Opening move rollouts
Alan Webb
I mention this as in Magriel's book "Backgammon", Magriel recommends
bringing 2 builders down with the opening roll 5-3 (13/8 13/10), rather than
the more common (and my current) move of making the 3 point. I understand
both the advantages and disadvantages of both these moves however I am
interested in what extensive rollouts have to say.
I still find myself varying the 4-3 opening roll for example. Everybody
seems to have their own preference and normally sound reasons to back them
up, but there must be some hard concrete evidence to show that play X is
better than play Y. I can imagine that the sample would have to be huge as
it is only the opening roll. One could also question as to whether rollouts
based on an opening roll could tell you anything of worth in any case.
regards
Alan Webb
JP White
> Has anyone done extensive rollouts on opening rolls and their combinations?
>
> I mention this as in Magriel's book "Backgammon", Magriel recommends
> bringing 2 builders down with the opening roll 5-3 (13/8 13/10), rather than
> the more common (and my current) move of making the 3 point. I understand
> both the advantages and disadvantages of both these moves however I am
> interested in what extensive rollouts have to say.
I suggest you visit http://www.northcoast.com/~mccool/kwopnrls.htm There is an
archive of a posting from Kit Woolsey from 1995 where he went over current
opening plays and theories. Some of the top players have rethought some of the
opening moves as a direct result of the bots analyses.
The 5-3 as per Magriel is now considered to be much inferior to making the 3
point as you are doing.
>
> I still find myself varying the 4-3 opening roll for example. Everybody
> seems to have their own preference and normally sound reasons to back them
> up, but there must be some hard concrete evidence to show that play X is
> better than play Y. I can imagine that the sample would have to be huge as
> it is only the opening roll. One could also question as to whether rollouts
> based on an opening roll could tell you anything of worth in any case.
>
Kit said that he thinks 13/10 13/9 and 24/20 13/10 are about equivalent. 24/21
13/9 as a close third.
Of course this is from a post from March 1995. Theory may have advanced once
again. It would be interesting to see what the top players think at the dawn of
a new millennium.
JP
--
JP White
Mailto:jp.w...@nashville.com
Alan Webb
JP White schrieb in Nachricht <36C64282...@nashville.com>...
>Alan Webb wrote:
>
>> Has anyone done extensive rollouts on opening rolls and their
combinations?
>>
>
an
>archive of a posting from Kit Woolsey from 1995 where he went over current
>opening plays and theories.
Tried that but got an error.
>The 5-3 as per Magriel is now considered to be much inferior to making the
3
>point as you are doing.
>
I had heard as much. There are a fair few opening rolls in Magriels book
which seem to go against the grain of modern strategic thinking. Not running
with 6-4 is another example.
I was of the opinion the rethink was due to extensive rollouts and that they
may have been published somewhere.
>> I still find myself varying the 4-3 opening roll for example. Everybody
>> seems to have their own preference and normally sound reasons to back
them
>> up, but there must be some hard concrete evidence to show that play X is
>> better than play Y. I can imagine that the sample would have to be huge
as
>> it is only the opening roll. One could also question as to whether
rollouts
>> based on an opening roll could tell you anything of worth in any case.
>>
>
>Kit said that he thinks 13/10 13/9 and 24/20 13/10 are about equivalent.
24/21
>13/9 as a close third.
>
Was this based on rollout analysis or Kits extensive experience?
>Of course this is from a post from March 1995. Theory may have advanced
once
>again. It would be interesting to see what the top players think at the
dawn of
>a new millennium.
>
Yes it would. If I had Snowie pro 2 I think I'd do extensive rollouts on
each. IF I could be reasonably certain that rollouts from an opening move
would be accurate. I have my doubts though due to the fact that it is only
the first move and the relevance could be lost in most rollouts by the end
of each rollout.
regards
Alan
Bob Stringer
> Alan Webb wrote:
>
> > Has anyone done extensive rollouts on opening rolls and their combinations?
