How much does luck and skill correlate?
lmfback
How much does luck and skill correlate?
After ending the longest "bad luck" BG streak I've had in my life (ok,
recorded streak) I think I have enough data to examine a hunch I've had
for some time. When two players are comparable in skill a -5 or -10 in
luck rate (Snowie terms) pretty much decides who wins. Sure enough, I
can't find one single match in my files where I've won at -10 or worse
in luck rate unless I've been at least two levels higher than my
opponent. Where the luck rate is around -2 I've won several matches
where I've outplayed my opponent.
Now let's assume I play a bad move (error, not blunder). For my next
roll my best play would have allowed 7 joker rolls. But with my worse
play, only 5 rolls qualify as jokers. Now, do bots take this into
account? Since 2 rolls now aren't great or maybe even horrible, what
difference will these two results show in the calculated luck rate?
Well, this should be easy to measure. Let's take a position like this
X to play (3 2)
+24-23-22-21-20-19-------18-17-16-15-14-13-+
| O O O O O O | | O |
| O O O O O O | | |
| | | | S
| | | | n
| | | | o
| |BAR| | w
| | | | i
| | | X | e
| | | X |
| O X X X X X | | X |
| O X X X X X | | X X |
+-1--2--3--4--5--6--------7--8--9-10-11-12-+
Pipcount X: 83 O: 102 X-O: 0-0/Money (1)
CubeValue: 1
Let's say X rolls 5-5 for all remaining rolls. O rolls 6-4 followed by
66 for all remaining rolls.
If X plays badly by playing 8/6 8/5 he will loose the game and his luck
rate is -142.
If X plays 11/6 correctly he will win the game (barely) with a luck
rate of 3.
Wait a second. Same rolls, one checker play error. Still the luck rate
is negative just because of one "tiny" misplay. Hmmm, so is my
"horrible luck" streak just me playing badly? Looking at the statistics
I haven't played much better opponents during my streak compared to
before. My rating hasn't increased noticeably either. So maybe I've
just have had bad luck then. But if I had played "out of my league",
would my luck rate be explainable by me playing so much worse than my
opponent?
I hope someone could give a reasonable explanation to this. Meanwhile
I'm setting up two players at home. One that will play against a bot
"beginner" and one that will play against bot "world-class". Maybe in
half a year or so I will have enough matches to look at. If the player
I use against the beginner bot has a (considerable) better luck
rating... well. Hmm.
Oh, and how long was my bad run? 13 months. I now have had one month
with positive luck. Another of those and I'm officially
out of the recession :-).
Eskimo
--
//------------------------------
//Remove absolutelynos-p-a-m to mail directly.
//Ascended:W,V (genopolywish),P(ill ath), T,K,H,S,B,C,P,W
(naked),Ro,Ra,A,W,almost pacifist A
//In progress:PAIN
Douglas Zare
lmfback wrote:
> How much does luck and skill correlate?
They certainly aren't mutually exclusive. The luck in backgammon means that
the game isn't decided by who makes the first mistake. The luck means that
it matters how much better than your opponent you play, not just whether
you play better.
The flip side of the coin is that you can determine how much better one
player is than another by determining how much luck is needed for each to
win. This works if you measure luck in mwc in an unbiased fashion, not in
EMG or EMG/move.
> After ending the longest "bad luck" BG streak I've had in my life (ok,
> recorded streak) I think I have enough data to examine a hunch I've had
> for some time. When two players are comparable in skill a -5 or -10 in
> luck rate (Snowie terms) pretty much decides who wins.
That might be reasonable for money play, if you play for a fixed amount of
time. In match play, you don't stop after a set period of time, but after
one side is ahead by 50% mwc.
If luck is properly measured, then between competent players, the luckier
would win the match over 99% of the time. The main reason you can find
examples that don't look like this is that Snowie's luck rate is a bad
measure.
See "A Measure of Luck" and "Hedging Toward Skill" by me in GammonVillage.
http://www.gammonvillage.com/backgammon/news/article_display.cfm?resourceid=411
http://www.gammonvillage.com/backgammon/news/article_display.cfm?resourceid=412
A luck rate in EMG mainly tells how the match was won, i.e., what match
scores were hit, as well as how easy the takes were for any doubles
accepted. Suppose in a 3 point match player A wins 1 point, then another
single game. Then player B wins the Crawford game, doubles player A in, and
wins at DMP. Who was luckier? Player B, obviously, unless A was playing
absolutely terribly. Player B won. Who had more luck in terms of EMG?
Perhaps player A, by about 0.200, and if the match took 200 moves the luck
rate for player A might be +1 millipoint per move. That's because with
perfect play, the luck in EMG for player A was +1 in the first game, about
+1.2 in the second game, -1 in the third game, and -1 in the fourth.
When expressed in mwc, the luck for player B should be about +50%. It is
50%+net skill. So, if player A made one little mistake and player B played
perfectly, the total luck for B was +49% mwc. If player A makes the same
mistake and then wins, the total luck for player A was +51% mwc. Even
though the luckier player wins, the weaker player needs more luck to win,
hence wins less often, since luck averages to 0.
Douglas Zare
Albert Silver
> How much does luck and skill correlate?
>
> After ending the longest "bad luck" BG streak I've had in my life (ok,
> recorded streak) I think I have enough data to examine a hunch I've had
> for some time. When two players are comparable in skill a -5 or -10 in
> luck rate (Snowie terms) pretty much decides who wins. Sure enough, I
> can't find one single match in my files where I've won at -10 or worse
> in luck rate unless I've been at least two levels higher than my
> opponent. Where the luck rate is around -2 I've won several matches
> where I've outplayed my opponent.
> Oh, and how long was my bad run? 13 months. I now have had one month
> with positive luck. Another of those and I'm officially
> out of the recession :-).
