Re: [Geopriv] Proposal for uncertainty

This is a reasonably-well known problem and seems similar to the "there is
a 30% chance that it rains today" - it either rains, or it doesn't. The
same can be said about a "home detection" algorithm that is based (for
example) on some inferences (e.g., time of day and last GPS coordinates).
In that context, it does make sense to say "there's an 80% chance that the
caller is at home", which really means "in 80% of cases with the same
amount of information, the person was indeed at home". It's relatively easy
to calculate the weather case (although there's the rumor that weather
forecasts over-predict rain since nobody complains if their BBQ does NOT
get rained out), but probably hard for our case, given the lack of
retrospective ground truth.

Henning

On Fri, May 23, 2014 at 4:12 PM, Martin Thomson <martin.thomson@gmail.com>wrote:

> I spoke about the one remaining issue with -uncertainty yesterday with
> Ray.  That conversation, along with Roger's comments regarding
> automated location determination helped me realize what I think the
> source of the disconnect here is.
>
> Based on this, I've tentatively concluded that the difference between
> civic and geodetic location is not what is at the core of the issue,
> though it's probably true that the issue only manifests with civic
> location information.  Primarily because of our better intuition
> regarding civic addresses.
>
> The core of the concern is that confidence - as a statistical measure
> - doesn't really make any sort of sense for location information that
> is a bald assertion, as opposed to locations that are generated by
> some measurement process (and here I give a nod to Carl for pointing
> out reverse geocoding as an example of a measurement process that ends
> with civic location).  For example, the assertion "I am at home" is
> verifiable and basically either correct or not, usually the former.
>
> It's only when taken in a larger context that terms like "confidence"
> even apply.  In particular, that means automated location
> determination.
>
> I've taken a stab at clarifying this in the draft.
>
> I've also permitted 100% confidence in the schema.  At the same time,
> I've retained the text describing this as impossible, but that text is
> all in the context of the statistical work, so I hope that the
> introductory material makes that clear enough.
>
> The changes are here:
>
> https://github.com/martinthomson/drafts/commit/c2207408af95de74d1f4bb7d790965d69d4b3045
> A readable copy is here:
> http://martinthomson.github.io/drafts/draft-ietf-geopriv-uncertainty.html
>
> Happy as always to take suggestions.  I'd like to think that this
> helps address Marc's concern without needing to cut Section 3.3
> entirely.
>
> _______________________________________________
> Geopriv mailing list
> Geopriv@ietf.org
> https://www.ietf.org/mailman/listinfo/geopriv
>