> >
> > I mention this as in Magriel's book "Backgammon", Magriel recommends
> > bringing 2 builders down with the opening roll 5-3 (13/8 13/10), rather than
> > the more common (and my current) move of making the 3 point. I understand
> > both the advantages and disadvantages of both these moves however I am
> > interested in what extensive rollouts have to say.
>
> I suggest you visit http://www.northcoast.com/~mccool/kwopnrls.htm There is an
> archive of a posting from Kit Woolsey from 1995 where he went over current
> opening plays and theories. Some of the top players have rethought some of the
> opening moves as a direct result of the bots analyses.
>
I clicked on the link you provided, and got a message that there was "an error
retrieving your document." A directory of customer web sites also was provided,
but I couldn't figure out how it led to the posting from Kit Woolsey.
Any advice how to get to the posting?
Thanks.
Bob Stringer
--
To reply, please delete
REMOVE from my address
JP White
>
>
> I clicked on the link you provided, and got a message that there was "an error
> retrieving your document." A directory of customer web sites also was provided,
> but I couldn't figure out how it led to the posting from Kit Woolsey.
>
> Any advice how to get to the posting?
>
> Thanks.
>
> Bob Stringer
> --
Hey Bob, I came across the article in DejaNews.
> Re: List of Opening Moves
> Author: Petko Boyadjiev
> Date: 1997/04/05
> Forum: rec.games.backgammon
>
> Opening Rolls
> Kit Woolsey
> Subject: Re: What are the best ways to play the opening rolls?
> ---------------------------------------------------------------------------
>
> It is quite true that rollout results from three backgammon playing
> computer programs (Expert Backgammon, TD-Gammon, and Jellyfish) have
> given
> us new insights into opening rolls and other phases of the game. Before
> taking any of these as gospel, there are several things to keep in mind:
>
> 1. The strengths and weakness of the programs. For example, a program
> which is weak in backgame play might downgrade early slotting plays
> in
> its rollouts since these plays will probably lead to backgames more
> often when the slotted blots are hit.
>
> 2. The rollouts do not take into account access to the cube, which
> might
> make a difference.
>
> 3. There can be quite a bit of variance in the rollouts due to luck,
> even
> with large sample sizes.
>
> 4. The rollouts do not take into account individual personalities of
> players playing. A player may well do better with an "inferior"
> opening play if it suits his personal style.
>
> 5. If playing a match, the match score may affect the value of
> different
> plays.
>
> Slotting the five was now generally considered inferior to splitting the
> back men to use the 1 for some of the opening rolls.
> ---------------------------------------------------------------------------
>
> Now, on the what I believe is an accurate synopsis of the 15 possible
> opening rolls:
>
> * 2-1: The slotting play 13/11, 6/5 and the splitting play 24/23,
> 13/11,
> the two most common plays, seem to be about equal. Nothing else is
> a
> serious contender.
>
> * 3-1: 8/5, 6/5 is obviously the only play.
>
> * 4-1: The splitting play 24/23, 13/9 has come out clearly superior
> to
> the slotting play 13/9, 6/5. Probably the reason is that with the
> builder on the 9 point there are so many good pointing numbers next
> turn anyway that you don't need the 5 point slotted.
>
> * 5-1: The splitting play 24/23, 13/8 has come out a bit better than
> the
> slotting play 13/8, 6/5. A third less common alternative, 24/18,
> came
> out clearly worse.
>
> * 6-1: The obvious 13/7, 8/7 is correct. Magriel's experiment of
> 13/7,
> 6/5 is awful.
>
> * 3-2: The splitting play 24/21, 13/11 came out a bit better than
> building with 13/10, 13/11.
>
> * 4-2: 8/4, 6/4 of course.
>
> * 5-2: The normal play for years has been 13/11, 13/8. However the
> newer
> splitting play, 24/22, 13/8, (shunned because of the crushing 5-5
> threat) has come out a bit better. The slotting play of 13/8, 6/4
> (which used to be my choice) did not survive the rollouts -- it was
> clearly inferior.
