• A good and interesting question about XG's evaluation

    From peps...@gmail.com@pepstein5@gmail.com to rec.games.backgammon on Fri Dec 29 10:18:12 2023
    From Newsgroup: rec.games.backgammon

    2A 2A play has been discussed ad nauseam and it would be tedious to
    rehash the details.
    Since, barring any checker blunders, doubling at the first opportunity is
    a version of optimal play, I was slightly surprised at XG's hold here.
    I would not be surprised if the double was optional but I thought that the overriding goal of XG is to play as accurately as possible so, for the sake
    of coding simplicity, I was expecting a double here.
    However, what I suspect is actually a bug is XG's determination that doubling is actually wrong. Surely doubling is fine even if holding is just as good? Any comments? I suppose occasionally XG will do something that a human
    will see as clearly non-optimal (or at least
    questionable). One of XG's evaluations is clearly slightly wrong but that might just be the price of the coding design.
    It reminds me of a position where XG had all its checkers and I had a single checker. My checker was on my deuce point, XG held my ace point and XG
    was on roll. If XG can't hit me and can't run off the backgammon then any player who isn't a rank beginner knows to hold the ace point hoping for the one percent parlay where I roll 11 and the opponent hits. 1% really isn't so small. But XG just ran off my acepoint with one checker and conceded the backgammon.
    This evaluation seems kind of like that.
    Having said all this, I now clearly remember that at 2A, 2A with my 52,
    holding is actually fine because XG's best sequence is 55, dance, which
    is a clear take.
    However, the determination that doubling is wrong seems suspect?
    Paul
    XGID=-b----E-D---eD---c-e--A-A-:0:0:-1:00:9:9:0:11:10
    X:XG Roller+ O:Daniel
    Score is X:9 O:9 11 pt.(s) match.
    +13-14-15-16-17-18------19-20-21-22-23-24-+
    | X O | | O X |
    | X O | | O X |
    | X O | | O |
    | X O | | O |
    | X | | O |
    | |BAR| |
    | | | X |
    | O | | X |
    | O X | | X |
    | O X | | X |
    | O X | | X O O |
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    Pip count X: 167 O: 160 X-O: 9-9/11
    Cube: 1
    X on roll, cube action
    Analyzed in XG Roller+
    Player Winning Chances: 49.72% (G:14.79% B:1.11%)
    Opponent Winning Chances: 50.28% (G:15.26% B:1.91%)
    Cubeless Equities: No Double=-0.014, Double=-0.022
    Cubeful Equities:
    No double: -0.015
    Double/Take: -0.022 (-0.007)
    Double/Pass: +1.000 (+1.015)
    Best Cube action: No double / Take
    Percentage of wrong pass needed to make the double decision right: 0.7%
    eXtreme Gammon Version: 2.10, MET: Kazaross XG2
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From Timothy Chow@tchow12000@yahoo.com to rec.games.backgammon on Fri Dec 29 23:57:27 2023
    From Newsgroup: rec.games.backgammon

    On 12/29/2023 1:18 PM, peps...@gmail.com wrote:
    I would not be surprised if the double was optional but I thought that the overriding goal of XG is to play as accurately as possible so, for the sake of coding simplicity, I was expecting a double here.

    That is certainly not the overriding goal. If it were, then XG would use
    some kind of massive bearoff database for bearing off, which it doesn't.
    As for coding simplicity, adding a special feature to make it double at
    the first opportunity at 2a2a is *less* simple than just letting it do
    its normal thing, and there's no reason to expect that letting it do its
    normal thing will cause it to double at the first opportunity at 2a2a.

    ---
    Tim Chow
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From Philippe Michel@philippe.michel7@free.fr.invalid to rec.games.backgammon on Sat Dec 30 07:10:39 2023
    From Newsgroup: rec.games.backgammon

    On 2023-12-29, peps...@gmail.com <pepstein5@gmail.com> wrote:

    I would not be surprised if the double was optional but I thought that the overriding goal of XG is to play as accurately as possible so, for the sake of coding simplicity, I was expecting a double here.
    However, what I suspect is actually a bug is XG's determination that doubling is actually wrong. Surely doubling is fine even if holding is just as good?

