How to measure luck in bg?
Richard Rasker
Hi all,
First of all: very best wishes for the new year!
Here's something I've been thinking about for quite
a while now: how to measure luck in bg.
I intended to write about this some time ago, but I
never got round to it. Recently the discussion about
the randomness of JF's dice has flared up again, which
made me decide to wait no longer and maybe make a fool
out of myself proposing an idea like this - it's just
that I keep wondering about it. What also seems funny
is that I never saw anyone bring it up before.
I know that the laws of statistics apply to bg - in the
long term that is. Yet it turns out that one (not necessarily
me) can have bad luck for weeks on end - I have seen
good players drop more than 150 rating-points over the
course of 3 or 4 weeks, whilst playing at least 3 matches
per day; in my opinion that counts as consistent bad luck.
For one thing: with a `personal luck-rating', one can no
longer whine about bad luck unless he or she really has
the numbers to show it :)
Here are some rather basic ideas as how to `measure luck':
- (Crude calculation) total number of pips rolled by a
player vs. the number of pips actually moved, so that
dancing and other situations where one can't move are
counted as bad luck.
- Number of shots vs. number of actual hits - I found
that when it's possible to hit a checker, most of the
time a player will hit. Not hitting anything with a given
number of shots implies bad luck.
- Number of possible ways to play a particular roll
(maybe I'm getting into skill-territory here), the
less ways a roll can be played, the less luck the
player has.
- (any other suggestion)
I'm quite aware that the level of skill will somehow
distort these calculations, so that unskilled players
probably seem to be less lucky than skilled players.
There must be lots of other snags in this whole concept,
and maybe the whole idea is just a load of rubbish, but
somehow it doesn't seem like complete nonsense.
I'd work it out further myself if I had the time to really
look into it - but I haven't even got time to play the actual
game lately. Besides, I don't think I have the necessary
insight in the matter, so I only hope that someone else
thinks it's at least an idea worth looking at.
Best regards,
Richard Rasker
(rasker on fibs)
-- remove .spam.kill from e-mail address when replying --
Donald Kahn
>Hi all,
>
>First of all: very best wishes for the new year!
>
>
>Here's something I've been thinking about for quite
>a while now: how to measure luck in bg.
>
(snip)
>
>I know that the laws of statistics apply to bg - in the
>long term that is. Yet it turns out that one (not necessarily
>me) can have bad luck for weeks on end - I have seen
>good players drop more than 150 rating-points over the
>course of 3 or 4 weeks, whilst playing at least 3 matches
>per day; in my opinion that counts as consistent bad luck.
>
>For one thing: with a `personal luck-rating', one can no
>longer whine about bad luck unless he or she really has
>the numbers to show it :)
>
>Here are some rather basic ideas as how to `measure luck':
>
>- (Crude calculation) total number of pips rolled by a
>player vs. the number of pips actually moved, so that
>dancing and other situations where one can't move are
>counted as bad luck.
>- Number of shots vs. number of actual hits - I found
>that when it's possible to hit a checker, most of the
>time a player will hit. Not hitting anything with a given
>number of shots implies bad luck.
>- Number of possible ways to play a particular roll
>(maybe I'm getting into skill-territory here), the
>less ways a roll can be played, the less luck the
>player has.
>
>- (any other suggestion)
Here is another way: (it would require an add-on to JellyFish or
similar system)
At every roll, evaluate the 36 possible equities produced by the next
roll, properly played. Note the player's actual roll, and keep a
running total of the cumulative difference between the equity produced
by his roll versus the median equity of the 36.
A comparison of the totals of the two players would give some
indication of their respective luck. Particularly over the course of
a long session, or a match, it seems to me it would be indicative.
deekay
Julian
In article <r.e.rasker....@student.spam.kill.utwente.nl>,
Richard Rasker <r.e.r...@student.spam.kill.utwente.nl> writes
>Here are some rather basic ideas as how to `measure luck':
>
>- (Crude calculation) total number of pips rolled by a
>player vs. the number of pips actually moved, so that
>dancing and other situations where one can't move are
>counted as bad luck.
