Some positions...
William R. Tallmadge
...from a match played by two of the world's top players.
Some JellyFish3.0 equities.
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 0
| | | X X X X X X |
| | | X X X X X |
| | | X X |
| | | |
| | | |
v| |BAR| | 11-point match
| | | |
| | | |
| | | |
| | | O |
| X | O | O X O |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 0
BAR: O-1 X-0 OFF: O-11 X-0 Cube: 2 (owned by X) turn: X
X rolls 3 and 2.
X moves 12-15 15-17 .
Move played -.126 JF's top move 12-14 21-24 -.007
========================================
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 0
| O X X | | X X X O |
| O X X | | X X X |
| | | |
| | | |
| | | |
v| |BAR| | 11-point match
| | | |
| | | |
| X | | O |
| X O O O | | O O |
| X O O O | | O O O X X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 2
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: O
O doubles.
X accepts the double.
No Double / Take .370 v.207
O rolls 2 and 4.
O moves 6-4 10-6 .
.325 6-4 24-20 .488
=========================
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 1
| X X O | | O O X |
| O | | O X |
| O | | O X |
| O | | O |
| O | | |
| |BAR| |v 11-point match
| X | | O |
| X | | O |
| X | | X O |
| X | | X O |
| X X | | X X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 4
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 1 and 4.
X moves 19-20 3-7 .
-.378 19-20 17-21 -.318
###
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 1
| O O O O O | | X O |
| O O O O | | X O |
| O O | | |
| | | |
| | | |
| |BAR| |v 11-point match
| | | |
| | | |
| X | | |
| X X X X X | | O |
| X X X X X X | | X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 4
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 2 and 3.
X moves 19-21 19-22 .
-.267 19-21 18-21 -.199
###
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 2
| X O O | | O X |
| O O | | O X |
| O | | O X |
| | | X |
| | | |
| |BAR| |v 11-point match
| X | | |
| X | | X O |
| X | | X O |
| O X | | X O |
| O O X | X | X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 4
BAR: O-0 X-1 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 2 and 2.
X moves bar-2 19-21 19-21 12-14 .
-.009 bar-4 21(2) +.061
###
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 2
| O X O O O | | O X |
| O X O O | | O X |
| O | | O X |
| | | |
| | | |
| |BAR| |v 11-point match
| | | |
| | | |
| X | | |
| X X X | | O X O |
| X X X X | | O X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 4
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 2 (owned by X) turn: X
X rolls 5 and 4.
X moves 12-17 17-21 .
-.196 17-22 17-21 -.119
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 2
| O X O O O | | O X |
| O X O O | | O X |
| O | | O |
| | | |
| | | |
| |BAR| |v 11-point match
| | | |
| X | | |
| X | | |
| X X X | | O X O |
| X X X X | | O X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 4
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 2 (owned by X) turn: O
O rolls 4 and 3.
O moves 8-4 5-2 .
.192 8-4 13-10 .243
###
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 2
| O O X O O O | O | X |
| O O O O O | | X |
| O O | | |
| | | |
| | | |
| |BAR| |v 11-point match
| | | |
| | | |
| X | | |
| X X X X | | O |
| X X X X X X | | O X |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 4
BAR: O-1 X-0 OFF: O-0 X-0 Cube: 2 (owned by X) turn: X
X doubles.
rjohnson whispers: opinions?
Chase whispers: drop
John_B whispers: take
petterb whispers: i'd drop
tommyh whispers: take
MontBlanc whispers: take
sunny whispers: drop
rjohnson whispers: drop, only due to score
rjohnson whispers: sorry, no, wait. score is the other way around. Not sure
O gives up. X wins 2 points.
No Double / Take
Wins Gammons/BG's
30.0 10.0 xx.x
70.0 25.0 xx.x
###
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 3
| O O O O | | O X |
| O O O O | | O X |
| O | | O |
| | | |
| | | |
| |BAR| |v 11-point match
| | | |
| | | |
| X X | | |
| X X | | X X X O |
| O X X | | X X X X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 6
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X doubles.
O accepts the double.
No Double / Take .324 v.315
X rolls 1 and 1.