    Analyzed in XG Roller+

    Maybe XG Roller+, a short truncated rollout with relatively low level evaluation of the leaf nodes, messes things up.

    I would expect regular evaluations at 3-ply or more to find this is an optional double / take.
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From Timothy Chow@tchow12000@yahoo.com to rec.games.backgammon on Sat Dec 30 09:44:05 2023
    From Newsgroup: rec.games.backgammon

    On 12/30/2023 2:10 AM, Philippe Michel wrote:
    I would expect regular evaluations at 3-ply or more to find this is an optional double / take.

    XGID=-b----E-D---eD---c-e--A-A-:0:0:-1:00:9:9:0:11:10

    Score is X:9 O:9 11 pt.(s) match.
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    | X O | | O X X |
    | X O | | O |
    | X O | | O |
    | X | | O |
    | | | O |
    | |BAR| |
    | O | | X |
    | O X | | X |
    | O X | | X |
    | O X | | X O |
    | O X | | X O |
    +13-14-15-16-17-18------19-20-21-22-23-24-+
    Pip count X: 160 O: 167 X-O: 9-9/11
    Cube: 1
    O on roll, cube action

    Analyzed in 3-ply
    Player Winning Chances: 49.96% (G:14.31% B:1.17%)
    Opponent Winning Chances: 50.04% (G:15.37% B:1.82%)

    Cubeless Equities: No Double=-0.020, Double=-0.014

    Cubeful Equities:
    No double: -0.002
    Double/Take: -0.014 (-0.012)
    Double/Pass: +1.000 (+1.002)

    Best Cube action: No double / Take
    Percentage of wrong pass needed to make the double decision right: 1.2%

    eXtreme Gammon Version: 2.19.211.pre-release, MET: Kazaross XG2

    ------------------------------------------------------------------------

    XGID=-b----E-D---eD---c-e--A-A-:0:0:-1:00:9:9:0:11:10

    Score is X:9 O:9 11 pt.(s) match.
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    | X O | | O X X |
    | X O | | O |
    | X O | | O |
    | X | | O |
    | | | O |
    | |BAR| |
    | O | | X |
    | O X | | X |
    | O X | | X |
    | O X | | X O |
    | O X | | X O |
    +13-14-15-16-17-18------19-20-21-22-23-24-+
    Pip count X: 160 O: 167 X-O: 9-9/11
    Cube: 1
    O on roll, cube action

    Analyzed in 4-ply
    Player Winning Chances: 49.96% (G:14.31% B:1.17%)
    Opponent Winning Chances: 50.04% (G:15.37% B:1.82%)

    Cubeless Equities: No Double=-0.020, Double=-0.004

    Cubeful Equities:
    No double: -0.002
    Double/Take: -0.004 (-0.002)
    Double/Pass: +1.000 (+1.002)

    Best Cube action: No double / Take
    Percentage of wrong pass needed to make the double decision right: 0.2%

    eXtreme Gammon Version: 2.19.211.pre-release, MET: Kazaross XG2

    ---
    Tim Chow
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From peps...@gmail.com@pepstein5@gmail.com to rec.games.backgammon on Sat Dec 30 11:56:19 2023
    From Newsgroup: rec.games.backgammon

    On Saturday, December 30, 2023 at 4:57:30rC>AM UTC, Timothy Chow wrote:
    ...
    As for coding simplicity, adding a special feature to make it double at
    the first opportunity at 2a2a is *less* simple than just letting it do
    its normal thing, and there's no reason to expect that letting it do its normal thing will cause it to double at the first opportunity at 2a2a.
    I assumed (with no justification whatsoever) that XG had some sort
    of special feature whereby it normally doubled as soon as legal at 2A 2A.
    If it did have some such feature (it probably doesn't) then code saying: [Always use that feature] would be simpler than code saying: [Use that
    feature unless...]
    Assuming you're an expert on how XG works, where did you gain that
    knowledge? Is it from public sources or personal correspondence
    with Xavier?
    Paul
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From Timothy Chow@tchow12000@yahoo.com to rec.games.backgammon on Sat Dec 30 23:50:42 2023
    From Newsgroup: rec.games.backgammon

    On 12/30/2023 2:56 PM, peps...@gmail.com wrote:
    On Saturday, December 30, 2023 at 4:57:30rC>AM UTC, Timothy Chow wrote:
    ...
    As for coding simplicity, adding a special feature to make it double at
    the first opportunity at 2a2a is *less* simple than just letting it do
    its normal thing, and there's no reason to expect that letting it do its
    normal thing will cause it to double at the first opportunity at 2a2a.

