Message boards :
Number crunching :
New Greenbank Files
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5
Author | Message |
---|---|
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Never cared for regulators, lol... Hmmm...rotating balls. Hmmm. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Never cared for regulators, lol... Yes, mathematically speaking, orbital oscillation in the temporal domain. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
To put some figures on the difference in runtimes: all tasks are VLAR, no -9 overflow, on I think my most stable machine. Arecibo (23 tasks): average 5,325.73 seconds, std. dev. 159.50 Guppi (4 tasks): average 4,441.34 seconds, std. dev. 241.30 Not as big a difference as I thought, but noticeable. |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
To put some figures on the difference in runtimes: all tasks are VLAR, no -9 overflow, on I think my most stable machine. Interesting to see what credit is granted by Credit Nu?. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
Interesting to see what credit is granted by Credit Nu?. You can watch if you like, but it's a long wait for wingmates. Valid tasks for computer 7118033 Arecibo - 2 VLAR validated, credit 105.98 and 108.96 Guppi - no validations at all yet. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I've had a few Validate, it's not much; http://setiathome.berkeley.edu/results.php?hostid=6906726&offset=500&show_names=1 If you run them on the GPU, it's even Less; http://setiathome.berkeley.edu/results.php?hostid=7258715&offset=520&show_names=1 |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
The problem is to get some for GPU. I disabled CPU work fetch for 24 hours and got nothing but Arecibo units. With each crime and every kindness we birth our future. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Of the first 18 that I completed, all are VLARs with an angle range between 0.005947 and 0.010613. 8 of the 18 have ended with -9 overflows. |
Chris Adamek Send message Joined: 15 May 99 Posts: 251 Credit: 434,772,072 RAC: 236 |
Same here. Nothing for my ATI cards from there yet. Chris |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
If such ratio will scale then some thresholds should be reconsidered IMO... |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
It doesn't seem to be scaling. I've now completed 87 guppis but the count of -9 overflows is only up to 18. I have no idea if that still seems high or not. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Most likely a reasonable indicator if the noise floor has changed, will be if you can get enough volume to see if the reissue (inconclusive) to pending ratio drops below 5% as with Arecibo tasks and a healthy system + app. Someproportion of that 5% (or less) going inconclusive, would be dodgy wingmen/apps (a sign of project health), natural variation cross platform, and a function of the noise-floor/threholds. If <=5% go inconclusive and reissue, yet the number of -9's remains high, then 'probably' can be interpreted as many apps+platforms+hosts agreeing that the noise floor is high (so thesholds should be raised) If, on the other hand, it refuses to converge on <=5% inconclusive, then some app or platform differences may remain (with or without -9 overflows present). Without excess -9's we could guess that there is some error growth going on in one or more calculations, one or more apps (possibly all). With excess -9's the cumulative error could be there also, but buried in the noise floor. I'm going to guess that this early there could be a bit of both noisy data on low thresholds going on, along with some possible app fine tuning needed if specific offenders can be pinpointed across enough hosts. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
If you follow the SoG thread to Beta, there is some indication that there may be a higher error rate than usual, on particular datasets. I think it's autocorr protection false positive. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Hard to say with GBT reobservation data IMO, as to what an appropriate safety limit might be, because self similarity implied by high autocorrelation values I'd imagine might be expected from deliberately staring at a radio source (self similar pattern, natural or otherwise) So yeah, probably a question for Eric, and lifting that might be the way (assuming it wasn't put there because of some driver/app/platform problem causing weird autocorrelations) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
Hard to say with GBT reobservation data IMO, as to what an appropriate safety limit might be, because self similarity implied by high autocorrelation values I'd imagine might be expected from deliberately staring at a radio source (self similar pattern, natural or otherwise) I think that was the intention, yes. But it raises - again - the problem of distinguishing high-significance autocorrelations caused by weird data, from ditto caused by weird computers. I suspect the answer may need to be "safeties to off" until Eric can characterise the data better for us. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
...(assuming it wasn't put there because of some driver/app/platform problem causing weird autocorrelations) I think it originally stemmed from quite a few rigs with ATI GPUs that were generating bogus -9 overflows with matching 30 autocorr counts that then validated against each other, causing legitimate results to be invalidated. (See Two wrongs make a right.) Joe suggested the sanity check and, once implemented, it seems to have alleviated that problem. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Thanks for link to post with original proposal. Since then level of 100 proved to be too low even for Arecibo's work. Current level of 135 isn't adequate for new GBT data too. I E-mailed Eric to check if we just have some weird batch of GBT data or it's inherent feature of such observation style... |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
A day later and I've now completed 143 guppi VLARs and the total of -9 overflows has only risen to 24, so the ratio seems to be dropping, now down to about 16.8%. For comparison, I took a look at the Arecibo VLARs in my archives for March. Out of 293 that I completed, 45 were -9 overflows, or about 15.4%. So, at least for my small sample sizes, the ratios are similar. However, what I did notice was that there seems to be a noticeably different distribution in the triggers for those overflows. To wit: -9 OVERFLOW TRIGGERS Arecibo VLARs (March, 2016) 30 spikes: 15 WUs 30 autocorrs: 1 WU 30 pulses: 0 WUs 30 triplets: 0 WUs 30 gaussian: 0 WUs 30 mixed signals: 29 WUs GBT (guppi) VLARs 30 spikes: 5 WUs 30 autocorrs: 3 WU 30 pulses: 2 WUs 30 triplets: 2 WUs 30 gaussian: 0 WUs 30 mixed signals: 12 WUs I don't know if that piques anybody's curiosity, but I figured as long as I had the data, I'd post it. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Maybe GBT has some different source of artifical signal noise than Arecibo. To have overflowed Triplets and Pulses w/o Spikes is very unusual indeed. Or, maybe we catch some really strong ET signal here.... fingers crossed :) |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Maybe GBT has some different source of artifical signal noise than Arecibo. To have overflowed Triplets and Pulses w/o Spikes is very unusual indeed. Perhaps some different thresholds might be worth considering, as I found that last night's processing yielded a whole bunch of triplets-only overflows. Out of 87 GBT tasks my hosts completed, 49 were 30-triplet overflows, all split from either HIP11048 (30 WUs) or HIP11048_OFF (19 WUs) files. Prior to those, I'd only seen 3 triplets-only overflows from GBT WUs, all from different sources, and only 3 others all month from Arecibo WUs. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.