Why is it that sometimes the total_work is lower than the statsrun before ? (i'm referring to the text dump http://www.seventeenorbust.com/stats/textStats.mhtml)
I'ts generating some negative values in my statsengine. (http://17orbust.qik.nl)
Printable View
Why is it that sometimes the total_work is lower than the statsrun before ? (i'm referring to the text dump http://www.seventeenorbust.com/stats/textStats.mhtml)
I'ts generating some negative values in my statsengine. (http://17orbust.qik.nl)
at one point Dave had a script cron'ed to go though and fix possible accumulated rounding errors. to my knowledge, it never actually ended up changing values. do you know when the drops are occuring?
-Louie
I use the 'update_time' (in rawstats) to gererate my import date (my timezone is EST +1, the Netherlands).
If the date is the same it update's the reccord, but if it's a new dat it inserts a new entry.
Sinds the difference is more than his daily output, it must be changed round 0:00 (the first statsrun witch has a new date)
Here is one user that has the problem : http://17orbust.qik.nl/index.php?tea...6-23&user=2300
After brooding over this with a friend a while ago we came to the following conclusion: The total work done by a person increases as "intemediate blocks" are submitted. If a person abandons a test somewhere in the middle, they've already received partial credit for it. Ten days later that same test is assigned to another person who crunches it and submits the final residual and at that time gets the full, formal credit for the test -- and at that time the original cruncher gets his/her preliminary partial credit removed which can lead to negative totals.
Now that sounds most excellent to me. If people/teams are just crunching partials to gain some stats points but then abandoning the proth test...then a removal of credit ten days later when someone else does ALL their work OVER again and finishes....well sounds great to me.
Give credit to those that actually accomplish something rather than a halfass effort.
:dump:
that scoring system does make some sense.
however, that is not what is used. Lagardo is mistaken.
-Louie
Louie - I may be (partially, at least) responsible for the notion that the scheme that Lagardo describes was what was being done. I came to this conclusion after watching what happened when ORU changed user IDs in the middle of a whole pile of tests. IIRC, this happened while you were taking finals and such, so that conjectures were all we had to go on...
Here's what happened: ORU's team production went through the roof - the "old" user still had partial credit for all the tests in progress and the "new" user got an astronomical increase in rate. People on other teams were getting understandably miffed at what appeared to be "cheating" (although no one ever said that this was intentional on ORU's part).
I watched the stats for a while and gradually the "old" user's production declined while the "new" user's rate declined as well. What I mean by this is *not* that the graph declined for the "new" user, but rather that the height of the spike (in the past) was gradually reduced.
The folks on the other teams saw this as well...they were still concerned, though - not so much with any transient effects of the changeover on the *rate* (which, after all, is always an estimate), but rather with the problem that it appeared that the ORU team had received double credit for part of this work - that is, once the "new" user reported a partial block, the server assumed that he had done all the block up until that point, whereas the "old" user had credit for the portion of the work he had actually done before the transfer. For example, let's say that the "old" user did 50% of a block, then transferred the block to the "new" user who first reported in at 55% completion. The team at that point has credit for 105% of the block in question.
Based on some rough estimates I had come up with by looking at the stats several different times over a period of a couple of days, I informed the folks on other teams who were concerned that my best guess was that by the time all the transferred Proth tests finished, all this would have sorted itself out. The only mechanism I could imagine for doing this was some variant of the scheme that Lagardo described above. Several days later, it did indeed look as though the total production of Team ORU was correct and so everyone was happy.
If this is *not* how the stats fixed themselves, how did they do so? Did they, in fact, correct themselves, or were we mistaken in our assessment that everything looked OK several days after that transfer? If there isn't some similar mechanism for doing this correction, does the current scheme allow "cheating" by transferring a block to several user ids within the same team? Again, I'm *not* alleging for a second that ORU did this intentionally, but were there unintended consequences of the change that they made?
Is it possible to move a work unit to a different userID after all?
You can, but you'll be flaged as a "suspicious" user for it. There's no reason you should be finishing other people's workunits.Quote:
Originally posted by Mystwalker
Is it possible to move a work unit to a different userID after all?
A few members of ORU are on the suspicion list. I'm going to have to look into it some more. So far, the vast majority of their work looks clean but there are a couple dozen shady looking block swaps between members that might not be cooincidences. We shall see.
-Louie
From talking to OberonBob, I think I can tell you what those block swaps were about - they had a bunch of machines that were ID'ed to OberonBob that should have been ID'ed to Macgyver. When it was noticed, those machines were swapped over to Macgyver - same machine finished the work unit that started it, just under another ID.
Well, I called it "our conclusion" because that's what it was -- not word from high up, just what we concluded. [shrugs]
Quite frankly I don't care about "cheating" or whatever, but if our conclusion was false I'd be mightily curious why the "total production" of a person is sometimes reported as negative (i.e. the number in the raw stats dump is lower than the one that was in the last dump). It took us a while to come up with any scenario that would do such a thing and now that we have real feedback from the project again, I'd really like to know...
We are definatley guilty of doing user switches mid-test. We never intended to cheat by doing that, and the stats seem to iron themselves out when the tests finish.
It is extremely easy for us to make one registry change to a bunch of computers at one time, and there have been some misunderstandings that have caused numerous computers to change user ids. I don't think it will happen in the future, as we are trying to be much more careful.
I could certainly understand if what we did would be concidered suspicious.
So, Louie, to be clear, you are saying that partial credit for a test lasts forever? It doesn't get cleared? If that is so, then ORU has a bit more total production to it's credit than it should.
It's still happening, users having less points than the statsrun before.
The only explanation (Lagardo's) is that someone else is 'finishing' the users test(s).
But....
....jjjL posted a quote from Louis that that's not the problem ?Quote:
Originally posted by jjjjL
however, that is not what is used. Lagardo is mistaken.
-Louie
I'm still wondering why this occasionally accures. 1 day ago : http://sob/index.php?team=34&more=us...ate=&user=1314
If you need any of my history etc.. please mail/post it.
Oops duke you made a booboo in your link ;)
This should be it
http://sob.qik.nl/index.php?team=34&...1-08&user=1314
Just some background clarification:
jjjjl is Louie. :p
Yes, tnx.. that's my 'local' link..Quote:
Originally posted by Sunnyboy
Oops duke you made a booboo in your link ;)
This should be it
http://sob.qik.nl/index.php?team=34&...1-08&user=1314
Already thought zo.. but was not sure ;)Quote:
Originally posted by Mystwalker
Just some background clarification:
jjjjl is Louie. :p
Any progress on this one ?
Sorry, but found another error in the stats.
12/13/03 11:22 CET:
textStats.mhtml,
Team 81 TeamPrimeRib 1019340632 525 27470 19 267 1.12838800724263e+15 2499643758886.18 1019244652 1071398412
12/14/03 07:22 CET:
textStats.mhtml,
Team 81 TeamPrimeRib 1019340632 525 27461 21 287 1.1270905088162e+15 2516299093365.22 1019244652 1071357614
1.1270905088162e+15 <- missing a 0 ?
1.12838800724263e+15
That's how an exponential value should be displayed Duke :(
But you can use a regular expression to find the e+ and then split and calculate the normal value can't you? If you need a hand gimmi a call