>
> * 6-2: The splitting play of 24/18, 13/11 comes out fairly clearly
> superior. Running with 24/16 is 2nd, but the run isn't far enough.
> Slotting with 13/5 (a common choice several years ago) was
> definitely
> in third place.
>
> * 4-3: The building play of 13/10, 13/9 and the common splitting play
> of
> 24/20, 13/10 were just about tied. The alternative split of 24/21,
> 13/9 was only a little behind.
>
> * 5-3: The simple 8/3, 6/3 is clearly best. The once common 13/10,
> 13/8
> has been found vastly inferior.
>
> * 6-3: The splitting 24/18, 13/10 comes out best, but the running
> play
> of 24/15 is not too far behind.
>
> * 5-4: Splitting with 24/20, 13/8 and building with 13/9, 13/8 come
> out
> quite close (that builder on the 9 point is powerful), with the
> split
> generally a tiny bit better. 24/15 is weaker still.
>
> * 6-4: Both running with 24/14 and splitting with 24/18, 13/9 are
> about
> equal. However the once laughed at 8/2, 6/2 has reared its head as
> a
> serious contender and comes out about equal with the other choices
> --
> nice play to try if you get familiar with it, since your opponent
> probably won't be.
>
> * 6-5: The simple 24/13 is clearly better than any other
> possibilities.
JP White
> JP White wrote:
>
> > I suggest you visit http://www.northcoast.com/~mccool/kwopnrls.htm There is an
> > archive of a posting from Kit Woolsey from 1995 where he went over current
> > opening plays and theories. Some of the top players have rethought some of the
> > opening moves as a direct result of the bots analyses.
> >
>
> I clicked on the link you provided, and got a message that there was "an error
> retrieving your document." A directory of customer web sites also was provided,
> but I couldn't figure out how it led to the posting from Kit Woolsey.
>
> Any advice how to get to the posting?
>
> Thanks.
>
> Bob Stringer
>
It would appear that the posting I retrieved in January of this year from the above
address is no longer there!
Sorry for the bum steer, I didn't bother to check the URL since it isn't long since
it was good.
To make up for this 'tease' I have researched rgb on Dejanews and came up with an
article by Midas, as Follows
> Opening Rolls Rollouts JellyFish Ver 3.0
>
> Author: Midas
> Date: 1997/09/23
> Forum: rec.games.backgammon
>
> For those of you interested I've rolled out all the opening moves top
> contenders. Trials 1296 Standard Deviations 0.008. I'm not sure of the
> statistical relevance of the data with this S.D. but they,re here for
> your perusal.A few points I noticed though were slotting ones own 5
> point with 41 is definitely an inferior choice, for 21 its definitely
> second place but for 51 seems like a good choice given the general
> accepted opinion is that JellyFish doesnt "steer" its games towards
> backgames.
> Spliting to the opponents 5 point with 43 and 54 produced better
> results in the rollouts.
> The inner board point making numbers 31 42 53 64 all produced equities
> higher than the evaluation.
