[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Bug-gnubg] Base of luck

From: Robert-Jan Veldhuizen
Subject: [Bug-gnubg] Base of luck
Date: Sat, 25 Jul 2009 16:19:39 +0200

---------- Forwarded message ----------
From: Robert-Jan Veldhuizen <address@hidden>
Date: Sat, Jul 25, 2009 at 4:19 PM
Subject: Re: [Bug-gnubg] Base of luck
To: Frank Berger <address@hidden>

I agree with luck being represented in MWC.

Also, I think the "luck per move" should be either removed, or at least, not be the basis of GnuBG's judgement about both players luck.

The total amount of luck (in MWC) should be used to tell how lucky you were. Luck per move makes no sense, as it will approach zero as the number of moves goes to inifinity. Ergo, in long matches GnuBG will nearly always try to tell us neither side was lucky, even when one player received f.i. +40% MWC luck.

Luck and errors (skill) should not be related too much IMO, they're mathematically different variables as errors are always negative and therefore the sum grows increasingly negative, whereas luck can be both positive and negative and the sum tends to zero in the long run.

If the overall statistics will be improved, I'd like to propose adding a Snowie-style "favourite" measure, with added ELO rating difference based on the matchlength. This is already done for the luck-adjusted result, but it would be nice to have the same measure for the error-based result (in MWC obviously). Especially in shorter matches, the luck-adjusted result can be quite  a way off it seems, and the error-based result, while biased, is usually more accurate.

This would also be interesting to compare both outcomes, and if there's a big discrepancy, it's interesting to do a "set analysis luckanalysis pl 1" and analyze the match again, with luck now measured at the 1-ply level instead of the default 0-ply. Perhaps it would be useful to make this command part of the GUI in the options/analysis section.

Thank you all for the ongoing development of GnuBG!

Greetings, Robert-Jan

On Wed, Jul 8, 2009 at 11:56 PM, Frank Berger <address@hidden> wrote:
Hi Christian,

hm... I see your point that makes skill and luck comparable. I agree with you that skill is EMG based, everything else makes no sense at all. Naturally an error at the beginning is less severe to the outcome, but to estimate the skill weighting with MWC.... no never.

For the luck I think it makes sense to use MWC, so my point of view is to take the red pill and have MWC for Luck and EMG for skill.


This argument pops up from time to time, and I'm off the unfortunate
opinion that both views are equally bad. The reason that we report
luck and skill in EMG is that it helps on the report and comparison of
games and matches.

Suppose you are at 17-17 away or 1-1 away. Loosing 5% matchequity by
bad luck would require a very unlucky game at the first score and
hardly any bad luck at the second score. Using EMG puts luck and skill
on a foundation that is independent of score and matchlength.
Unfortunately the foundation isn't very good.

Every argument about EMG or MWC you want to make about luck may be
made for skill as well. For skill our decision is firmly for EMG. For
luck the argument for MWC is stronger, but it isn't 100% convincing. I
like to have luck and skill reported the same way, and if somebody
took the time to make the code changes to report both EMG and MWC for
both luck and skill in a concise way I wouldn't mind at all.

Bug-gnubg mailing list

Robert-Jan Veldhuizen

Robert-Jan Veldhuizen

reply via email to

[Prev in Thread] Current Thread [Next in Thread]