X moves 14-15 15-16 15-16 14-15 .
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 3
| O O O O | | O X |
| O O O O | | O X |
| O | | O |
| | | |
| | | |
| |BAR| |v 11-point match
| | | |
| | | |
| X X | | |
| X X | | X X X O |
| O X X | | X X X X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 6
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 2 (owned by O) turn: O
O rolls 2 and 6.
Pipcounts: X 117 O 113
O moves 8-2 8-6 .
-.579 13-7 13-11 -.547 ?!
###
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 3
| O O O O O | | O O |
| O O O O | | O |
| O O | | |
| | | |
| | | |
| |BAR| |v 11-point match
| | | |
| | | |
| X X | | X |
| X X | | X X X X |
| O X X | | X X X X |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 6
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 2 (owned by O) turn: X
X rolls 3 and 5.
X moves 19-22 17-22 .
.437 23(2) .543 ?? (doubt it very much - JF screwy in Prime vs P)
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O X | | X X O |
| O X | | X X O |
| O X | | X |
| O | | X |
| O | | X |
v| |BAR| | 11-point match
| | | |
| | | O |
| X | | O |
| X O | | O O X |
| X O | | O O X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 8
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: O
O rolls 6 and 3.
O moves 13-7 13-10 .
-.061 13-4 -.035
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O X | | X X O O |
| O X | | X X O |
| O X | | X |
| | | X |
| | | X |
v| |BAR| | 11-point match
| | | |
| | | O |
| X | | O |
| X O | | O O |
| X X O | O | O O X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 8
BAR: O-1 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X doubles.
O accepts the double.
No Double / Take .430 v.183
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: score: 3
| O O X X | | O X X |
| X X | | O X X |
| X X | | X X |
| | | X |
| | | |
v| |BAR| | 11-point match
| | | |
| | | |
| | | |
| O | | O O O O X |
| O O | | O O O O X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 8
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 2 (owned by O) turn: X
X rolls 1 and 2.
X moves 1-2 22-24 .
-.879 2(2) -.810
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O O | | X X X X |
| | | X X X X |
| | | X X X |
| | | X X |
| | | |
v| |BAR| | 11-point match
| | | |
| | | O |
| | | O O |
| O | | O O O O |
| O X | | O O O O X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 8
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 2 (owned by O) turn: O
O rolls 3 and 3.
O moves 8-5 5-2 5-2 8-5 .
1.015 16-7* 5-2* 1.061
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O X | | X X O O |
| O X | | X X |
| O X | | X |
| O | | X |
| | | X |
v| |BAR| | 11-point match
| | | |
| | | O |
| X | | O |
| X O | | O O X |
| X O O | | O O X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 10
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 2 and 2.
X moves 19-21 21-23 19-21 21-23 .
-.070 (gammon chances 11%) 22(2) 4(2) .081 (gammon chances 8%)
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O X | | X X X O |
| O X | | X X X |
| O X | | X |
| O | | |
| | | |
v| |BAR| | 11-point match
| | | |
| | | O |
| X | | O |
| X O | | O O X |
| X O O | O | O O X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 10
BAR: O-1 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 5 and 6.
X moves 19-24 12-18 .
-.058 (G C 11.1) 13 +.150 (GC 7.3)
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O X | | X X O X X |
| O X | | X X X X |
| O | | |
| O | | |
| | | |
v| |BAR| | 11-point match
| | | |
| | | O |
| | | O |
| X O | | O O X |
| X O O | O | O O X X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 10
BAR: O-1 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 6 and 1.
X moves 3-9 17-18 .
.126 9,2 .188
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O | | X X O X X X |
| O | | X X O X X X |
| O | | |
| O | | |
| | | |
v| |BAR| | 11-point match
| | | |
| | | O |
| | | O |
| O | | O O X |
| O | O | O X O X X X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 10
BAR: O-1 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 4 and 2.
X moves 1-3 5-9 .
-.141 1-7 -.078
###
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O X | | X X O X X X |
| O | | X X O X X X |
| | | O |
| | | |
| | | |
v| |BAR| | 11-point match
| | | |
| O | | |
| O | | O O |
| X O | | O O |
| X O | | O X O X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 10
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: O
O rolls 3 and 2.