>
> Top moves marked with *
>
> Dice Move Move L7 EV L6 Roll Dice Move Move
> L7 EV L6 Roll
> 21* 13/11 24/23 -.005 +.017 62 24/18 13/11
> +.010 +.006
> 21 13/11 6/5 -.007 -.008 62* 24/16
> -.003 +.008
> 21 24/21 -.018 -.008 62 13/5
> -.020 -.015
> 21 13/10 -.026 -.027 62 24/18 24/22
> -.041 -.042
> 21 24/22 24/23 -.035 -.056
>
> 43 13/9 24/21
> +.010 -.003
> 31* 8/5 6/5 +.140 +.155 43 13/9 13/10
> +.005 -.004
> 43* 24/20 13/10
> +.001 +.009
> 41* 13/9 24/23 -.010 -.006 43 24/20 24/21
> -.007 +.003
> 41 13/8 -.022 -.037
>
> 41 24/20 24/23 -.026 -.023 53* 8/3 6/3
> +.052 +.077
> 41 13/9 6/5 -.032 -.031 53 13/8 24/21
> +.029 +.024
> 41 24/20 6/5 -.048 -.066 53 13/8 13/10
> +.008 -.003
>
>
> 51* 13/8 24/23 +.008 +.004 63* 24/18 13/10
> +.002 .000
> 51 13/8 6/5 -.014 -.004 63* 24/15
> +.002 .000
> 51 24/18 -.025 -031 63 24/18 24/21
> -.031 -.047
> 63 13/4
> -.039 -.037
> 61* 13/7 8/7 +.098 +.107
>
> 54* 13/8 24/20
> +.025 +.034
> 32* 24/21 13/11 +.002 +.011 54 13/8 13/9
> +.022 +.015
> 32 13/10 13/11 -.005 +.004 54 24/15
> +.002 .000
> 32 13/10 24/22 -.007 -.008
>
> 32 13/8 -.022 -.037 64 24/14
> +.014 +.007
> 32 24/21 24/22 -.027 -.027 64 24/18 13/9
> +.011 +.008
> 32 13/10 6/4 -.049 -.038 64* 8/2 6/2
> +.005 +.011
>
>
> 42* 8/4 6/4 +.095 +.122 65* 24/13
> +.079 +.066
>
>
> 52 13/8 24/22 +.010 +.001
>
> 52* 13/8 13/11 +.008 +.007
>
> 52 13/8 6/4 -.030 -.041
>
> 52 13/6 -.040 -.054
>
Hope this helps.
If I get the time, I'll scan the post and repost if you're interested.
JP White
Alan Webb wrote:
> >
> >Kit said that he thinks 13/10 13/9 and 24/20 13/10 are about equivalent.
> 24/21
> >13/9 as a close third.
> >
>
> Was this based on rollout analysis or Kits extensive experience?
>
I just don't know. A bit of both I wouldn't wonder.
>
> >Of course this is from a post from March 1995. Theory may have advanced
> once
> >again. It would be interesting to see what the top players think at the
> dawn of
> >a new millennium.
> >
>
> Yes it would. If I had Snowie pro 2 I think I'd do extensive rollouts on
> each. IF I could be reasonably certain that rollouts from an opening move
> would be accurate. I have my doubts though due to the fact that it is only
> the first move and the relevance could be lost in most rollouts by the end
> of each rollout.
>
> regards
>
> Alan
See also my reply to Bob Stringer in this thread. I posted an analysis of
opening moves that Midas performed with Jelly 3.0 some time back.
stig...@yahoo.com
> I suggest you visit http://www.northcoast.com/~mccool/kwopnrls.htm There is an
> > archive of a posting from Kit Woolsey from 1995 where he went over current
> > opening plays and theories. Some of the top players have rethought some of
the
> > opening moves as a direct result of the bots analyses.
> >
>
> I clicked on the link you provided, and got a message that there was "an error
> retrieving your document." A directory of customer web sites also was
provided,
> but I couldn't figure out how it led to the posting from Kit Woolsey.
>
> Any advice how to get to the posting?
>
> Thanks.
>
> Bob Stringer
> To reply, please delete
> REMOVE from my address
>
Paal Fladsrud has this article on his page:
http://classic.csc.no/~paalf/BG/opening.txt
Stig Eide
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Robert-Jan Veldhuizen
<we...@hannover.sgh-net.de> wrote:
>I had heard as much. There are a fair few opening rolls in Magriels book
>which seem to go against the grain of modern strategic thinking. Not running
>with 6-4 is another example.
I thought the 6-4 opening roll was one of those examples where the
differences are small and matchscore comes into play.
At DMP 24/14;
trailing post-crawford when gammons (will) count 24/18 13/9;
needing only one point but gammons (will) count for your opp 8/2 6/2;
otherwise 24/18 13/9 or 24/14.
if I recall corectly but maybe recent bots have changed the opinions
here?