>- Number of shots vs. number of actual hits - I found
>that when it's possible to hit a checker, most of the
>time a player will hit. Not hitting anything with a given
>number of shots implies bad luck.
>- Number of possible ways to play a particular roll
>(maybe I'm getting into skill-territory here), the
>less ways a roll can be played, the less luck the
>player has.
>
>- (any other suggestion)
>
I had a think about this. There's no real way to measure luck using
pips, hits, etc. since their value is dependent on context. However, the
only part of the game which involves luck is the rolling of the dice.
So, you can measure the "luck" of each throw as the increase in the
match/game equity achieved by rolling - that is, the difference between
your equity before rolling (a weighted average over all 21 possible
rolls) and after (when you know the roll). The "overall luck" for the
game would then be found by averaging the luck per roll.
You could equally define a "skill" associated with the move, as the
difference in equity between having made the roll (which assumes you
make the optimal move) and after completing the move - that is, the
equity lost by making a sub-optimal move.
---------------------------------------------------------------------------
Julian Hayward 'Booles' on FIBS jul...@ratbag.demon.co.uk
+44-1344-640656 http://www.ratbag.demon.co.uk/
---------------------------------------------------------------------------
All generalizations are false, including this one.
---------------------------------------------------------------------------
Wai Mun Yoon
Richard Rasker wrote:
<snip>
> For one thing: with a `personal luck-rating', one can no
> longer whine about bad luck unless he or she really has
> the numbers to show it :)
>
> Here are some rather basic ideas as how to `measure luck':
>
> - (Crude calculation) total number of pips rolled by a
> player vs. the number of pips actually moved, so that
> dancing and other situations where one can't move are
> counted as bad luck.
> - Number of shots vs. number of actual hits - I found
> that when it's possible to hit a checker, most of the
> time a player will hit. Not hitting anything with a given
> number of shots implies bad luck.
> - Number of possible ways to play a particular roll
> (maybe I'm getting into skill-territory here), the
> less ways a roll can be played, the less luck the
> player has.
>
> - (any other suggestion)
>
> I'm quite aware that the level of skill will somehow
> distort these calculations, so that unskilled players
> probably seem to be less lucky than skilled players.
> There must be lots of other snags in this whole concept,
> and maybe the whole idea is just a load of rubbish, but
> somehow it doesn't seem like complete nonsense.
>
> I'd work it out further myself if I had the time to really
> look into it - but I haven't even got time to play the actual
> game lately. Besides, I don't think I have the necessary
> insight in the matter, so I only hope that someone else
> thinks it's at least an idea worth looking at.
I find the concept of luck, whatever it is, and varying personal
attitudes to it totally facinating. However, I don't think that
ratiocination, in the manner you suggest will resolve very much, since
most of the discussion on this subject is qualitative.
- Read Danny Kleinmann's Vision Laughs at Counting where he discusses
the relative contribution of luck and skill in BG, and concludes that it
is impossible to determine the importance of each since BG is a game of
'luck-handling' skills.
- The volatility of the dice essentially expands the stochastic
search-space for a player. However, the *concept* of 'luck' is different
for humans than for 'bots, since 'bots are 'emotionally' indifferent to
luck and do not steam. This means that a human with good 'luck-handling'
skills is one who not only prepares for nightmare 5-5's, but who avoids
steaming after fanning four times on a two-point board, while for a 'bot
only the former situation applies. All experienced BG players recognise
this and exploit it. Remember Robertie's problem in 'Advanced
Backgammon' when he says that a take is a good practical option [though
technically incorrect] since the opponent will be severely shaken after
a loss? (implying that the short-term loss of equity in taking is more
than outweighed by dividends from future steaming). Thus, a quantitative
approach to rationalising 'luck', even if it were possible, would not
address the real problem, which is that different people *react*
differently to the same dice rolls [same amount of 'luck']
- Your mention of a 'personal luck rating' raises some interesting
questions. Are some players luckier than others? Are some versions of
JellyFish luckier than others? [judging by some of the recent posts, the
answer is yes :-) ] Joe Dwek in his intro to 'Backgammon for Profit'
admits that he believes that some players are fundamentally luckier than
others and so survive in schools stronger than their skills should merit
[note: he says he *believes* this, since this is unprovable], while
weaker, unlucky players give up the game quickly.