O moves 8-6 4-1 .
-.001 8-5*-3 +.074
+13-14-15-16-17-18-------19-20-21-22-23-24-+ X: - score: 3
| O X | | X X O X X X |
| O | | X X O X X X |
| | | O |
| | | |
| | | |
v| |BAR| | 11-point match
| | | |
| | | O |
| O | | O |
| X O | | O O |
| X O | | O X O X O |
+12-11-10--9--8--7--------6--5--4--3--2--1-+ O: - score: 10
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 1 turn: X
X rolls 5 and 2.
X moves 5-10 10-12 .
-.024 7(2) +.096
###
+-1--2--3--4--5--6--------7--8--9-10-11-12-+ O: - score: 5
| O O O O | | |
| O O O O | | |
| O O O | | |
| O O | | |
| O | | |
| |BAR| |v 11-point match
| X | | |
| X | | |
| X | | X |
| X X X | | X X |
| O X X X | | X X X |
+24-23-22-21-20-19-------18-17-16-15-14-13-+ X: - score: 10
BAR: O-0 X-0 OFF: O-0 X-0 Cube: 2 (owned by X) turn: X
X rolls 3 and 1.
X moves 15-18 19-20 .
.516 19-22 19-20 .577 (How do you like this one?)
X wins the 11 point match 12-5 .
FLMaster39
One thing I noticed in looking at these positions is that in many cases
JellyFish prefered the simple, more natural-looking play, while the
players prefered cuter, more aggressive ones.
Of course, we're "assuming" here that JellyFish is actual superior to the
players in evaluating the positions. If it is, then it would be
interesting - that the gap between the best and the intermediate players
might be greater than we think. Since the best players obviously win
more, and do so by making inferior moves that an intermediate player would
find, there must be huge swings elsewhere in the game. (Again, assuming
that JellyFish is right.)
As a separate question, I've been wondering about something for some time.
Computers are equal to the best at Chess, vastly inferior at bridge, and
better at backgammon. Why? I have a personal suspicion, but I'd be
interested in comments.
When I play backgammon I play by positional principles. I try to look
ahead to the position I want to create, then go after it. If I have three
options for a move, I don't really count, after that move, how many Great,
Good, Average, Poor, and Just Plain Rotten rolls my opponents has next
roll, let alone my responses. I doubt that I could even identify that
play A gives me 13 rolls to make my 5-point while leaving 6 hit numbers,
while play B gives me 14 rolls to amke my five-point while leaving only 5
shots.
A computer can do those things instantly. Often when I play against
weaker players they will complain about my luck, and I will say "Good
players make their own luck." They do, by playing so as to maximize the
number of favorable rolls. I wonder if a computer doesn't have a huge
advantage just by being able to quickly calculate a one-ply lookahead,
where a human can't do the math over the board.
Suppose you could get just one extra solidly good roll per game. Not a
miracle, like the doubles that enter 3 from the bar against a 5-point
board and hit besides, but.... well, say an opening 3-1. At best, 3-1
makes you 60% to win the game. Now, suppose the average game consists of
36 rolls (probably a bit high but allow me for this example). Suppose
that on each roll your opponents could have just ONE more roll that turned
a mediocre roll into a good one. So once per game they get the equivalent
of an opening 3-1 (since no opening roll leaves you worse than about 49%
to win). Now, would Kit or KG or any of those want to play me and spot me
an opening 3-1 every game? I've never pretended to be close to their
class, but I still doubt they'd give me that much of a handicap.
Anyway, I wonder if anyone would care to comment on my speculation about
why computers play so well.
For those who care (probably damn few) the bridge Nationals start today,
and there is a contest in which some computer bridge programs will play
against each other and against humans. It will be interesting to see what
the results are. However, some bridge programs actually DO cheat!! The
authors admit it. They peek at the opponent's cards. The authors justify
this on the grounds that people do not necessarily expect to "compete"
even-up with computers, rather they just want a competitive game to
practice against. Since they can't program a strong game, they strengthen
the game in other ways. However it would be very easy to catch a program
cheating. Just feed it a hand, then rearrange the opponent's cards but
not its own and see if it bids and plays the same way.