Another thing to consider is your opponent's and your own play style and
ability.
24/18 13/9 more often leads to more complicated games, priming games or
lots of men back.
24/14 prepares for a sometimes pretty simple race, depending on how well
you know how to play with (and your opponent how to defend against) one
man back.
8/2 6/2 can be useful for blitzing purposes or opponents that are (too)
afraid to split their backmen after this move, it can also lead to more
easy games; at least you won't get hit after this move :-)
--
Robert-Jan/Zorba
Kevin Bastian
1995 Kit Woolsey newsgroup post.
By the way, the article is still there. The address was just posted
incorrectly (by one letter).
Correct address is: http://www.northcoast.com/~mccool/kwopnrls.html
Humorous anecdote: I was actively learning backgammon in 95, and read Kit's
list of opening moves with great interest. I had previously memorized
recommended openers from older books I'd read, but modified one or two
moves based on his then-recent article.
In April of 96 McCool (Kate) and I were at the San Diego Gran Prix and
decided to enter the doubles tournament. Only 4 couples entered, and when
names were drawn for the pairings, we heard the words no sane doubles team
would EVER want to hear ("Kevin and Kate, you are playing the Woolseys.")
(Single elimination. Quickest hundred bucks I ever lost since I had my
wallet stolen in college!)
Anyway, at one point, Kit and Sally rolled, and I don't remember what the
roll was, but Kit moved it differently than he had recommended in his post!
I said, "Kit, I thought you recommended so-and-so in your newsgroup posting
last year." He wryly said "I don't remember ever recommending any such
thing." (Kate and Sally were chuckling at this point.)
Just to be sure, after the tournament, I rechecked the list. Sure enough,
he HAD recommended it. So even Kit changes his mind, or else just likes to
mix things up. Probably was worth it to him just to pull my chain a bit :-)
Oh well. I can still remember how nice and crisp those two $50 bills
were... *sigh*
Bob Stringer <REMOVE...@pacbell.net> wrote in article
<36C7173B...@pacbell.net>...
> JP White wrote:
>
> > Alan Webb wrote:
> >
> > > Has anyone done extensive rollouts on opening rolls and their
combinations?
> > >
> > > I mention this as in Magriel's book "Backgammon", Magriel recommends
> > > bringing 2 builders down with the opening roll 5-3 (13/8 13/10),
rather than
> > > the more common (and my current) move of making the 3 point. I
understand
> > > both the advantages and disadvantages of both these moves however I
am
> > > interested in what extensive rollouts have to say.
> >
Daniel Murphy
wrote:
Hi JP and all;
The article you found by Midas and by Kit Woolsey through Dejanews ...
>> Opening Rolls Rollouts JellyFish Ver 3.0
>> Author: Midas
>> Date: 1997/09/23
> Opening Rolls
> Kit Woolsey
> Subject: Re: What are the best ways to play the opening rolls?
... are available (with 100's of others) in the rec.games.backgammon
newsgroup archive at: Backgammon Galore, www.bkgm.com. See:
http://www.bkgm.com/rgb/rgb.cgi?view+383.
http://www.bkgm.com/rgb/rgb.cgi?view+179
_______________________________________________
Daniel Murphy http://www.cityraccoon.com
Humlebæk BG Klub http://www.hbgk.dk
Raccoon on FIBS http://www.fibs.com
mcc...@northcoast.com
see the look on Kevin's face after Kit made that first move--the chuckles I
have remembering it I consider dividends on an investment :o)
Kate
http://www.northcoast.com/~mccool/bg.html
In article <01be589e$22bde560$b1c86420@kbastian>,
"Kevin Bastian" <kba...@ibm.net> wrote:
(snip)
> (Single elimination. Quickest hundred bucks I ever lost since I had my
> wallet stolen in college!)
>
> Anyway, at one point, Kit and Sally rolled, and I don't remember what the
> roll was, but Kit moved it differently than he had recommended in his post!