13 MONTHS of bad luck? I realize this is theoretically possible,
however the improbability is so significant I would look for another
explanation. Do you go over your moves with the bots seeing the moves
you got right and studying the moves you got wrong? The bots are
tremendous teachers and go a long long way in helping improve one's
game. There is also a large amount of instructive material in the
guise of books or online texts at sites such as GammonVillage
(http://www.gammonvillage.com) or Gammonline
(http://www.gammonline.com). If you get the Gold subscription
($50/year) at GammonVillage you can access all the archived classes
including a fantastic series by Walter Trice called Beginner's Boot
Camp. The title is misleading as it presents information ranging from
novice to advanced, but the value is unquestionable IMHO.
Albert Silver
>
>
> Eskimo
Reef Fish
> lmfback wrote:
>
> > How much does luck and skill correlate?
>
> They certainly aren't mutually exclusive. The luck in backgammon means that
> the game isn't decided by who makes the first mistake. The luck means that
> it matters how much better than your opponent you play, not just whether
> you play better.
That's an odd question to ask.
That still leaves the terms 'luck' and 'skill' not well-defined. Luck is
at least in part rolling the 'perfect roll' (at whatever the current
board situation), no matter what the probability of that roll is.
So, I'll answer a slightly different question, 'how much does luck
and winning a match correlate?' My answer to the question is:
It's asymptotally ZERO. Otherwise, the longer the match, the smaller
the role luck plays. The more skilled player will win most of the
time.
I'll have to think much harder to see if your question makes sense
from a technical, statistical point of view. I hold a Ph.D degree in
Statistics. :-)
-- Bob.
Gregg Cattanach
news:f9846eb9.03022...@posting.google.com...
> lmfback <peter.b...@absolutelynos-p-a-m.lmf.ericsson.se> wrote in
message news:<MPG.18c59493...@news.lmf.ericsson.se>...
> > How much does luck and skill correlate?
> >
>
> > with positive luck. Another of those and I'm officially
> > out of the recession :-).
>
> 13 MONTHS of bad luck? I realize this is theoretically possible,
> however the improbability is so significant I would look for another
> explanation. Do you go over your moves with the bots seeing the moves
> you got right and studying the moves you got wrong?
If he is using Snowie's account manager and getting an average luck factor
for all of the matches played each month, it isn't astronomically hard to
have this average be a negative number 13 months in a row. It is a 8192 to
one shot do to so, (2 ^ 13).
If he's saying he's been unluckier than his opponent in EVERY match he's
played for 13 months, that's another situation :)
Gregg C.
lmfback
gcattana...@prodigy.net says...
> "Albert Silver" <silver...@hotmail.com> wrote in message
> news:f9846eb9.03022...@posting.google.com...
> > lmfback <peter.b...@absolutelynos-p-a-m.lmf.ericsson.se> wrote in
> message news:<MPG.18c59493...@news.lmf.ericsson.se>...
> > > How much does luck and skill correlate?
> > >
> >
> > > Oh, and how long was my bad run? 13 months. I now have had one month
> > > with positive luck. Another of those and I'm officially
> > > out of the recession :-).
> >
> > 13 MONTHS of bad luck? I realize this is theoretically possible,
> > however the improbability is so significant I would look for another
> > explanation. Do you go over your moves with the bots seeing the moves
> > you got right and studying the moves you got wrong?
>
> If he is using Snowie's account manager and getting an average luck factor
> for all of the matches played each month, it isn't astronomically hard to
> have this average be a negative number 13 months in a row. It is a 8192 to
> one shot do to so, (2 ^ 13).
Well, my data is not all that statistically valid. In the beginning I
was only keeping a Good/Bad luck +1 count. Which means I could well
have had positive average luck during one month.
When I got Snowie 3 Student I went over a lot of previous matches for
twenty days. Ok, if like Zare says the Snowie luck rate isn't a good
measure... Oh well.
Did a similar run when I upgraded to Snowie 4 Student.
Add to that that I've been using other bots too like GnuBG and there's
probably enough errors measure-wise to make the data almost invalid
:-).
I still wonder if it's beacause I play like a schmuck though... but
I'll try to rephrase myself when responding to Zare.
lmfback
za...@math.columbia.edu says...
>
>
> lmfback wrote:
>
> The flip side of the coin is that you can determine how much better one
> player is than another by determining how much luck is needed for each to
> win. This works if you measure luck in mwc in an unbiased fashion, not in
> EMG or EMG/move.
>
>
>
> http://www.gammonvillage.com/backgammon/news/article_display.cfm?resourceid=411
>
> http://www.gammonvillage.com/backgammon/news/article_display.cfm?resourceid=412
I reread those. Apart from certain parts where you completely lost me
(especially in 412) I don't think they answer my basic question, which
pretty much is the position I showed. For the two different plays I
would like to see a luck rate figure that is equal.
> A luck rate in EMG mainly tells how the match was won, i.e., what match
> scores were hit, as well as how easy the takes were for any doubles
> accepted. Suppose in a 3 point match player A wins 1 point, then another
> single game. Then player B wins the Crawford game, doubles player A in, and
> wins at DMP. Who was luckier? Player B, obviously, unless A was playing
> absolutely terribly. Player B won. Who had more luck in terms of EMG?
> Perhaps player A, by about 0.200, and if the match took 200 moves the luck
> rate for player A might be +1 millipoint per move. That's because with
> perfect play, the luck in EMG for player A was +1 in the first game, about
> +1.2 in the second game, -1 in the third game, and -1 in the fourth.
>
> When expressed in mwc, the luck for player B should be about +50%. It is
> 50%+net skill. So, if player A made one little mistake and player B played
> perfectly, the total luck for B was +49% mwc. If player A makes the same
> mistake and then wins, the total luck for player A was +51% mwc. Even
> though the luckier player wins, the weaker player needs more luck to win,
> hence wins less often, since luck averages to 0.
