Re: [chrony-dev] [GIT] chrony/chrony.git branch, master, updated. 1.25-pre1-18-g20a4340

[Bill Unruh <unruh@xxxxxxxxxxxxxx>]

On Wed, 13 Apr 2011, Miroslav Lichvar wrote:

> On Wed, Apr 13, 2011 at 09:01:43AM -0700, Bill Unruh wrote:
> > Hm, whether one used normalised or unnormalised weights should not make a
> > difference to the fit, or the variance in the slope. I looked at
> > that code a few years ago, but have forgotten it by now. Will have
> > to look again.
>
> It doesn't. But in the weights calculation is included variance from
> the previous regression, which can apparently create a positive
> feedback when the weighted average of samples variance is used.
>
> > From the code:
>
> sd = sqrt(inst->variance)
> sd_weight = 1.0 + SD_TO_DIST_RATIO * (peer_distances[i] - min_distance) / sd;
> weights[i] = sd_weight * sd_weight;

That use of sd in the weights is new in the code. It used to be be that the weighting was
purely by the ratio of the excess distance over the min_distance. Not sure
why the sd is in there instead now. Did you add it?
 This equation implies that as sd gets
small, the "1" becomes unimportant and small differences in the peer_distance
become more and more important.

But what is worrying is that the weights are entering into the subsequent
variance, rather than the relative weights. I would have expected just the
relative weights to be important. Will have to look at the Regression code
again. Seems there is a problem there.

>

--
William G. Unruh   |  Canadian Institute for|     Tel: +1(604)822-3273
Physics&Astronomy  |     Advanced Research  |     Fax: +1(604)822-5324
UBC, Vancouver,BC  |   Program in Cosmology |     unruh@xxxxxxxxxxxxxx
Canada V6T 1Z1     |      and Gravity       |  www.theory.physics.ubc.ca/

---
To unsubscribe email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "unsubscribe" in the subject.
For help email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "help" in the subject.
Trouble?  Email listmaster@xxxxxxxxxxxxxxxxxxxx.