Slightly off-topic I know. Sorry.
Chuck Bower
In article <5rd30a$hpg$1...@dismay.ucs.indiana.edu>,
Chuck Bower <bo...@bigbang.astro.indiana.edu> wrote:
> I heard
> through the grapevine (danger! danger!) that recently Malcolm Davis
> conducted an "experiment" in Texas. Nick Ballard and Mike Senkiewicz
> each played a 300 game (REAL) money session against JFv3.0 level-7.
> Harvey Huie backed Jellyfish in the $200/point one-on-one contest.
> JF had "Caution" setting turned on. The final results: one player
> finished up 58 pts, the other down by the same amount. All games
> were recorded. Just a rumor...
>
(snip)
>
>Hmmm. Based on the light level generated in our newsgroup over SUSPECTED
>cheating by bots, I bet the bridge newsgroup is at times solar powered!
>
And before our newsgroup again becomes lucent, let me add that I also
heard that for Davis's above mentioned experiment, ALL DICE ROLLS were
generated by humans.
Chuck
bo...@bigbang.astro.indiana.edu
c_ray on FIBS
Chuck Bower
In article <19970725200...@ladder02.news.aol.com>,
FLMaster39 <flmas...@aol.com> wrote:
(snip)
>
>As a separate question, I've been wondering about something for some time.
> Computers are equal to the best at Chess, vastly inferior at bridge, and
>better at backgammon.
I think you are exaggerating a bit here. Last year in a post to this
newsgroup, Kit Woolsey estimated that the best neural net BG players
were on a par with the best humans. (I'm paraphrasing.) I heard
through the grapevine (danger! danger!) that recently Malcolm Davis
conducted an "experiment" in Texas. Nick Ballard and Mike Senkiewicz
each played a 300 game (REAL) money session against JFv3.0 level-7.
Harvey Huie backed Jellyfish in the $200/point one-on-one contest.
JF had "Caution" setting turned on. The final results: one player
finished up 58 pts, the other down by the same amount. All games
were recorded. Just a rumor...
(snip)
>Suppose you could get just one extra solidly good roll per game. Not a
>miracle, like the doubles that enter 3 from the bar against a 5-point
>board and hit besides, but.... well, say an opening 3-1. At best, 3-1
>makes you 60% to win the game.
Recent JFv3.0 level-6 cubeless rollouts say 31 opener is 5::4 favorite
to win (JF vs. JF!!) and when you throw in gammons, +0.164 in cubeless
equity.
>Now, suppose the average game consists of
>36 rolls (probably a bit high but allow me for this example). Suppose
>that on each roll your opponents could have just ONE more roll that turned
>a mediocre roll into a good one. So once per game they get the equivalent
>of an opening 3-1 (since no opening roll leaves you worse than about 49%
>to win). Now, would Kit or KG or any of those want to play me and spot me
>an opening 3-1 every game? I've never pretended to be close to their
>class, but I still doubt they'd give me that much of a handicap.
I've been reviewing some of my FIBS matches lately, using JFv3.0 level-7
evaluation as my guru. I get JF's opinion after EVERY play and record and
tally the results. For example, in a recent 7-point match (where I lost
6-7), ten games were played. I averaged pitching 0.16 points per game
on checker plays; my opp. did worse (according to JF) giving back 0.27
equity (average) per game. (This is money, cubeless equity, with
adjustments made for late in the match when gammons count differently
than at $ play.) I also kept track of cubes (which is trickier to
tally). Here I fared worse than my opp. -0.06/game compared to -0.04.
It is interesting that MY per game checker play deficit (-0.16 cubeless
equity) just about equals the advantage of the 31 opening! Maybe
FLMaster (is that your REAL name?) is onto something here.
(snip)
> However, some bridge programs actually DO cheat!! The
>authors admit it. They peek at the opponent's cards. The authors justify
>this on the grounds that people do not necessarily expect to "compete"
>even-up with computers, rather they just want a competitive game to
>practice against. Since they can't program a strong game, they strengthen
>the game in other ways.