Ok, this is more like it. What I really would like to see is a measure
for this for the example I gave. i.e., ***can I somewhere see that
because I blundered one single move, my luck rate shows up bad***.
Pointing me in the correct direction where to look in Gnu or Snowie
would be appreciated. I'd probably have to violate the license for
another 20 days on my secondary PC for that but if I'm ever to upgrade
to Snowie Pro I'd better figure out the benefits a bit better than I do
right now (sorry Oasya, but getting more than my initial euros is
probably why the trial exists).
As it is now I have a hard time figuring out if my play is improving or
not just because my luck rate has been "awful" for so long. It might
well be that my play has improved a lot, but because I was having a bad
run I don't see as much gain as I should have. This is a bit of a
motivation issue.
Thanks,
Douglas Zare
lmfback wrote:
> In article <3E5B9FDC...@math.columbia.edu>,
> za...@math.columbia.edu says...
> >
> >
> > lmfback wrote:
> >
> > http://www.gammonvillage.com/backgammon/news/article_display.cfm?resourceid=411
> >
> > http://www.gammonvillage.com/backgammon/news/article_display.cfm?resourceid=412
>
> I reread those. Apart from certain parts where you completely lost me
> (especially in 412) I don't think they answer my basic question, which
> pretty much is the position I showed. For the two different plays I
> would like to see a luck rate figure that is equal.
The luck for a roll is independent of how you play it. However, there is no reason to
expect the luck of a particular subsequent roll or sequence (which you specified) to
be the same, no matter how you play this one.
If you win the opening roll, 2-1, splitting is about the same as slotting. If I
respond 4-3, this is a lucky roll for me if you have slotted, since I get to hit. It
is an unlucky roll for me if you have split. Of course, to balance this, there are
rolls which are luckier if you have split than if you have slotted, such as 5-5.
I don't see any reasonable measure of luck that says, a priori, what the luck was of
the sequence 2-1, 4-3.
> >[...]
> > When expressed in mwc, the luck for player B should be about +50%. It is
> > 50%+net skill. So, if player A made one little mistake and player B played
> > perfectly, the total luck for B was +49% mwc. If player A makes the same
> > mistake and then wins, the total luck for player A was +51% mwc. Even
> > though the luckier player wins, the weaker player needs more luck to win,
> > hence wins less often, since luck averages to 0.
>
> Ok, this is more like it.
That's what I meant by this excerpt: "A strong player throwing away .2 less equity
than I do will still have to be lucky by +.8 to beat me (in a match). Of course, this
will happen 60% of the time."
Also relevant was, "Your expected luck, the average luck you should expect in the
future, is 0, regardless of your play. " The context of the word "play" was
different, but it is also true that if luck is properly measured, blundering now does
not affect your expected total luck for the rest of the game.
> What I really would like to see is a measure
> for this for the example I gave. i.e., ***can I somewhere see that
> because I blundered one single move, my luck rate shows up bad***.
I don't think that exists.
> [...]
> As it is now I have a hard time figuring out if my play is improving or
> not just because my luck rate has been "awful" for so long. It might
> well be that my play has improved a lot, but because I was having a bad
> run I don't see as much gain as I should have. This is a bit of a
> motivation issue.
Why focus on the results or your luck rate? Much more useful is to consider your
error rate or your net skill advantage. It makes more sense to ask whether you play
as well when your luck so far in a match has been bad as you do when your luck has
been good.
Douglas Zare
Michael Crane
Michael
--
** People who do not have adequate or up-to-date virus protection are
responsible for spreading viruses via emails etc.Install virus protection
software and help put an end to viruses. **
"Douglas Zare" <za...@math.columbia.edu> wrote in message
news:3E5DF50A...@math.columbia.edu...
lmfback
za...@math.columbia.edu says...
> > In article <3E5B9FDC...@math.columbia.edu>,
> > za...@math.columbia.edu says...
>
> If you win the opening roll, 2-1, splitting is about the same as slotting. If I
> respond 4-3, this is a lucky roll for me if you have slotted, since I get to hit. It
> is an unlucky roll for me if you have split. Of course, to balance this, there are
> rolls which are luckier if you have split than if you have slotted, such as 5-5.
>
> I don't see any reasonable measure of luck that says, a priori, what the luck was of
> the sequence 2-1, 4-3.
I kinda didn't expect it to exists yet. But it should be theorethically
possible for a bot to analyze moves so that if you do another move than
the best one the bot would calculate the luck based on the best move
and your move regarding your opponents next roll (or the top ten moves
but you get the picture). Then again, all this just might mesh into
what the bots currently do and I'm just not clever enough to understand
there is no or not much of a difference.
> > As it is now I have a hard time figuring out if my play is improving or
> > not just because my luck rate has been "awful" for so long. It might
> > well be that my play has improved a lot, but because I was having a bad
> > run I don't see as much gain as I should have. This is a bit of a
> > motivation issue.
>
> Why focus on the results or your luck rate? Much more useful is to consider your
> error rate or your net skill advantage. It makes more sense to ask whether you play
> as well when your luck so far in a match has been bad as you do when your luck has
> been good.
You are right. I'm trying to figure out ways on how to improve my game
further and this was one possible path. If I could "prove" that giving
my opponent one additional joker roll possibility per game would show
up as me being unlucky I could then concentrate on playing better in
general not giving that joker possibility thus improving my "luck" too
(read: try not to become an underdog). Likewise if I found out I really
have been unlucky I should try to improve my game when I'm an underdog.
So far I've been looking at the amount of errors/blunders before and
after certain events which is, erm, rather subjective and inaccurate.