> I said, "Kit, I thought you recommended so-and-so in your newsgroup posting
> last year." He wryly said "I don't remember ever recommending any such
> thing." (Kate and Sally were chuckling at this point.)
(snip)
VSG
Thanks for that JP :)
6-4 and 6-2 were interesting.. Jelly steering no doubt :)
Tim Goodwin
wrote:
>Bob Stringer wrote:
>
>> JP White wrote:
>>
>> > I suggest you visit http://www.northcoast.com/~mccool/kwopnrls.htm
Just make it .html instead of .htm and you should find the file.
Tim
Chuck Bower
Alan Webb <we...@hannover.sgh-net.de> wrote:
>Has anyone done extensive rollouts on opening rolls and their combinations?
>
>I mention this as in Magriel's book "Backgammon", Magriel recommends
>bringing 2 builders down with the opening roll 5-3 (13/8 13/10), rather than
>the more common (and my current) move of making the 3 point. I understand
>both the advantages and disadvantages of both these moves however I am
>interested in what extensive rollouts have to say.
>
>I still find myself varying the 4-3 opening roll for example. Everybody
>seems to have their own preference and normally sound reasons to back them
>up, but there must be some hard concrete evidence to show that play X is
>better than play Y. I can imagine that the sample would have to be huge as
>it is only the opening roll. One could also question as to whether rollouts
>based on an opening roll could tell you anything of worth in any case.
I have done several million rollouts of REPLIES to opening rolls using
JFv3.0. It is possible to combine these to give the best plays for
the OPENING rolls, and CJC put the results in Flint Area Backgammon News
last year (or maybe it was in 1997; time flies). There are some things one
must keep in mind when considering this data, however.
1) Jellyfish is playing CUBELESS money games.
2) Jellyfish is playing against itself.
3) Many of the rollout results are so close that both statistical and
systematic uncertainties swamp the outcome, even with as many rollouts
as I did.
I also used these data in an Inside Backgammon article ("Post Crawford
Plays and Cubes") which was published last summer. There I showed how
the different post Crawford match scores affect the choice of opening plays.
(But don't forget the above three warnings!)
As far as the above mentioned rolls, 53 played 13/8, 13/10 is inferior.
The three point is quite valuable. Make it. 43 is one of the closest
choices with 13/9, 13/10; 24/20, 13/10; and 24/21, 13/9 all being about
equal in the rollouts. 24/20, 24/21 is at least as good as the single
split plays when leading a match and being unable to use gammons when playing
against an opponent who CAN profit from a gammon. (e.g. 1-away, 2-away
Crawford.)
One last comment. The candidates are so close that if your opponent is
unfamiliar with defending a certain play then you are better off choosing that
option, even if the rollouts indicate it is inferior. This can make plays
like the just mentioned 24/20, 24/21 with 43 (even at money play) and making
the 2-point with 64 tip the scales in your favor.
Chuck
bo...@bigbang.astro.indiana.edu
c_ray on FIBS
David desJardins
> Thus a difference of 2 X 0.0023 = 0.0046 is a two standard deviation
> result, meaning that there is about a 2.3% chance that an infinite
> (replace with a very large number if you don't like that word ;)
> number of trials with the same robot would reverse the order of the
> two plays.
No, unfortunately it doesn't mean that. What it does mean is, "The
probability that a play which would rank lower in an infinite
setquence of rollouts, would rank this much higher in this finite trial,
is no more than 2.3%."
You have to make some set of additional assumptions to estimate the
probability that the lower-ranked play actually would rank higher in an
infinite sequence of rollouts. I wouldn't mention, except that there's
been so much debate about statistical hypothesis testing lately.
Note that with what I personally consider plausible assumptions, you
would get a figure *lower* than 2.3%. So I'm not complaining that your
data aren't sufficient; I'm just saying that your language is not precise.