However, one should note that students of the game continue to study
(and rollout) only because they, in turn, *believe* in eventual
reversion to the mean or what is known as 'the long run'. Caveat: "Ars
longa, vita brevis." :-)
wm
(who would prefer to be lucky, than skillful)
Hank Youngerman
I'm one of those who believes that the law of large numbers will even
out the luck. My FIBS experience is over 4000, which means that I've
had probably over 100,000 dice rolls (mine and opponents). Let's say
it's 108,000, a nice mulitple of 36.
My calcultion is that over that time, it is 99.7% likely that the
number of doubles for the two players will be within about 365. In
other words, even if I'm extremely lucky - I will average one more
doubles per 11-point match. And if extremely unlucky, the reverse.
However, if one were to really want to measure luck, and do it in a
reasonable way (that is, not trying to count every change in equity) I
would measure:
1. Number of doubles. In general, doubles are such strong rolls that
they will swing a match powerfully.
2. Dancing percentages vs. 1, 2, 3, 4, and 5-point boards. Again -
hits are important, but hits are STILL sometimes two-edged swords.
While there are of course positions where you want to dance, it's
pretty rare.
But while we're on the subject...... I'll play anyone with an equal
rating to mine. You bring your lucky dice, your lucky talisman, a
beautiful girl to blow on your dice. I'll use your least lucky set of
dice. But I get to let Kit Woolsey make all my cube decisions. Any
takers?
(Yes, Kit, you can have half the action <grin>!)
Donald Kahn
>I'm one of those who believes that the law of large numbers will even
>out the luck. My FIBS experience is over 4000, which means that I've
>had probably over 100,000 dice rolls (mine and opponents). Let's say
>it's 108,000, a nice mulitple of 36.
>
>My calcultion is that over that time, it is 99.7% likely that the
>number of doubles for the two players will be within about 365. In
>other words, even if I'm extremely lucky - I will average one more
>doubles per 11-point match. And if extremely unlucky, the reverse.
>
>However, if one were to really want to measure luck, and do it in a
>reasonable way (that is, not trying to count every change in equity) I
>would measure:
>
>1. Number of doubles. In general, doubles are such strong rolls that
>they will swing a match powerfully.
>
>2. Dancing percentages vs. 1, 2, 3, 4, and 5-point boards. Again -
>hits are important, but hits are STILL sometimes two-edged swords.
>While there are of course positions where you want to dance, it's
>pretty rare.
>
>
These seem good indications. And more specifically and in my
experience most important, hitting shots off the bar. But I was not
suggesting that any human should keep track of the equity variations.
This is a job for a program, and would give specific information about
a player's luck during, for instance, a match or series of matches.
(If such a program should ever become available, one could feed into
it the matches of the participants in important events. We could give
a prize to the player who overcame the worst luck.)
deekay
Morten Wang
* Hank Youngerman
| 1. Number of doubles. In general, doubles are such strong rolls that
| they will swing a match powerfully.
| 2. Dancing percentages vs. 1, 2, 3, 4, and 5-point boards. Again -
| hits are important, but hits are STILL sometimes two-edged swords.
| While there are of course positions where you want to dance, it's
| pretty rare.