[this is what you say above too I think but I haven't been as
articulate as you :-]
Gregg Cattanach
news:2831c30c.03030...@posting.google.com...
>
> "Error rate" according to what...? According to "Snowie"
> can we suppose...?? :))
>
> Unless one submits to the fallacy that "Snowie is the god
> given measuring stick of backgammon skill", the luck rate
> (even as measured by Snowie itself) would be much less of
> a biased comparison...!
>
> Luck rate is much easier to agree on than the skill rate.
>
When anyone refers to an error rate they always are talking about a bot
evaluation of every play in a match or a set of matches. This kind of
numerical analysis isn't possible without the aid of a computer.
Snowie is not a *god* given measuring stick of backgammon skill, of course.
However, Snowie *is* a measuring stick, and along with GnuBG, Jellyfish and
a few other bots they are the only measuring tools that we currently have
that can give reliable numerical results for error rates or luck rates.
Understand that the way a bot measure a luck rate is intimately related to
the way it comes up with an error rate. The luck rate depends on its
ability to give an equity for the best play with the 36 possible rolls, and
compare the average of those equities with the equity of the best play using
the actual roll. The same evaluation engine is uses to rate moves as errors
as well as deriving the luck factor for each roll.
Certainly the various bots are not perfect. But they clearly give us the
best information we have available today for evaluating backgammon positions
and matches, far better than just guessing or asking an 'expert'. Just
because they are imperfect and may disagree with one's opinion about how to
play the game doesn't mean they should be discounted and ignored.
The vast majority if not all world-class players depend on bot analysis to
improve their game. MK does not. You pick...
Gregg C.
Frank Mazza
I haven't responded to your latest missives because I am facing
deadline pressures at work and hence have less time to write both here
and at Gammonline.
If you want to persist in thinking that by agreeing with Zare I am
kissing his ass, go right ahead. In fact considering I agree with just
about everyone who argues positions contrary to yours (not a small
group), by your standards I'm a world class ass kisser.
I don't focus on my luck rate at all. I almost never look at the luck
rate after analyzing my matches with Snowie. I think most experienced
players have a pretty good feel for when they have been lucky/unlucky
in a match. I just care about my error rate and going over those
positions/cube decisions where my judgement was faulty.
I don't think Snowie is an infallible "God," but there aren't many
times when Snowie says I have made an error that I can't see upon
reflection that Snowie was right.
Snowie and Gnu at their strongest sail close to the top in ratings at
any server they play at. And as Gregg said, virtually all the world's
top players use Snowie and or Gnu/Jellyfish as analytical tools.
Your contention that Gnu cheats is as groundless as ever. I have
played long sessions with both Gnu and Snowie and have experienced, as
expected, lucky streaks and unlucky streaks. Nothing re. the dice
seems in any way different from what I experience playing over a real
board using manual dice, or playing against humans at the three
servers I play at.
Frank Mazza
On 2 Mar 2003 00:58:45 -0800, mu...@compuplus.net (Murat Kalinyaprak)
wrote:
>Douglas Zare wrote 3E5DF50A...@math.columbia.edu
>
>> Why focus on the results or your luck rate? Much more
>> useful is to consider your error rate or your net skill
>> advantage.
>
>"Error rate" according to what...? According to "Snowie"
>can we suppose...?? :))
>
>Unless one submits to the fallacy that "Snowie is the god
>given measuring stick of backgammon skill", the luck rate
>(even as measured by Snowie itself) would be much less of
>a biased comparison...!
>
>Obviously Zare can't comprehend this. Otherwise he would
>not have written the above paragraph in the first place.
>
>Luck rate is much easier to agree on than the skill rate.
>
>But what I really would like to know is what happened to
>our "triple major at Philosophy/Psychology/English at a
>quite good private college" Frank Mazza...???
>
>I posted a reply to his last article twice in this group.
>Perhaps he hasn't received them somehow...??
>
>Will he use his triple major and/or degree to still kiss
>Zare's ass in this typical "sick-gambler", "world-class",
>"phd-maathematician-level" bullshit ...?
>
>MK
Douglas Zare
Frank Mazza wrote:
> I'm still around Murat.
>
> I haven't responded to your latest missives because I am facing
> deadline pressures at work and hence have less time to write both here
> and at Gammonline.
I haven't responded to him since I haven't read most of what he has
written.
I've been using Google.com's archive to read some of the older articles on
rec.games.backgammon. It is remarkable that Murat Kalinyaprak didn't used
to be a repetitive netkook. He has contributed to several informative
discussions, though not usually the information.
His anal obsession is recent. In the last year (March 4 to March 4), it
appears he has authored 176 posts in rec.games.backgammon containing the
words "ass" or "shit," a solid majority of his 279 posts. I think the
record shows that in almost all of those threads, he was the first to use
the terms. In contrast, only 116 contained the words "checker," "cube," or
"double."
Historically, the majority of his posts were silly, but contained no
references to excrement or anuses. Before this past year, he brought them
up only 81 times out of 488 posts.
Anyway, I just thought it might be nice to add some numerical tools to the
analysis of how lucky we are to have Murat. I'm glad that other people can
stomach reading and responding to his posts, although I think a direct
discussion of backgammon is more interesting for me.
Douglas Zare
Frank Mazza
Maybe the pressure of ranting constantly and not being taken seriously
by anyone has exacerbated Murat's obssesive tendencies. Who knows?
But In the future I will only respond to his posts only if they are
totally free of gratuitious vulgarity and insults.
Enough, as they say, is enough.
Frank Mazza
Michael Sullivan
> When anyone refers to an error rate they always are talking about a bot
> evaluation of every play in a match or a set of matches. This kind of
> numerical analysis isn't possible without the aid of a computer.
