donnerstag, 14. August 2008 15:03
> To: Andrew Johnson
> Cc: psamp; Juergen Quittek
> Subject: Re: [PSAMP] PSAMP-INFO IE realtiveError
> 
> Hi Andrew,
> 
> great. So it seems that after a lot of mice stepping on and off the
> scale we have a common understanding of the terms :-)
> So to summarize:
> - we don't need an offset IE
> - we definitely include the absoluteError IE
> - we can also include the relativeError IE (maybe you as authors
> decide?)
> 
> Kind regards
> Tanja
> 
> > -----Original Message-----
> > From: [e-mail] [mailto:[e-mail]] On
> Behalf
> > Of Andrew Johnson
> > Sent: Thursday, August 14, 2008 12:13 AM
> > To: Zseby, Tanja
> > Cc: psamp; Juergen Quittek
> > Subject: Re: [PSAMP] PSAMP-INFO IE realtiveError
> >
> > Hi Tanja
> >
> >
> > On Sat, 2008-08-09 at 01:19 +0200, Zseby, Tanja wrote:
> > > Hi Andrew,
> > >
> > > > -----Original Message-----
> > > > From: [e-mail] [mailto:[e-mail]] On
> > > > Behalf Of Andrew Johnson
> > > > Sent: Friday, August 08, 2008 8:48 PM
> > > > To: Zseby, Tanja
> > > > Cc: psamp; Juergen Quittek
> > > > Subject: Re: [PSAMP] PSAMP-INFO IE realtiveError
> > > >
> > > >
> > > > On Fri, 2008-08-08 at 18:07 +0200, Zseby, Tanja wrote:
> > > > > Hi Andrew,
> > > > >
> > > > > lets first forget about the fixed error and say we agree that
> we
> > > need
> > > > > something like an absolute error that defines the maximum error
> > > > > that can happen at each measurement (given the real error is
> > unknown).
> > > > > Then it was unclear for me why you report this together with a
> > > > > relative error which provides exactly the same information but
> > > > > only
> > > > as
> > > > > percentage of the measured value. It is only for convenience
> that
> > > > > we can report either format.
> > > > >
> > > > > relError=abserror/measured value
> > > > >
> > > > >
> > > > > e.g. you can e.g. say: The absolute error is +/- 0.2 kg:
> > > > > Person:   80.50kg +/- 0.2kg
> > > > > Mouse:     0.50 kg +/- 0.2kg
> > > > >
> > > > > That corresponds to the relative errors:
> > > > > Person:   0.249 %
> > > > > Mouse:   40%
> > > >
> > > > In the Accuracy Report Interpretation you only provide one
> accuracy
> > > for
> > > > the field, you don't report the accuracy per measurement.  The
> idea
> > > > is to provide the margin of error for all measurements of a
> certain
> > type.
> > > >
> > > > The only use of the error field types that was originally
> intended
> > > > was in the Accuracy Report Interpretation, where the error is
> > scoped
> > > > to
> > > the
> > > > field (and optionally template).  Unless you use a new template
> per
> > > > record I'm not sure how you would scope the error value to an
> > > > individual measurement.
> > >
> > > But this is absolutely in-line with the above. You can provide one
> > > absolute error for all timestamps or all byte measurements (or all
> > > weight measurements). With this you say what is the maximum error
> > when
> > > measuring the timestamp or bytes. This maximum error usually
> depends
> > > on the measurement method and therefore the absolute error (= the
> > > maximum possible error) is usually the same for all of the values
> > > measured with this method. I think this is exactly the same that
> you
> > > want to express, correct?
> >
> > Ah I see what you mean now.  In general I think that only one type of
> > error will be reported at once and that most of the time that will be
> > an absolute error.
> >
> > I think there is still value in providing a way to express a relative
> > error though.  For example, clocks tend to drift over time so the
> > larger the time measurement the larger the error, i.e. accurate to
> > within 1 second per day is 0.0011574%.
> >
> > Cheers
> >
> > Andrew
> >
> >
> > > >
> > > >
> > > > > Or you could say: The relative error is +/- 10 %. Then you get
> > the
> > > > > corresponding absolute errors:
> > > > > Person:   80.50kg +/- 8.05 kg
> > > > > Mouse:     0.50 kg +/ 0.05 kg
> > > > >
> > > > > If this is o.k., the second question would be:  do we need
> > > > > something like an offset/fixed error ?