(snip)
Hmmm. Based on the light level generated in our newsgroup over SUSPECTED
cheating by bots, I bet the bridge newsgroup is at times solar powered!
Fredrik Dahl
Chuck Bower wrote:
>
> In article <5rd30a$hpg$1...@dismay.ucs.indiana.edu>,
>
> Chuck Bower <bo...@bigbang.astro.indiana.edu> wrote:
>
> > through the grapevine (danger! danger!) that recently Malcolm Davis
> > conducted an "experiment" in Texas. Nick Ballard and Mike Senkiewicz
> > each played a 300 game (REAL) money session against JFv3.0 level-7.
> > Harvey Huie backed Jellyfish in the $200/point one-on-one contest.
> > JF had "Caution" setting turned on. The final results: one player
> > finished up 58 pts, the other down by the same amount. All games
> > were recorded. Just a rumor...
> >
(snip)
>
> heard that for Davis's above mentioned experiment, ALL DICE ROLLS were
> generated by humans.
>
Yes, this is true. I like to believe that Jelly was unlucky, but I hope
others believe the opposite, so that it can be done again. I would also
happily back JF3.0 level 7 against any human in short matches.
--
- Fredrik Dahl
helmet
In article <33DA9E...@sn.no> Fredrik Dahl <fre...@sn.no> writes:
>Path: ihug.co.nz!chi-news.cic.net!199.60.229.3!newsfeed.direct.ca!news-feed.inet.tele.dk!uninett.no!sn.no!not-for-mail
>From: Fredrik Dahl <fre...@sn.no>
>Newsgroups: rec.games.backgammon
>Subject: JF vs humans
>Date: Sat, 26 Jul 1997 18:04:04 -0700
>Organization: SN Internett
>Lines: 28
>Message-ID: <33DA9E...@sn.no>
>References: <5ra8vp$e...@prometheus.acsu.buffalo.edu> <19970725200...@ladder02.news.aol.com> <5rd30a$hpg$1...@dismay.ucs.indiana.edu> <5rd47i$3le$1...@dismay.ucs.indiana.edu>
>Reply-To: fre...@sn.no
>NNTP-Posting-Host: lorenskog6-2.ppp.sn.no
>Mime-Version: 1.0
>Content-Type: text/plain; charset=us-ascii
>Content-Transfer-Encoding: 7bit
>X-Mailer: Mozilla 3.0 (Win16; I)
>Xref: ihug.co.nz rec.games.backgammon:8558
>Chuck Bower wrote:
>>
>> In article <5rd30a$hpg$1...@dismay.ucs.indiana.edu>,
>>
>> Chuck Bower <bo...@bigbang.astro.indiana.edu> wrote:
>>
>> > I heard
>> > through the grapevine (danger! danger!) that recently Malcolm Davis
>> > conducted an "experiment" in Texas. Nick Ballard and Mike Senkiewicz
>> > each played a 300 game (REAL) money session against JFv3.0 level-7.
>> > Harvey Huie backed Jellyfish in the $200/point one-on-one contest.
>> > JF had "Caution" setting turned on. The final results: one player
>> > finished up 58 pts, the other down by the same amount. All games
>> > were recorded. Just a rumor...
>> >
>(snip)
>>
>> And before our newsgroup again becomes lucent, let me add that I also
>> heard that for Davis's above mentioned experiment, ALL DICE ROLLS were
>> generated by humans.
very sensible i wouldnt trust jf to roll its own dice its not named fishy for
nothing
helmet
Daniel Murphy
On 26 Jul 1997 14:52:58 GMT, bo...@bigbang.astro.indiana.edu (Chuck
Bower) wrote:
> I think you are exaggerating a bit here. Last year in a post to this
> newsgroup, Kit Woolsey estimated that the best neural net BG players
> were on a par with the best humans. (I'm paraphrasing.) I heard
> through the grapevine (danger! danger!) that recently Malcolm Davis
> conducted an "experiment" in Texas. Nick Ballard and Mike Senkiewicz
> each played a 300 game (REAL) money session against JFv3.0 level-7.