> Snowie is not a *god* given measuring stick of backgammon skill, of course.
> However, Snowie *is* a measuring stick, and along with GnuBG, Jellyfish and
> a few other bots they are the only measuring tools that we currently have
> that can give reliable numerical results for error rates or luck rates.
> Understand that the way a bot measure a luck rate is intimately related to
> the way it comes up with an error rate. The luck rate depends on its
> ability to give an equity for the best play with the 36 possible rolls, and
> compare the average of those equities with the equity of the best play using
> the actual roll. The same evaluation engine is uses to rate moves as errors
> as well as deriving the luck factor for each roll.
> Certainly the various bots are not perfect. But they clearly give us the
> best information we have available today for evaluating backgammon positions
> and matches, far better than just guessing or asking an 'expert'. Just
> because they are imperfect and may disagree with one's opinion about how to
> play the game doesn't mean they should be discounted and ignored.
But here's a question... Have any world class expert players played a
long series of matches with the best bots, and how have they fared?
Are the best human players stronger than the bots? That's the
impression that I have so far. If this is true, then it seems odd to
talk about their "error rate" as measured by said bots. Clearly there
must be moves which the bot rates as errors, but which are actually
superior to the bot choice.
Michael
Gregg Cattanach
news:1frbi4m.3432lvlj69yN%m...@panix.com...
>> But here's a question... Have any world class expert players played a
> long series of matches with the best bots, and how have they fared?
>
> Are the best human players stronger than the bots? That's the
> impression that I have so far. If this is true, then it seems odd to
> talk about their "error rate" as measured by said bots. Clearly there
> must be moves which the bot rates as errors, but which are actually
> superior to the bot choice.
Nack Ballard and Mike Senkiewicz each played 300 money games against
Jellyfish for real $$ stakes (a millionaire backed Jellyfish.) This was
something 6-8 years ago, I'm not exactly sure of the date. One player won
about 59 points and the other lost 59 points. So at least in this trial the
bot was clearly playing at world-class level. The dice were rolled manually
and a person input the rolls into the computer and reported the play that
Jellyfish would make; the actual play was on a real backgammon board.
I also agree that the best players in the world probably do play better than
Snowie, et al, and some of their 'error rate' are actually plays that are
better than the bot's choice. But I'd guess that Snowie 3-ply would belong
somewhere on the list of the top 64 players in the world.
Gregg C.
Richard McIntosh
stronger than robots. By definition, they must make better moves at the
margin than the robots do.
The cases where it might be true are the sort of thing that experts talk
about with other experts. For example, it is widely known that earlier
robots played backgames and prime-vs-prime positions poorly (so do most
humans, by the way). Even intermediates are aware of this, and some
intermediates claim that they can "beat the robots" by steering into
high-cube backgames. The newer robots do not make so many mistakes in these
areas, since their authors have incorporated new neural nets that are
trained for those situations.
Having acknowledged that it there are cases where some humans make better
plays than robots, the most important questions are "Can we learn from
robots? Are robots the best tools available to guide us and to measure our
progress? How will we recognize moves that are better than robot moves?"
Unless you are willing (and can afford) to pay the price for thousands of
sessions with human experts, robots are the best tools available to get to
the level where you can answer the last question. Even human teachers will
guide you to the robots for practice, practice, practice.
When you are able consistently to achieve an error rate below 3.0 measured
by Snowie or GnuBG, you will be in elite company indeed, and it may well be
that your "true" error rate will be lower, since you will probably be better
than the robots. The number of human players for whom this is true can be
measured in two digits. For most of us, it is not an issue, since we play
at nowhere near that level. If your (robot-measured) error rate is 15, it
doesn't matter that you sometimes make plays that are better than the robot
plays, since you have no way of knowing which they are.
Meanwhile, the robots can quickly point out your blunders. As in all
interesting human endeavors, the leap from talented amateur to expert is
larger than amateurs realize, and by definition, amateurs cannot produce the
nuances of expert thought -- if they could, they would be experts. They may
marvel when they see expert plays, and claim to understand, but that is a
different thing. (Sometimes you can make a perfect drive off the tee; that
doesn't make you a professional golfer.) Consistency is key.
The robots will continue to get better, and humans who use them as part of
their training arsenals -- and think about why robots do what they do --
will get better faster than those who don't. All you have to do to
appreciate this point is to look at the top players at any major tournament.
With very few exceptions, they are dedicated robot students. (I cannot name
any exceptions, but I accept that there may be some.)
More importantly, the overall improvement in human play over the last ten
years can be ascribed in large part to the univeral use of robots, and the
need to play competitively against humans who use these tools. I am a
middling intermediate player, and I think it safe to say that I play as well
as the world champions did 30 years ago. Unfortunately for me, they are
mostly still around and, thanks to robots, they play so much better than
they did 30 years ago that I am still a middling intermediate player.
Enjoy the show.
Richard McIntosh
"Michael Sullivan" <m...@panix.com> wrote in message
news:1frbi4m.3432lvlj69yN%m...@panix.com...
> Gregg Cattanach <gcattana...@prodigy.net> wrote:
>
[...elision...]
Brad Davis
> I also agree that the best players in the world probably do play better than
> Snowie, et al, and some of their 'error rate' are actually plays that are
> better than the bot's choice. But I'd guess that Snowie 3-ply would belong
> somewhere on the list of the top 64 players in the world.
>
> Gregg C.
I disagree.
I would back Snowie 4 against ANY player in the world in matchplay for good money.
Brad Davis
> Your theoretical argument is correct: There are (a few) humans who are
> stronger than robots. By definition, they must make better moves at the
> margin than the robots do.
Please list the names of the humans you think can outplay Snowie 4 in matchplay?
I'd be prepared to put up my own good money to back SW4.
