|
From: | Ian Shaw |
Subject: | Re: [Bug-gnubg] Wrong luck adjusted result on Grandmaster analysis |
Date: | Tue, 6 Jan 2015 17:47:52 +0000 |
Richard,
I’m not sure I “get it”, either! I’m just using my best guess. 1.
The result in the game I ran was indeed a win for “tophat”. Yes, the difference is smaller than your examples, because both sides had good rolls. However, the maths
is consistent, so I can’t see where there is any error. I don’t think anything is mathematically wrong, but the way it’s presented ,may not be the most comprehensible, as outlines in my earlier post. Do you agree with this assessment? 2.
Of course, EMG and mEMG are both affected by cube value. I don’t use the database (too lazy), so I can’t comment on that aspect.
3.
I got 2.4 from 1.791 - -0.642 = 2.433. That is, the sum of your bad luck and your opponent’s good luck. The gnubg developers are fairly inactive at the moment, so I’m not sure who’d be willing and able to reply, but I’ve copied this post to the mailing list. I haven’t got the
source code loaded on any PC right now, so I can’t try and decipher it myself. n
Ian From: RICHARD BEAGLEY [mailto:address@hidden
Ian At the risk of losing your good will....
"Gnubg reports EMG Total luck as (-0.642 and +1.791). That’s a 2.4 game swing, which looks like a lot of luck for a single
game, consistent with the MWC report. Dividing by the number of moves (20, 19) gives the reported mEMG luck values of (-32.1, +94.3)." Is there somebody at gnubg support who could fill in your blanks? Thanks for your patience so far and apologies for not "getting it" Richard From: Ian Shaw <address@hidden> I would interpret this as his winning being 98% luck and only 2% due to relative chequer/cube play. (You did make SOME errors, after all.) Statisticians may well be critical
of this interpretation. However, I’ve just analysed a 1-pointer with gnubg playing both sides (so the result is entirely due to luck), and got the following result: tregurtha tophat
Luck total EMG (MWC) +0.742 (+37.093%) +1.392 (+69.579%) Luck rate mEMG (MWC) +17.7 ( +0.883%) +32.4 ( +1.618%) Luck rating None Good dice, man!
Luck adjusted result -17.51% +17.51%
I’m not entirely sure what to make of this. n
Ian From: RICHARD BEAGLEY [mailto:address@hidden]
Ian I have just noticed your comment: "In other words, without the luck, tophat would have got to 1.95% instead of +50%, and you would have got to -1.95% instead of
-50%. His victory was almost entirely by luck." and you could have continued ...........but he would still have won (+2% probability) How??? This is not the most important thing in my life but I would like to understand the logic. Regards Richard From: RICHARD BEAGLEY <address@hidden> Ian I monitor the mEMG luck differential - this case (-126.4) was the most adverse in my 383 (ignoring 1 point matches) results todate on DG (OK I'm new here). My luck adjusted
success rate todate is 66.5% cf 63.6% actual and I have had many results changed when luck adjusted. This was an apparent glaring anomaly. My chequerplay was Expert (v Beginner) My Cube was Supernatural (v Awful) Logic would say that I would probably have won if the luck distribution was less extreme. Surely there is a flaw in your formula which says otherwise! Regards Richard From: Ian Shaw <address@hidden> Hi Richard. The luck adjusted result looks OK to me. You had -12.342% luck and tophat had +35.705% luck. The actual result was you -50%, tophat +50%. The Luck-Adjusted result formula = Actual Result – Own Luck + Opponent’s Luck. Your luck-adjusted result = -50 - -12.342 + 35.705 = -1.953% Tophat’s luck-adjusted result = 50 – 35.705 + -12.342 = -1.953% In other words, without the luck, tophat would have got to 1.95% instead of +50%, and you would have got to -1.95% instead of -50%. His victory was almost entirely by luck. You aren’t the first person to find this confusing.
If the formula were instead (50 – Own Luck + Opponent’s Luck), we would have got 98.05% and 1.95% in your favour.
May be it’s the Actual Result that is the source of the confusion. If this were reported as +100 and 0, rather than +50 and -50, the current formula (Result – Own Luck
+ Opponent’s Luck), would give 48.05% and 51.95%. This indicates that, with the luck removed, your opponent only managed to get from 50% to 51.95%. Again, his victory was almost entirely luck. I think that (50 – Own Luck + Opponent’s Luck) gives the most intuitive result. Does anyone know how the other two active bots, XG and BgBlitz, report the luck adjusted result? It might be sensible to standardize on a definition and use that. I’ve
a feeling that XG uses (50 – Own Luck + Opponent’s Luck), but this is only based on my hazy recollection of the discussions at DailyGammon when Miran ran some luck-adjusted tournaments. n
Ian From:
bug-gnubg-bounces+address@hidden [mailto:bug-gnubg-bounces+address@hidden]
On Behalf Of RICHARD BEAGLEY Version GNU backgammon 0.90.0 Aug 8 2011 via XQuartz 2.7.7 (xoorg-server 1.15.2) on MAC OSX 10.6.8 Analysis - grandmaster 3 ply This a single game result (3 point) with extreme differences in move, cube and luck ratings. I should have had a luck adjusted win! Is this a known bug and / or is my software out of date? Regards Richard Beagley (tregurtha) |
[Prev in Thread] | Current Thread | [Next in Thread] |