David desJardins
Chuck Bower
>Alan Webb <we...@hannover.sgh-net.de> wrote:
>
>>Has anyone done extensive rollouts on opening rolls and their combinations?
>>
>>I mention this as in Magriel's book "Backgammon", Magriel recommends
>>bringing 2 builders down with the opening roll 5-3 (13/8 13/10), rather than
>>the more common (and my current) move of making the 3 point. I understand
>>both the advantages and disadvantages of both these moves however I am
>>interested in what extensive rollouts have to say.
>>
>>I still find myself varying the 4-3 opening roll for example. Everybody
>>seems to have their own preference and normally sound reasons to back them
>>up, but there must be some hard concrete evidence to show that play X is
>>better than play Y. I can imagine that the sample would have to be huge as
>>it is only the opening roll. One could also question as to whether rollouts
>>based on an opening roll could tell you anything of worth in any case.
Here are the data from my rollouts. This is more or less what was
pubbed in FLINT AREA BACKGAMMON NEWS, Holiday (Dec) 1997 issue, with some
additions and corrections in the Jan/Feb 1998 issue. These data are based
upon 800,000+ JFv3.0 level-6 rollouts of replies to opening rolls. I then
combined the 'correct' replies to come up with these opening roll cubeless
equities:
21
11,23 -0.006
11,5 -0.011
31
5(2) +0.160
41
9,23 -0.001
9,5 -0.027
51
8,23 +0.005
8,5 -0.012
61
7(2) +0.102
32
21,11 +0.003
10,11 -0.004
42
4(2) +0.118
52
8,22 +0.008
8,11 -0.002
62
18,11 +0.013
16 -0.011
43
8,9 +0.009
9,21 +0.007
20,10 +0.006
20,21 -0.006
53
3(2) +0.062
63
18,10 +0.004
15 -0.003
54
8,20 +0.022
8,9 +0.012
64
2(2) +0.010
14 +0.007
18,9 +0.004
65
13 +0.068
Statistical uncertainty (tandard deviations) are estimated at
0.0016 for a given rollout and 0.0023 for comparing two rollouts.
Thus a difference of 2 X 0.0023 = 0.0046 is a two standard deviation
result, meaning that there is about a 2.3% chance that an infinite
(replace with a very large number if you don't like that word ;)
number of trials with the same robot would reverse the order of the
two plays. (NOTE: 2.3% is the area under the standard normal curve
above an abscissa of 2. It is merely a coincidence that a single
standard deviaion was 0.0023--2.3% divided by 10.)
If there are someone's "favorite" candidates missing, that is
partially personal bias, but it is based on previous rollouts (I've
been doing opening rolls and replies to opening rolls for ~6 years)
using previous robots (two versions of Expert Backgammon and the
earlier two versions of Jellyfish) which indicated rather significant
inferiority. You can, for example use the above data to estimate
other openings. E.g. 53 played 13/8, 13/10 should be reasonably
close to 52 played 13/8, 13/11. Slotting the 5-point with 62 is
not hugely different from playing 51 13/8, 6/5.
Finally, I guess I could have kept these data secret in the hopes
of publishing a book and making lots of money. I long ago decided
against that, because "life is too short". (Or maybe it's because
"there aren't enough paying customers." ;)
Chuck Bower
Chuck Bower wrote:
This is wrong. Some of the rollouts were for "inferior" replies so
thosedon't count. The standard deviation for a single rollout is more like 0.0022
and joint standard deviation (for comparing two plays) 0.0031. Those are
the numbers I claimed in the FABN article. I did things right, then (I think).
> Thus a difference of 2 X 0.0023 = 0.0046 is a two standard deviation
make that 2 X 0.0031 = 0.006
> result, meaning that there is about a 2.3% chance that an infinite
> (replace with a very large number if you don't like that word ;)
> number of trials with the same robot would reverse the order of the
> two plays. (NOTE: 2.3% is the area under the standard normal curve
> above an abscissa of 2. It is merely a coincidence that a single
> standard deviaion was 0.0023--2.3% divided by 10.)