Michael Sullivan
> Meanwhile, the robots can quickly point out your blunders.
Clearly, I am not attempting to argue that one shouldn't use bots to
learn. For measuring me against some other intermediate player, they
should do just fine.
The question comes when trying to use them as a general measuring stick
for players that are roughly as strong as they are. I don't think that
makes a whole lot of sense.
I'm sure it's still worthwhile for top players to study the bots and
what they think are mistakes, because if they don't have a good reason
to think the bot is wrong, it probably isn't, and one can always check
with a rollout. Obviously rollouts improve but aren't the final answer
either.
It seems that 5-10 years ago, bots were weak enough at certain types of
games that their evaluations of any position likely to lead to those
games were very suspect, and rollouts don't help unless 5-6ply is deep
enough for most of the tree threads to resolve into positions the bot
can evaluate well.
There was a thread here about a month ago where Doug Zare pointed out a
weakness in the evaluation of one of the bots, where it used cubeless
money evaluation at one point in the minimax tree, and then applied that
evaluation to a problem with a cube in match play -- basically it
recommended a move that only made sense if the opponent responded (given
appropriate dice) by playing moves that would make sense at cubeless
money play, but would be hopeless blunders at the existing match score.
That's clearly a bug. It's the kind of bug that doesn't matter much in
most positions, but occasionally produces a boner move (like the
example) that smells bad even to a player like me. Presumably this kind
of analysis allso leads to very small errors that experts can detect
(but I cannot) far more often.
Michael
Richard McIntosh
1. I don't remember specifying match play. I was thinking of money play.
2. Naming the players I believe qualify is pointless; it will not cause
them to take your action.
If you specify a sufficient amount of your good money, perhaps one of them
will accept your offer. It's not for me to say.
Meanwhile, I stick by my contention. 8^)
Richard
"Brad Davis" <backg...@email.com> wrote in message
news:bcf2ceef.03030...@posting.google.com...
Frank Mazza
bots are probably stronger than any player in the world. Even the best
human players will make blunders through a lapse in attention or
fatigue that I would think would cost a not insignificant amount of
equity over time.
The bots can calculate the equity of positions/cube decisions that
even the best humans have to approximate.
If Snowie 4, say, was entered in human backgammon tournaments the way
chess programs like Fritz and Deep Junior are sometimes allowed to
play in Grandmaster chess tournaments, I would predict an impressive
showing for Snowie.
Frank Mazza
Brad Davis
> B Davis wrote bcf2ceef.03030...@posting.google.com
>
> My question is whether you will bet money on gnudung (at
> equal stregth settings as Snowie) against any human...??
>
> If so, why don't you try to get a world-class sick gambler
> to vouch for gnudung and back you up with your bet...? And
> I'm not talking about 10 matches of 7 points here. Rather
> 100 matches of 64 points...! If I win, I get to spit on the
> face of your world-class partner and call him a conspiring
> sick bastard in front of an audience. If you win, you can
> split the $1,000 that I will be betting... Easy money... :)
> In fact, "free money" if you guys really believe what comes
> out if your own mouths... Come and get it... :)))
>
> MK
Yes I'm prepared to back the latest version of GNU against you for $1000
We need to clarify what constitutes equal strength settings as Snowie.
You can spit in MY face if you win.
Name the place I'll be there.
Brad Davis
>
> > Yes I'm prepared to back the latest version of GNU against
> > you for $1000 We need to clarify what constitutes equal
> > strength settings as Snowie. You can spit in MY face if you
> > win. Name the place I'll be there.
>
> are you...??
>
> Are you among the top 10 players in the world...? If not, why
> would I bet to spit on your dumb face...?? Fuck off and make
> room for the real world-class conspiring sick bastards...!!
>
> MK
So you don't want to back my dollars against yours to prove that
GNU cheats?
Who do _you_ consider to be the top 10 players in the world?
Nobody
> So you don't want to back my dollars against yours to prove that
> GNU cheats?
> Who do _you_ consider to be the top 10 players in the world?
Brad,
You must realize that Murat is a anal obsessive idiot who's only
pleasure in life is taunting people in this newgroup. Trying to engage
him in any sort of meaningful discussion is a waste of time, and only
encourages him by feeding his pathological ego.
Osman Guner
"Frank Mazza" <fxmaz...@yahoo.com> wrote in message
news:tjmh6v0oiug46lutg...@4ax.com...
> Most everything you say makes sense to me, except I think the latest
> bots are probably stronger than any player in the world. Even the best
> human players will make blunders through a lapse in attention or
> fatigue that I would think would cost a not insignificant amount of
> equity over time.
>
> The bots can calculate the equity of positions/cube decisions that
> even the best humans have to approximate.
Hello Frank... its been a while I visited this list... colorful discussions
as always.
One thing the bots can't do is to "take chance". Someone who is able to
exploit this "weakness" may beat the bots, but may not necessarily be a
"world-class" backgammon player. There were such specialist people who
would consistently outperform the early versions of the chess computers.
Douglas Zare
Osman Guner wrote:
> One thing the bots can't do is to "take chance". Someone who is able to
> exploit this "weakness" may beat the bots, but may not necessarily be a
> "world-class" backgammon player. There were such specialist people who
> would consistently outperform the early versions of the chess computers.
This seems backward. A common human mistake is not to risk a "sure" win to try
for a gammon. Bots are better at handling volatile positions, and at judging
whether an improvement means 5% extra gammons or 2%.
Do you mean something other than "to take a chance" when you say "take chance?"
Douglas Zare
Frank Berger
> One thing the bots can't do is to "take chance". Someone who is able to
> exploit this "weakness" may beat the bots, but may not necessarily be a
> "world-class" backgammon player. There were such specialist people who
> would consistently outperform the early versions of the chess computers.