> Harvey Huie backed Jellyfish in the $200/point one-on-one contest.
> JF had "Caution" setting turned on. The final results: one player
> finished up 58 pts, the other down by the same amount. All games
> were recorded. Just a rumor...
Of the true variety. Notice that Jellyfish broke even, proving, said
one crouton, that Jellyfish is just an average player.
(Yes, that was a joke.)
Question: Assuming two *equal* players sit down and play 300 games,
what's the chance of one of them winning or losing 58 points?
______________________________________________________________________
Daniel Murphy San Francisco, California rac...@cityraccoon.com
FIBS, the First Internet Backgammon Server
http://www.fibs.com, telnet://fibs.com:4321
Backgammon By the Bay: monthly tourneys in San Francisco, annotated games:
http://www.backgammon.org/bgbb/
Fredrik Dahl
Daniel Murphy wrote:
>
> Question: Assuming two *equal* players sit down and play 300 games,
> what's the chance of one of them winning or losing 58 points?
>
A single game with normal use of the cube has a standard deviation of
about 3 points. Multiply this with the square root of 300, and get
approximately 50. So the standard deviation after 300 games is about 50
points. About a 3rd of the time one of the players will win more than 1
standard deviation, so a win or loss of 58 points is pretty normal.
It is surprising how random the results from long playing sessions are.
Even if the players are exactly equal in strength, it can happen (2% or
so) that one side wins more than 100 points in 300 games.
It makes you wonder how significant it is to win a tournament, doesn't
it?
--
- Fredrik Dahl
Chuck Bower
>Daniel Murphy wrote:
>>
>> Question: Assuming two *equal* players sit down and play 300 games,
>> what's the chance of one of them winning or losing 58 points?
>>
>
In article <33DB92...@sn.no>, Fredrik Dahl <fre...@sn.no> wrote:
>A single game with normal use of the cube has a standard deviation of
>about 3 points. Multiply this with the square root of 300, and get
>approximately 50. So the standard deviation after 300 games is about 50
>points. About a 3rd of the time one of the players will win more than 1
>standard deviation, so a win or loss of 58 points is pretty normal.
>
>It is surprising how random the results from long playing sessions are.
>Even if the players are exactly equal in strength, it can happen (2% or
>so) that one side wins more than 100 points in 300 games.
(snip)
Fredrik didn't say where he got the std dev. value (3) he used in
his calculation. Below are compilations of results for ME vs. three
different bots (each running highest available level):
points JF1.0 TDGammon JF2.0
1 592(36.5) 222(39.9) 263(45.6)
2 654(40.4) 229(41.2) 202(35.0)
4 315(19.4) 88(15.8) 90(15.6)
6 12( 0.7) 0 5( 0.9)
8 45( 2.8) 17( 3.1) 14( 2.4)
16 2( 0.1) 0 3( 0.5)
--------- --------- ---------
totals 1620 556 577
std. dev 2.73 2.56 2.75
(numbers in parentheses are % of total games played against that bot)
I consider myself tighter with the cube than most. Thus these
numbers, though somewhat smaller than 3, add credibility to Fredrik's
calculation.
Alexander Nitschke
Chuck Bower wrote:
>
> Fredrik didn't say where he got the std dev. value (3) he used in
> his calculation. Below are compilations of results for ME vs. three
> different bots (each running highest available level):
>
> points JF1.0 TDGammon JF2.0
>
> 1 592(36.5) 222(39.9) 263(45.6)
> 2 654(40.4) 229(41.2) 202(35.0)
> 4 315(19.4) 88(15.8) 90(15.6)
> 6 12( 0.7) 0 5( 0.9)
> 8 45( 2.8) 17( 3.1) 14( 2.4)
> 16 2( 0.1) 0 3( 0.5)
> --------- --------- ---------
> totals 1620 556 577
>
> std. dev 2.73 2.56 2.75
>
> (numbers in parentheses are % of total games played against that bot)
>
> I consider myself tighter with the cube than most. Thus these
> numbers, though somewhat smaller than 3, add credibility to Fredrik's
> calculation.
