Mangled 3d file
Wookey
wookey at wookware.org
Fri Mar 6 08:25:15 GMT 2015
+++ Olly Betts [2015-03-06 05:55 +0000]:
> On Wed, Mar 04, 2015 at 01:22:54PM +0000, Wookey wrote:
> > I wonder if one should change the SD to fit
> > the 'expected' tolerance, or add an explict 'tolerance' warning
> > threshold? The two numbers are related but not actually identical.
>
> I think the current 2 sigma test isn't actually the right test, since
> with a correctly specified SD it's actually expected to have a fairly
> high false positive rate, and that isn't helpful. I've changed that
> to 3 sigma in commit 4dc9899, which should reduce the noise.
That seems like a good plan - cheers. And you've fixed the 'ERROR'
thing too, which I think should solve the practical problem.
> As point out below, an appropriate tolerance may vary with other
> readings on the leg. Perhaps the tolerance should be specifiable as an
> angle or in "sigmas", with the default being "3 sigmas". That would
> allow for the current approach or for automating a project policy on
> the tolerance.
That's a good idea.
> > A large project inevitably collects a lot of fore/back sights that are
> > out of tolerance over time, and thus compiling a big system gives a
> > lot of warnings, which obscures genuine errors in the new survey you
> > just added. Perhaps just having an option (like gcc) to turn off
> > specific warnings is one way to deal with this?
>
> Perhaps an optional flag per leg indicating how to resolve fore vs back
> with options like average, quiet average, use only fore, use only back
> - this provides the ability to "ring the preferred reading". I guess
> it would need to be specifiable for clino and compass separately.
Interesting idea. Currently the data is 'nobbled' by replacing the
duff readings with '-' or more often no-one knows which reading is
best. I did have one trip this year where one distoX was clearly more
wonky than the other. In that case preferring the good one when too
different was a sensible choice.
Being able to say 'yes I've seen the warnings but I don't really care
in future' (i.e 'quiet average') is definitely useful.
Being able to record which inst was preferred is also useful, but that
can also be done with commenting+rewriting. Is it worth adding a new
construct? It'd be neater.
> > (Sighting compasses should strictly have an SD function
> > that takes into account the clino angle (as opposed to a constant)- so
> > there is a step at 15 degrees where it starts to stick, then it rises
> > to several degrees at +89 clino).
>
> There's actually a formula in use which effectively increases the
> expected compass error for steeper legs (in "normal" style), but it
> currently does its job when we calculate the covariance matrix, so the
> SDs used for the tolerance check are currently exactly those specified
> with *SD (or the defaults).
OK - that's good then.
Wookey
--
Principal hats: Linaro, Debian, Wookware, ARM
http://wookware.org/
More information about the Survex
mailing list