> > > > > e.g. Offset: 0.25
> > > > > Person (real value):   80.50kg
> > > > > Person (measured):     80.75kg
> > > > >
> > > > > Mouse (real value):   0.50 kg
> > > > > Mouse (measured):     0.75 kg
> > > > >
> > > > > The only thing that might be confusing is if you have an offset
> > > > > and report it together with a relative error, since the
> relative
> > > > > error should still refer to the real value (without offset).
> But
> > > > > we
> > > > probably
> > > > > do not need the offset value.
> > > >
> > > > I don't think we have any need for an offset.
> > >
> > > o.k. I agree.
> > >
> > > >
> > > > > Hope this was not even more confusing...
> > > >
> > > > I think I understand how you may have misunderstood how the error
> > > > IEs were intended to be used.  I hope I'm making things
> clearer...
> > > > perhaps the draft needs some clarification.
> > >
> > > I think we have the same understanding (see above). The
> absoluteError
> > > gives the maximum value from which a measured value can differ from
> > > the real value. The error is usually bound to the measurement
> method
> > > or system (i.e. the same for subsequent values).
> > >
> > > Kind regards,
> > > Tanja
> > >
> > > >
> > > > Cheers
> > > >
> > > > Andrew
> > > >
> > > >
> > > > > Kind regards
> > > > > Tanja (starting to see white mice)
> > > > >
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Andrew Johnson [mailto:[e-mail]]
> > > > > > Sent: Friday, August 08, 2008 5:14 PM
> > > > > > To: Zseby, Tanja
> > > > > > Cc: Paul Aitken; psamp; Juergen Quittek
> > > > > > Subject: Re: [PSAMP] PSAMP-INFO IE realtiveError
> > > > > >
> > > > > > [SNIP]
> > > > > > > > >> The intention was to say that the clock is 5 minutes
> > > > > > > > >> slow,
> > > > > > > > >> +/-
> > > > > > 10
> > > > > > > > >> seconds - so there's both an absolute error and a
> > > > > > > > >> relative
> > > > > > error.
> > > > > > > > >>
> > > > > > > > >
> > > > > > > > > NOW I finally understand what you meant by fixedError
> > > > > > > > > initially
> > > > > > !!
> > > > > > > > > This I would not consider as error. Any fixed deviation
> I
> > > > > > > > > would
> > > > > > > > rather
> > > > > > > > > name "offset"...
> > > > > > > > > If it is known couldn't you add it to the value and
> > report
> > > > the
> > > > > > > > correct
> > > > > > > > > time?
> > > > > > > >
> > > > > > > > In that case there would be no need to ever report
> > > > "absoluteError"
> > > > > > -
> > > > > > > > because all the original measurements can be corrected
> > > > > > > > before
> > > > > being
> > > > > > > > exported.
> > > > > > >
> > > > > > > Maybe for clarification:
> > > > > > > The absoluteError that I propose is different from what you
> > > > > > > intended by fixedError. absoluteError is a maximum error
> that
> > > you
> > > > > > > would
> > > > > expect
> > > > > > > due to the inaccurate measurement (e.g. the timestamp may
> > vary
> > > by
> > > > > +/-
> > > > > > 5 ms).
> > > > > > > The real error that you make during measurements is unknown
> > > > > > > and can vary. Your fixedError is different. It is a fixed
> and
> > > > > > > known offset
> > > > > > for
> > > > > > > the measured values, correct?
> > > > > >
> > > > > > I think the absoluteError is the same as the originally
> > > fixedError.
> > > > > In
> > > > > > Paul's example above the fixedError was +/- 10 seconds.  I'm
> > not
> > > > > > sure how you would communicate the "5 minutes slow" part...
> > > > > >
> > > > > > The original idea was fixedError would say this is accurate
> to
> > > > > > within
> > > > > X
> > > > > > units.  Both the fixed and the absolute error can be used
> > > together,
> > > > > but
> > > > > > you just have to go with the least accurate one.  For
> example,
> > > > > > if
> > > > my
> > > > > > bathroom scales have a fixed error of 0.25kg and a relative
> > > > > > error
> > > > of
> > > > > > 0.5%, then they can weigh a person very accurate, but are
> > > > > > rubbish for weighing mice:
> > > > > >   Person1:   81.50kg +/- 0.4kg
> > > > > >   Mouse1:     0.25kg +/ 0.25kg
> > > > > >
> > > > > > Cheers
> > > > > >
> > > > > > Andrew
> > > > >
> > >
> _______________________________________________
> PSAMP mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/psamp

_______________________________________________
PSAMP mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/psamp