>
> Chuck
> bo...@bigbang.astro.indiana.edu
> c_ray on FIBS
I made such statistics too and got to a similar result: The standard
deviation is about 2.7. But I doubt your calculations are fully correct,
because you seem to assume a mean of zero. It changes the standard
deviation, though, if the mean was different from zero, eg +0.2 per game
for you. So you have to note all occurences of -1, -2, -4, ... as the one
of +1, +2, +4, ... separately. But if you didn't want to show your
performance, I can understand this :-))
Best wishes,
Alexander
lee
In article <33DA9E...@sn.no>, Fredrik Dahl <fre...@sn.no> writes
>
>Yes, this is true. I like to believe that Jelly was unlucky, but I hope
>others believe the opposite, so that it can be done again. I would also
>happily back JF3.0 level 7 against any human in short matches.
>
--
----------------------------------------------
l...@infoplus.demon.co.uk
Web site: http://www.ibmpcug.co.uk/~oak/info/
Email for links to/from this site
----------------------------------------------
Robert-Jan Veldhuizen
On 26-jul-97 22:44:41, helmet wrote:
>>> And before our newsgroup again becomes lucent, let me add that I also
>>> heard that for Davis's above mentioned experiment, ALL DICE ROLLS were
>>> generated by humans.
> very sensible i wouldnt trust jf to roll its own dice its not named
> fishy for nothing
> helmet
So why are you called helmet then?
--
Zorba/Robert-Jan
Chuck Bower
In article <33DBCD...@mailszrz.zrz.tu-berlin.de>,
Alexander Nitschke <nits...@mailszrz.zrz.tu-berlin.de> wrote:
>Chuck Bower wrote:
>>
>> Fredrik didn't say where he got the std dev. value (3) he used in
>> his calculation. Below are compilations of results for ME vs. three
>> different bots (each running highest available level):
>>
>> points JF1.0 TDGammon JF2.0
>>
>> 1 592(36.5) 222(39.9) 263(45.6)
>> 2 654(40.4) 229(41.2) 202(35.0)
>> 4 315(19.4) 88(15.8) 90(15.6)
>> 6 12( 0.7) 0 5( 0.9)
>> 8 45( 2.8) 17( 3.1) 14( 2.4)
>> 16 2( 0.1) 0 3( 0.5)
>> --------- --------- ---------
>> totals 1620 556 577
>>
>> std. dev 2.73 2.56 2.75
>>
>> (numbers in parentheses are % of total games played against that bot)
>>
>> I consider myself tighter with the cube than most. Thus these
>> numbers, though somewhat smaller than 3, add credibility to Fredrik's
>> calculation.
>>
>> Chuck
>> bo...@bigbang.astro.indiana.edu
>> c_ray on FIBS
Alexander's response:
>
>I made such statistics too and got to a similar result: The standard
>deviation is about 2.7. But I doubt your calculations are fully correct,
>because you seem to assume a mean of zero. It changes the standard
>deviation, though, if the mean was different from zero, eg +0.2 per game
>for you. So you have to note all occurences of -1, -2, -4, ... as the one
>of +1, +2, +4, ... separately. But if you didn't want to show your
>performance, I can understand this :-))
Correct about using the true mean instead of zero. I was mainly
trying to get a "ballpark" feel as to the veracity of Fredrik's contention
that REAL money games have a standard deviation of 3 points. I then used
my data and ASSUMED that a typical money game BETWEEN EQUAL PLAYERS
would have a similar distribution of outcomes (that is games ending with
the winner garnering 1, 2, 4, 6, 8,... points).
As far as Alexander's "tongue-in-cheek" speculation about the mean
of my results with the three bots, I think I already posted that a few
months back. GOTCHA! My recollection is that my money play results
were on the order -0.3 against JFv1.0 level-6, -0.2 against TD-Gammon
(the version that come with the "Family Funpack" for OS/2 Warp 3.0)
and -0.4 against JFv2.0 level-7. Play was always money, Jacoby, NO
automatics (at least where these rules are even available). Keeping
with the tradition of playing the lastest, bestest commercially available
bot, I now have graduated to JFv3.0 level-7. It's too early to give
results on my performance (since I've played it less than 100 games).