Too many players consider XG/other-bots to be almost infallible. bots are much stronger than humans, but how much Elo they shed to a perfect player we will never know.
-Joseph On Tue, 3 Nov 2020 at 06:40, pierre zakia <[email protected]> wrote: > Thanks Tim and Joseph for your explanations > > I would have expected that at 4 ply, the tricky intrication between > gammons percentage, cube access and score will be solved. > > But I never suspected that software is faulty, just trying to understand > the quantum leap from 2ply to 3 ply to 4ply. > > Pierre > > > ------------------------------------------------------------------------------------------------ > > Le lundi 2 novembre 2020 à 18:00:12 UTC+1, [email protected] < > [email protected]> a écrit : > > > Send Bug-gnubg mailing list submissions to > [email protected] > > To subscribe or unsubscribe via the World Wide Web, visit > https://lists.gnu.org/mailman/listinfo/bug-gnubg > or, via email, send a message with subject or body 'help' to > [email protected] > > You can reach the person managing the list at > [email protected] > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Bug-gnubg digest..." > > > Today's Topics: > > 1. Re: Why such discrepancy for the ranked (obviously wrong) 1st > choice between 4 ply and simulation ? (Timothy Y. Chow) > 2. Re: Why such discrepancy for the ranked (obviously wrong) 1st > choice between 4 ply and simulation ? (Joseph Heled) > 3. Re: Why such discrepancy for the ranked (obviously wrong) 1st > choice between 4 ply and simulation ? (pierre zakia) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Sun, 1 Nov 2020 15:14:08 -0500 (EST) > From: "Timothy Y. Chow" <[email protected]> > To: [email protected] > Subject: Re: Why such discrepancy for the ranked (obviously wrong) 1st > choice between 4 ply and simulation ? > Message-ID: <[email protected]> > Content-Type: text/plain; charset="iso-8859-15"; Format="flowed" > > Pierre Zakia wrote: > > > In the position here below, I am curious to understand what in Gnubg > > engine yields such discrepancy between 4 ply and roll out. > > > > > > GNU Backgammon ID de position: G27HAAiY2+ABAw > > ID de match : MAHyAAAAAAAE > > Thanks for posting this position---it's very interesting! > > First let me mention that eXtreme Gammon (XG) finds this position tricky > as well. XG 3-ply plays 13/9(2) 13/5. (XG and GNU use slightly > definitions of "ply" I think, but that's not important here.) So I don't > think that 13/9(2) 13/5 is "obviously wrong." Two things to note: > > 1. 13/9(2) 13/5 wins more gammons than 20/16(2) 13/9(2), so it's probably > the better play at certain match scores. > > 2. Suppose you change the position slightly---instead of having the cube > centered, have X own the cube on 2. I think that if you roll it out then > you'll see that 20/16(2) 13/9(2) comes out ahead by a much smaller margin. > So part of the reason that 20/16(2) 13/9(2) comes out so far ahead in your > original position is the somewhat subtle fact that O is hoping to double > next turn---but if O gets hit then she won't be able to double. So with > the cube in the middle, O is much more cautious about leaving shots. > > Anyway, I don't think that this is a fault of the software. It's just > that the position is more subtle than you may have thought at first > glance. > > Tim > > ------------------------------ > > Message: 2 > Date: Mon, 2 Nov 2020 09:58:42 +1300 > From: Joseph Heled <[email protected]> > To: pierre zakia <[email protected]> > Cc: "[email protected]" <[email protected]> > Subject: Re: Why such discrepancy for the ranked (obviously wrong) 1st > choice between 4 ply and simulation ? > Message-ID: > <CAG8x8-0D6dimLHYmfX2s8HbG=icytrn2ahgnf0bce-5xujy...@mail.gmail.com> > Content-Type: text/plain; charset="utf-8" > > You can probably write a 5kg book about positions where a bot plays > different moves at 0-ply and 1-ply. No bot is good at all types of > positions.And remember, to get the higher plies the bot need to "play" the > position to this depth. With a tricky position many moves might be > inaccurate, so the higher ply is not always correct as well. > Same for rollouts. > > -Joseph > > On Mon, 2 Nov 2020 at 04:02, pierre zakia <[email protected]> wrote: > > > Hi everybody, > > > > > > In the position here below, I am curious to understand what in Gnubg > > engine yields such discrepancy between 4 ply and roll out. > > > > > > GNU Backgammon ID de position: G27HAAiY2+ABAw > > > > ID de match : MAHyAAAAAAAE > > > > +12-11-10--9--8--7-------6--5--4--3--2--1-+ O: gnubg > > > > | O O | | O O X | 0 point > > > > | O O | | O O | Dés jetés 44 > > > > | | | O | > > > > | | | | > > > > | | | | > > > > ^| |BAR| | Match en 7 points (Videau > > : 1) > > > > | | | | > > > > | O | | | > > > > | O X | | X | > > > > | O X X X | | X O X X | > > > > | O X X X | | X O X X | 0 point > > > > +13-14-15-16-17-18------19-20-21-22-23-24-+ X: pierrez > > > > Pip counts : O 148, X 106 > > > > > > *4 ply results *: > > > > *13/5 13/9(2) *+0,579 > > > > 0,684 0,135 0,005 - 0,316 0,098 0,002 > > > > *20/16(2) 13/9(2) *+0,570 (* -0,008*) > > > > 0,675 0,083 0,003 - 0,325 0,082 0,001 > > > > *13/5(2) *+0,436 ( -0,143) > > > > 0,611 0,122 0,003 - 0,389 0,065 0,001</td> > > > > > > *Roll out :* > > 1. Roll out *20/16(2) 13/9(2)* Eq.: +0,783 > > > > 0,709 0,078 0,003 - 0,291 0,089 0,000 CL +0,415 CF +0,783 > > > > [0,011 0,006 0,000 - 0,011 0,012 0,012 CL 0,026 CF 0,053] > > 2. Roll out *13/5 13/9(2) * Eq.: +0,576 (* -0,207*) > > > > 0,675 0,138 0,004 - 0,325 0,101 0,002 CL +0,389 CF +0,576 > > > > [0,002 0,003 0,000 - 0,002 0,002 0,000 CL 0,004 CF 0,007] > > 3. Roll out *13/5(2) * Eq.: +0,549 ( -0,233) > > > > 0,634 0,105 0,003 - 0,366 0,044 0,056 CL +0,338 CF +0,549 > > > > [0,008 0,009 0,001 - 0,008 0,007 0,055 CL 0,020 CF 0,049] > > > > > > I am more interested in the relative equity (bold font) between the first > > 2 moves than in the respective absolute equity of the same moves. > > > > > > Thanks in advance to shed some light on this. > > > > > > Pierre > > > > > -------------- next part -------------- > An HTML attachment was scrubbed... > URL: < > https://lists.gnu.org/archive/html/bug-gnubg/attachments/20201102/444af80c/attachment.html > > > > ------------------------------ > > Message: 3 > Date: Mon, 2 Nov 2020 14:04:09 +0000 (UTC) > From: pierre zakia <[email protected]> > To: Joseph Heled <[email protected]> > Cc: "[email protected]" <[email protected]> > Subject: Re: Why such discrepancy for the ranked (obviously wrong) 1st > choice between 4 ply and simulation ? > Message-ID: <[email protected]> > Content-Type: text/plain; charset="utf-8" > > I am aware of what you wrote, but the position doesn't look so tricky to > get fooled at 4 ply (compare to roll home a massive backgame for instance), > hence my question. > Pierre > ------------------------------- > Le dimanche 1 novembre 2020 à 21:58:55 UTC+1, Joseph Heled < > [email protected]> a écrit : > > You can probably write a 5kg book about positions where a bot plays > different moves at 0-ply and 1-ply. No bot is good at all types of > positions.And remember, to get the higher plies the bot need to "play" the > position to this depth. With a tricky position many moves might be > inaccurate, so the higher ply is not always correct as well.Same for > rollouts. > > -Joseph > > On Mon, 2 Nov 2020 at 04:02, pierre zakia <[email protected]> wrote: > > > Hi everybody, > > > > > In the position here below, I am curious to understand what in Gnubg > engine yields such discrepancy between 4 ply and roll out. > > > > > GNU Backgammon ID de position: G27HAAiY2+ABAw > > ID de match : MAHyAAAAAAAE > > +12-11-10--9--8--7-------6--5--4--3--2--1-+ O: gnubg > > | O O | | O O X | 0 point > > | O O | | O O | Dés jetés 44 > > | | | O | > > | | | | > > | | | | > > ^| |BAR| | Match en 7 points (Videau > : 1) > > | | | | > > | O | | | > > | O X | | X | > > | O X X X | | X O X X | > > | O X X X | | X O X X | 0 point > > +13-14-15-16-17-18------19-20-21-22-23-24-+ X: pierrez > > Pip counts : O 148, X 106 > > > > > 4 ply results : > > 13/5 13/9(2) +0,579 > > 0,684 0,135 0,005 - 0,316 0,098 0,002 > > 20/16(2) 13/9(2) +0,570 ( -0,008) > > 0,675 0,083 0,003 - 0,325 0,082 0,001 > > 13/5(2) +0,436 ( -0,143) > > 0,611 0,122 0,003 - 0,389 0,065 0,001</td> > > > > Roll out :1. Roll out 20/16(2) 13/9(2) Eq.: +0,783 > 0,709 0,078 0,003 - 0,291 0,089 0,000 CL +0,415 CF +0,783 > > [0,011 0,006 0,000 - 0,011 0,012 0,012 CL 0,026 CF 0,053] > 2. Roll out 13/5 13/9(2) Eq.: +0,576 ( -0,207) > 0,675 0,138 0,004 - 0,325 0,101 0,002 CL +0,389 CF +0,576 > > [0,002 0,003 0,000 - 0,002 0,002 0,000 CL 0,004 CF 0,007] > 3. Roll out 13/5(2) Eq.: +0,549 ( -0,233) > 0,634 0,105 0,003 - 0,366 0,044 0,056 CL +0,338 CF +0,549 > > [0,008 0,009 0,001 - 0,008 0,007 0,055 CL 0,020 CF 0,049] > > > > > I am more interested in the relative equity (bold font) between the first > 2 moves than in the respective absolute equity of the same moves. > > > > > Thanks in advance to shed some light on this. > > > > > Pierre > > > > -------------- next part -------------- > An HTML attachment was scrubbed... > URL: < > https://lists.gnu.org/archive/html/bug-gnubg/attachments/20201102/65b5698e/attachment.html > > > > ------------------------------ > > Subject: Digest Footer > > _______________________________________________ > Bug-gnubg mailing list > [email protected] > https://lists.gnu.org/mailman/listinfo/bug-gnubg > > > ------------------------------ > > End of Bug-gnubg Digest, Vol 210, Issue 2 > ***************************************** >