>
> If there are someone's "favorite" candidates missing, that is
> partially personal bias, but it is based on previous rollouts (I've
> been doing opening rolls and replies to opening rolls for ~6 years)
> using previous robots (two versions of Expert Backgammon and the
> earlier two versions of Jellyfish) which indicated rather significant
> inferiority. You can, for example use the above data to estimate
> other openings. E.g. 53 played 13/8, 13/10 should be reasonably
> close to 52 played 13/8, 13/11. Slotting the 5-point with 62 is
> not hugely different from playing 51 13/8, 6/5.
Chuck
bo...@bigbang.astro.indiana.edu
c_ray on FIBS
Chuck Bower
David desJardins wrote:
> Chuck Bower <bo...@bigbang.astro.indiana.edu> writes:
> > Thus a difference of 2 X 0.0023 = 0.0046 is a two standard deviation
> > result, meaning that there is about a 2.3% chance that an infinite
> > (replace with a very large number if you don't like that word ;)
> > number of trials with the same robot would reverse the order of the
> > two plays.
>
> No, unfortunately it doesn't mean that. What it does mean is, "The
> probability that a play which would rank lower in an infinite
> setquence of rollouts, would rank this much higher in this finite trial,
> is no more than 2.3%."
>
> You have to make some set of additional assumptions to estimate the
> probability that the lower-ranked play actually would rank higher in an
> infinite sequence of rollouts. I wouldn't mention, except that there's
> been so much debate about statistical hypothesis testing lately.
>
> Note that with what I personally consider plausible assumptions, you
> would get a figure *lower* than 2.3%. So I'm not complaining that your
> data aren't sufficient; I'm just saying that your language is not precise.
Well, I always like to err on the side of stating something too weakly
as oppose to too strongly! I personally find some of these statistics
debates hard to follow, which I associate with my poor knowledge of
the subject (and not to the validity of the arguments nor the skill of
the debaters in explaining).
I have contended for a long time that probability and statistics are
way undertaught, at least in this country. High schools teach calculus
to their upper level students and ignore P&S. In most cases, a person
is unlikely to ever use calculus once leaving school. P&S
come up every day for everyone (including non-BG players)!
Even in college, the subject tends to good a bad shake. Basically
there are two kinds of P&S classes: those required for non-math
majors (of which 95% of the students couldn't care less and are only
there because they are being forced) and those for mathematicians
who really enjoy epsilon-delta arguments. In one case you are looking
at the moon through a 4-inch (10 cm) telescope. In the other you are
studying a moon rock with an electron microscope. For those of us
who would like something in between, we're left holding the (marble) bag.
Steve Harris
bo...@bigbang.astro.indiana.edu says...
>43
> 8,9 +0.009
I presume here you meant 9,10, right?
I'm not trying to be nitpicky. I appreciate your posting. Just thought that, if
you're like me, you would want any errors (even slight ones) corrected.
Regards,
Steve
JP White
One question though. The rollouts include doubles as the first players
play. I thought that doubles were disallowed as the first throw of a
game.
Thanks for posting them. I'll digest the rollouts once I've found enough
paper to print them on!
VSG
>21
> 11,23 -0.006
> 11,5 -0.011
Thanks Chuck!
Thats pretty much what I've been looking for. I shall bear the results in mind
and weigh them up with my own and opponents strengths, score etc.
Notation threw me out a bit though.. I wondered how with a 21 you could move
from the 11 to 23 point :).
MAXINE W MCLEOD
could follow easily? I do not mean to sound ignorant but, well maybe I just
am.
thanks
magnolia
Kevin Bastian
I'm convinced both are true!
Kevin
Chuck Bower <bo...@bigbang.astro.indiana.edu> wrote in article
<7al3lh$lsp$1...@jetsam.uits.indiana.edu>...
>
<snip>
"life is too short". (Or maybe it's because
> "there aren't enough paying customers." ;)
>
>