What do you mean with take a chance? I personally feel, that the bots
are very coldminded and don't have any fear.
To exploit the weaknesses of a chess programm hav been quite easy (in
earlier days), becuase the strategic judgement of the humans was
better. This is a different story in Backgammon. I think only in deep
backgames is a chance to fool the bots, and I'm don't even sure on
that.
ciao
frank
Albert Silver
He's referring to the mighty David Bronstein, a tactical wizard and
world champion contender in the 1950s, who would play some incredibly
creative games against the early chess computers by sacrificing
enormous amounts of material and beating them in spectacular style.
The thing is that even then the issue wasn't so much that he took a
risk as the computers he was playing were so much weaker than he was.
He was truly no longer in his heydey but all the same represented a
solid 300 Elo point average *minimum* against the silicon opponents.
It was fun to watch but I was never convinced the reason he won had
anything to do with the style of play. Another very significant issue
was the fact that chess programs, stand-alones then, suffered from a
sever lack of understanding of king-safety. Until quite recently this
was still a bad issue so you can imagine how it was when the programs
calculated half as deep and had less knowledge to boot.
Albert Silver
Douglas Zare
Albert Silver wrote:
> Douglas Zare <za...@math.columbia.edu> wrote in message news:<3E74032C...@math.columbia.edu>...
> > Osman Guner wrote:
> >
> > > One thing the bots can't do is to "take chance". Someone who is able to
> > > exploit this "weakness" may beat the bots, but may not necessarily be a
> > > "world-class" backgammon player. There were such specialist people who
> > > would consistently outperform the early versions of the chess computers.
> >
> > This seems backward. A common human mistake is not to risk a "sure" win to try
> > for a gammon. Bots are better at handling volatile positions, and at judging
> > whether an improvement means 5% extra gammons or 2%.
> >
> > Do you mean something other than "to take a chance" when you say "take chance?"
> >
> > Douglas Zare
>
> He's referring to the mighty David Bronstein, a tactical wizard and
> world champion contender in the 1950s, who would play some incredibly
> creative games against the early chess computers by sacrificing
> enormous amounts of material and beating them in spectacular style.
I think grandmasters could spot chess programs a queen until the late 1970s, so who cares how they chose
to beat them before then?
I snipped some context. Osman Guner was responding to someone who didn't mention chess (IIRC), in a
backgammon newsgroup. He mentioned backgammon afterwards. So, if his comment was not supposed to apply
to backgammon, but rather to chess, it was absurd. I think it was supposed to apply to backgammon, but I
don't know how.
I think there is a tendency for people to make up weaknesses of bots based on too little evidence, or to
imagine that anything that produces numbers and says you are wrong must be perfect. It takes a lot of
work to discover the real weaknesses of bots, and of human players, for that matter.
Unless you try to have 15 checkers sent back, it won't happen, and you will never encounter the problems
with misevaluating an outside prime being rolled home over a single checker. Bots have had more
weaknesses in backgames than this, but it is much more important that Snowie 3 often misevaluated
holding games by 0.1 cubeless (Snowie 4 is a bit better, but still surprisingly weak there), and that
Snowie 4 undervalues the blitz in the opening, and plays strangely against a potential backgame.
Jellyfish is much too aggressive at the start of a contact bearoff, and misevaluates containment
positions and some primes terribly, and it doesn't handle the cube as well as Snowie.
I haven't worked with BGBlitz or gnu to determine their weaknesses, but I'd bet that they exist. I
wouldn't guess their weaknesses ahead of time.
Douglas Zare
Douglas Zare
Douglas Zare wrote:
> I snipped some context. Osman Guner was responding to someone who didn't mention chess (IIRC), in a
> backgammon newsgroup. He mentioned backgammon afterwards. So, if his comment was not supposed to apply
> to backgammon, but rather to chess, it was absurd. I think it was supposed to apply to backgammon, but I
> don't know how.
Sorry for the bad line-breaks. I'm using Netscape Communicator 4.78. I have my preferences set to wrap at 65
characters, yet the line breaks are often inserted only after 90 or more characters, and it varies. I'll try
to switch to a different newsreader. If anyone knows how to fix this bug, please send an e-mail to me
directly. A web search produced nothing.
Douglas Zare
az-willie
========
Why are you using such an out of date program? Netscape 7.0 is out and
also you could be using Mozilla 1.3.
Douglas Zare
> Douglas Zare wrote:
>
>> Sorry for the bad line-breaks. I'm using Netscape Communicator 4.78.
>> I have my preferences set to wrap at 65
>> characters, yet the line breaks are often inserted only after 90 or
>> more characters, and it varies. I'll try
>> to switch to a different newsreader. If anyone knows how to fix this
>> bug, please send an e-mail to me
>> directly. A web search produced nothing.
>>
> Why are you using such an out of date program? Netscape 7.0 is out and
> also you could be using Mozilla 1.3.
Older does not mean obsolete. It was not the newest program when I last
installed it, but it had a better set of features than many newer
programs. Communicator 4.7* is particularly good at displaying threads
and avoiding large amounts of dead space. The bug I mentioned seems to
be active some of the time and not at other times.
Douglas Zare
Albert Silver
> Albert Silver wrote:
>
> > Douglas Zare <za...@math.columbia.edu> wrote in message news:<3E74032C...@math.columbia.edu>...
> > > Osman Guner wrote:
> > >
> > > > One thing the bots can't do is to "take chance". Someone who is able to
> > > > exploit this "weakness" may beat the bots, but may not necessarily be a
> > > > "world-class" backgammon player. There were such specialist people who
> > > > would consistently outperform the early versions of the chess computers.
> > >
> > > This seems backward. A common human mistake is not to risk a "sure" win to try
> > > for a gammon. Bots are better at handling volatile positions, and at judging
> > > whether an improvement means 5% extra gammons or 2%.
