WU Inconclusive...Why?


log in

Advanced search

Message boards : Number crunching : WU Inconclusive...Why?

Author Message
Profile cov_route
Avatar
Send message
Joined: 13 Sep 12
Posts: 303
Credit: 7,602,313
RAC: 4,760
Canada
Message 1313404 - Posted: 10 Dec 2012, 3:21:12 UTC

http://setiathome.berkeley.edu/workunit.php?wuid=1109451566

My result:

Spike count: 0 Pulse count: 0 Triplet count: 2 Gaussian count: 0

Wingman's result:

Spike count: 0 Pulse count: 0 Triplet count: 2 Gaussian count: 0

What else feeds into the decision logic?

Profile ignorance is no excuse
Avatar
Send message
Joined: 4 Oct 00
Posts: 9529
Credit: 44,433,321
RAC: 0
Korea, North
Message 1313407 - Posted: 10 Dec 2012, 3:25:10 UTC - in response to Message 1313404.


Flopcounter: 47198076140224.703000


Flopcounter: 55876472160744.562000

notice a difference
____________
In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

End terrorism by building a school

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5079
Credit: 74,105,639
RAC: 7,326
Australia
Message 1313428 - Posted: 10 Dec 2012, 6:01:09 UTC - in response to Message 1313404.
Last modified: 10 Dec 2012, 6:18:28 UTC

What else feeds into the decision logic?


With the various kinds of signals there are a number of parameters checked/compared server side. In the case of those triplets there are known inaccuracies in both the CPU 6.03 application (mostly normalisation summing noise) and in the 6.10 application there is some chirp inaccuracy.

This can mean occasional variation in detected signal characteristics, which 'usually' in most cases doesn't matter, or cause any problems / disagreement, as the difference is usually small. However, as the project uses discrete thresholds without hysteresis or confidence intervals for detection, when the signals are around threshold, one or another application can show detected signals the other doesn't. In addition, there are 'best' signals not displayed in stderr output.

Technologically speaking both those applications are dated now, and substantial refinement work has gone into converging CPU & GPU applications leading up to V7 multibeam.

As floating point arithmetic is not dealing with exact representations, and Fourier analysis in particular has accuracy limitations that vary by implementation in the third party (FFT) libraries used, bit-level agreement is not a practical requirement for seti@home, also in part due to the data source being noise. Consider that as where the validation mechanism comes in, as Boinc doing 'its job'.

This all means for the purpose of seti@home firstly that however much application refinement is applied, there will always be *some* variation between applications especially cross-platform, and secondly that the nature of the data source imposes real restrictions on how much accuracy can be reasonably maintained (number of significant digits in the numeracy sense).

Jason
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Josef W. SegurProject donor
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4331
Credit: 1,113,296
RAC: 1,149
United States
Message 1313580 - Posted: 10 Dec 2012, 17:07:48 UTC - in response to Message 1313404.

http://setiathome.berkeley.edu/workunit.php?wuid=1109451566

My result:

Spike count: 0 Pulse count: 0 Triplet count: 2 Gaussian count: 0

Wingman's result:

Spike count: 0 Pulse count: 0 Triplet count: 2 Gaussian count: 0

What else feeds into the decision logic?

For that WU, the validator is also comparing one best_spike, one best_pulse, one best_triplet, and one best_gaussian from each host. The best_triplet has to be identical to one of the reported triplets, though the hosts might have disagreed on which of the two was best. Very tiny calculation differences can swing the choice, and also for the other three "best" signals.
Joe

Message boards : Number crunching : WU Inconclusive...Why?

Copyright © 2014 University of California