"Josef W. Segur" <jsegur@frontiernet.net> skrev i melding
news:rmcbf2tit5spbnap1mlc6rrfjdjmci5nkf@4ax.com...

Martin 53N 1W <ml_news@ddnospamddml1dd.co.uk.dd> wrote:

Knut Arvid Keilen wrote:
[...]

But now then, the current running job (not finished onces, important to
say)
reads the following for the best Gaussian detected :
Power: 1.922769, Fit: 1.326398, Score: -0.15188 (Yes ! A negative number
for
the Score result).

Compare this with my earlier example which was as follows:
Power: 2.306740, Fit 1.337945, Score 2.414156 .

Having a closer look at it, the only thing that puzzles me is the number
for
the Score in the Gaussians (both positive and negative numbers).

Why so ?

Perhaps that negative score has got everyone puzzled...

The Chi-Square Fit should only ever give a positive result, or is there
a version/variation that can give negative scores?...

Or have you found an old bug?!


Aside, note:
http://boinc-wiki.ath.cx/index.php?title=Gaussian
####
"Fit" is a measure of how well a rising and falling signal fits an ideal
Gaussian (bell curve) profile. A lower "fit" value means a better fit.
(It's actually a chi-square fit, i.e., a measure of how far the data
departs from an ideal Gaussian.) Since noise can sometimes randomly
simulate a Gaussian, the SETI@Home system only returns a Gaussian
stronger than 3.2 times the average noise level with a fit less than 10.
####


Keep searchin',
Martin

The calculation of Gaussian score was changed for setiathome_enhanced.
The documentation probably hasn't been updated for that change. Here's
the comment section in the source code:

// Gauss score used for "best of" and graphics.
// This score is now set to be based upon the probability that a signal
// would occur due to noise and the probability that it is shaped like
// a Gaussian (normalized to 0 at thresholds).  Thanks to Tetsuji for
// making me think about this. The Gaussian has 62 degrees of freedom and
// the null hypothesis has 63 degrees of freedom when gauss_pot_length=64;

I've always ignored the score shown for Gaussians since it has never been
used to determine whether the signal should be reported to be put into
the master science database. And although there's a field in the result
file for reporting the score with a Gaussian, it's always 0.

Here is an interesting point I took notice of :

Using Seti@home-MapView 6.54, one of my results in setiathome_enhanced
version 5.15 lists the value -12.000000 in the
Score field for the Gaussian found (or not found) in that delivered job. The
gaussians that are detected while carrying out the workunit
are typically low (slightly higher than -12.000000 . There may perhaps be
some lower values as well, for a reason I really do not know.

The threshold value in the settings for Gaussians are set at 0.4, meaning
that Gaussians having a score (not perhaps a fit as you may think) may be
of an interest. My guess is that the programmers have changed things a
little bit in the enhanced version.

Also the threshold setting for spikes are 0.05 and for pulses 1.02.

So therefore, my current running job may perhaps deliver a quite strong
result when finished (it is only 38.534 % and running quite slow).

The result as of now lists the following:

Spike: power 25.29106, score -0.19909.
Pulse: power 1.259953, period 0.360038 (score 1.028788).
Gaussian: power 2.439109, fit 1.384596 (score 4.908314)

And a triplet as well - power 9.977741 (score 9.977741)

fjernhode