IIRC both of these where tested for in the dicethrows of all the
bigbrother matches and the results posted earlier this year. I
believe it was Tom Keith who posted the results, drop by Deja News or
his site and look for it. and IISRC Patti Beadles checked for the
number of doubles in the 5 million dicerolls she did, and the result
was that there weren't too many doubles. if my memory is incorrect
please correct me.
| But while we're on the subject...... I'll play anyone with an equal
| rating to mine. You bring your lucky dice, your lucky talisman, a
| beautiful girl to blow on your dice. I'll use your least lucky set of
| dice. But I get to let Kit Woolsey make all my cube decisions. Any
| takers?
what's your rating? (I'm so used to playing JellyFish that I believe
somebody with its cube-decisions but without its checkers play should
be beatable. does not neccessarily mean I'll back my view with money,
though)
Morten!
--
"God does not deduct from our alloted life span
the time spent playing backgammon."
--> Morty on FIBS
--> Backgammon homepage: http://home.sol.no/~warnckew/gammon/
Donald Kahn
On Sat, 03 Jan 1998 17:00:55 GMT, news.salzburg.co.at (Donald Kahn)
wrote:
>On Sat, 3 Jan 1998 15:07:01 UNDEFINED,
>r.e.r...@student.spam.kill.utwente.nl (Richard Rasker) wrote:
>
>>Hi all,
>>
>>First of all: very best wishes for the new year!
>>
>>
>>Here's something I've been thinking about for quite
>>a while now: how to measure luck in bg.
>>
>(snip)
>>
>>I know that the laws of statistics apply to bg - in the
>>long term that is. Yet it turns out that one (not necessarily
>>me) can have bad luck for weeks on end - I have seen
>>good players drop more than 150 rating-points over the
>>course of 3 or 4 weeks, whilst playing at least 3 matches
>>per day; in my opinion that counts as consistent bad luck.
>>
>>For one thing: with a `personal luck-rating', one can no
>>longer whine about bad luck unless he or she really has
>>the numbers to show it :)
>>
>>Here are some rather basic ideas as how to `measure luck':
>>
>>- (Crude calculation) total number of pips rolled by a
>>player vs. the number of pips actually moved, so that
>>dancing and other situations where one can't move are
>>counted as bad luck.
>>- Number of shots vs. number of actual hits - I found
>>that when it's possible to hit a checker, most of the
>>time a player will hit. Not hitting anything with a given
>>number of shots implies bad luck.
>>- Number of possible ways to play a particular roll
>>(maybe I'm getting into skill-territory here), the
>>less ways a roll can be played, the less luck the
>>player has.
>>
>>- (any other suggestion)
>
>Here is another way: (it would require an add-on to JellyFish or
>similar system)
>
>At every roll, evaluate the 36 possible equities produced by the next
>roll, properly played. Note the player's actual roll, and keep a
>running total of the cumulative difference between the equity produced
>by his roll versus the median equity of the 36.
>
>A comparison of the totals of the two players would give some
>indication of their respective luck. Particularly over the course of
>a long session, or a match, it seems to me it would be indicative.
>
>deekay
And, come to think of it, would it not be a way of putting to bed once
and for all the question whether JellyFish's luck is better than
average, over the long run? (See the theories of Professor van den
Doel, as posted, interminably, in the Jellyfish thread.)
Stuart Katz, MD
On Mon, 05 Jan 1998 15:01:46 GMT, news.salzburg.co.at (Donald Kahn)
wrote:
>(We could give
>a prize to the player who overcame the worst luck.)
>
My own experience is that "bad luck" is further complicated by a loss
of equity from trying to find the best of a bad set of candidate
plays. Maybe others have a different feeling but when I score my play
against JF L7 I consistently do better finding the best play on a
"lucky shake" but more frequently make errors on poor tosses.
Stuart
John Goodwin
I suspect that this is because we have a sort of 'best course' mapped
out as we play, (or possibly several in ranked order of preference),
and because we think most about these we are 'primed' when we get
compatible throws. On the other hand we probably spend less time
thinking about all the bad throws that could happen.
Perhaps that's one area where expert players score. It's certainly a
(fair) advantage that BG programs have (if they're written properly).
J.G.