> > >
> > > Do you mean something other than "to take a chance" when you say "take chance?"
> > >
> > > Douglas Zare
> >
> > He's referring to the mighty David Bronstein, a tactical wizard and
> > world champion contender in the 1950s, who would play some incredibly
> > creative games against the early chess computers by sacrificing
> > enormous amounts of material and beating them in spectacular style.
>
> I think grandmasters could spot chess programs a queen until the late 1970s, so who cares how they chose
> to beat them before then?
>
> I snipped some context. Osman Guner was responding to someone who didn't mention chess (IIRC), in a
> backgammon newsgroup. He mentioned backgammon afterwards. So, if his comment was not supposed to apply
> to backgammon, but rather to chess, it was absurd. I think it was supposed to apply to backgammon, but I
> don't know how.
His reference to players outperforming chess computers by taking risks
was IMO a clear reference to Bronstein among others. Bronstein
trounced the top programs of the day until 1991-1992 or so in such a
manner. How that would be applied to backgammon programs is not clear
to me, but I was merely explaining his reference.
Albert Silver
Osman Guner
the cube is at 4. When I see a few good rolls (or jikers) that will allow
me to achieve an easy win, I turn the cube at a point, perhaps, Snowie would
consider a blunder.
I am taking a chance here: if I do get the good rolls, I win; if I don't get
the good rolls, I will have to play play Crawford next and will have to beat
the bot 5 times in a row in order to win.
So I turn the cube... I don't think the bots would do that.
"Frank Berger" <fr...@bgblitz.com> wrote in message
news:aab8a7e5.0303...@posting.google.com...
A Modern Caveman
>the cube is at 4. When I see a few good rolls (or jikers) that will allow
>me to achieve an easy win, I turn the cube at a point, perhaps, Snowie would
>consider a blunder.
>
>I am taking a chance here: if I do get the good rolls, I win; if I don't get
>the good rolls, I will have to play play Crawford next and will have to beat
>the bot 5 times in a row in order to win.
>
>So I turn the cube... I don't think the bots would do that.
The computers are pretty lousy cube managers.
I can beat JellyFish on Level 7 in a "500 point match" a good percentage of the
time just by getting a slightly inferior position and then doubling and
redoubling. A good "bar bet" is to take 20-1 odds on a "long-term" 500-point
match that you can then reduce to one game with a 15-20 percent chance of
winning.
Ray Gordon: BACK by popular demand!
Limited time only!
Everything you need to know about women. FREE!
http://www.cybersheet.com/library.html
Douglas Zare
Osman Guner wrote:
> Here is an example. I am behind 6-3 against a strong computer, 11 pt game,
> the cube is at 4. When I see a few good rolls (or jikers) that will allow
> me to achieve an easy win, I turn the cube at a point, perhaps, Snowie would
> consider a blunder.
Snowie understands desperation doubles far better than most human players. I
doubt very much that Snowie would consider your redouble to be a blunder if you
had market losers. Do you have a concrete example?
It used to be a big weakness of gnu that it doubled far too aggressively when
behind in the match, but I think the developers have mostly fixed that problem.
> I am taking a chance here: if I do get the good rolls, I win; if I don't get
> the good rolls, I will have to play play Crawford next and will have to beat
> the bot 5 times in a row in order to win.
>
> So I turn the cube... I don't think the bots would do that.
Bots try to play well rather than guess when they will get lucky. So do good
human players. Are you trying to say how human players might outplay bots, or
just manners in which human players play badly and still win some amount less
than 50% of the time?
Douglas Zare
Nis Jorgensen
wrote:
>The computers are pretty lousy cube managers.
>
>I can beat JellyFish on Level 7 in a "500 point match" a good percentage of the
>time just by getting a slightly inferior position and then doubling and
>redoubling. A good "bar bet" is to take 20-1 odds on a "long-term" 500-point
>match that you can then reduce to one game with a 15-20 percent chance of
>winning.
Perhaps it is the people you are betting against, rather than
Jellyfish, who are lacking in understanding.
BTW: I wouldn't put 20-1 odds on a reasonably good player against JF
(or gnu), even if the cube level was restricted.
BTW: You can set up gnu to use a MET which assumes a difference in
skills - which would probably limit your wins significantly.
--
Nis Jorgensen
Amsterdam
Please include only relevant quotes, and reply below the quoted text. Thanks
jthyssen
> BTW: You can set up gnu to use a MET which assumes a difference in
> skills - which would probably limit your wins significantly.
Unfortunately gnubg can only play 64-pt matches, because rounding
errors in the MET start to show up around these scores, as gnubg only
uses single precision floats.
Jørn
A.Melon
http://www.ray-gordon.com/ (now over 25,000 hits)
In an effort to inform the many people on Usenet who have been offended,
provoked, harassed, or otherwise puzzled during the past eight years (yes, EIGHT
YEARS) by the individual calling himself "Ray Gordon", the above website was
created to detail some of his malicious behaviors, opinions, mental illness,
threats to sue people on an almost daily basis, and his tendency to proclaim
himself an expert on pretty much everything from gymnastics to the stock market.
All this despite a lack of any credentials and the fact that he still lives at
home with his mother at the age of 35. The site is NOT affiliated with one
Gordon Parker of Pennsylvania, who has made numerous threats of legal action
against it.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
For information on Paranoid Personality Disorder please visit
http://www.toad.net/~arcturus/dd/paranoid.htm and
http://60.1911encyclopedia.org/P/PA/PARANOIA.htm
(scroll down to the section on litigious paranoia)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
In article <20030329182640...@mb-fm.aol.com>
amodern...@aol.com (A Modern Caveman) wrote:
>