    I assumed (with no justification whatsoever) that XG had some sort
    of special feature whereby it normally doubled as soon as legal at 2A 2A.
    If it did have some such feature (it probably doesn't) then code saying: [Always use that feature] would be simpler than code saying: [Use that feature unless...]

    Assuming you're an expert on how XG works, where did you gain that
    knowledge? Is it from public sources or personal correspondence
    with Xavier?

    I'm not an expert on how XG works, but this whole thread is presenting
    evidence that XG has no such special feature. Since there is no evidence
    that XG does have such a special feature, the preponderance of the
    evidence suggests that no such special feature exists.

    ---
    Tim Chow

    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From Stick Rice@bananaboater315@gmail.com to rec.games.backgammon on Mon Jan 1 11:49:01 2024
    From Newsgroup: rec.games.backgammon

    On Friday, December 29, 2023 at 1:18:14rC>PM UTC-5, peps...@gmail.com wrote:
    2A 2A play has been discussed ad nauseam and it would be tedious to
    rehash the details.
    Since, barring any checker blunders, doubling at the first opportunity is
    a version of optimal play, I was slightly surprised at XG's hold here.
    I would not be surprised if the double was optional but I thought that the overriding goal of XG is to play as accurately as possible so, for the sake of coding simplicity, I was expecting a double here.
    However, what I suspect is actually a bug is XG's determination that doubling
    is actually wrong. Surely doubling is fine even if holding is just as good? Any comments? I suppose occasionally XG will do something that a human
    will see as clearly non-optimal (or at least
    questionable). One of XG's evaluations is clearly slightly wrong but that might just be the price of the coding design.

    It reminds me of a position where XG had all its checkers and I had a single checker. My checker was on my deuce point, XG held my ace point and XG
    was on roll. If XG can't hit me and can't run off the backgammon then any player who isn't a rank beginner knows to hold the ace point hoping for the one percent parlay where I roll 11 and the opponent hits. 1% really isn't so small. But XG just ran off my acepoint with one checker and conceded the backgammon.
    This evaluation seems kind of like that.

    Having said all this, I now clearly remember that at 2A, 2A with my 52, holding is actually fine because XG's best sequence is 55, dance, which
    is a clear take.
    However, the determination that doubling is wrong seems suspect?

    Paul

    XGID=-b----E-D---eD---c-e--A-A-:0:0:-1:00:9:9:0:11:10

    X:XG Roller+ O:Daniel
    Score is X:9 O:9 11 pt.(s) match. +13-14-15-16-17-18------19-20-21-22-23-24-+
    | X O | | O X |
    | X O | | O X |
    | X O | | O |
    | X O | | O |
    | X | | O |
    | |BAR| |
    | | | X |
    | O | | X |
    | O X | | X |
    | O X | | X |
    | O X | | X O O |
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    Pip count X: 167 O: 160 X-O: 9-9/11
    Cube: 1
    X on roll, cube action

    Analyzed in XG Roller+
    Player Winning Chances: 49.72% (G:14.79% B:1.11%)
    Opponent Winning Chances: 50.28% (G:15.26% B:1.91%)

    Cubeless Equities: No Double=-0.014, Double=-0.022

    Cubeful Equities:
    No double: -0.015
    Double/Take: -0.022 (-0.007)
    Double/Pass: +1.000 (+1.015)

    Best Cube action: No double / Take
    Percentage of wrong pass needed to make the double decision right: 0.7%

    eXtreme Gammon Version: 2.10, MET: Kazaross XG2
    The bot shouldn't double at 2a 2a unless it sees at least one market losing sequence. Are there any MLS here?
    Stick
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From peps...@gmail.com@pepstein5@gmail.com to rec.games.backgammon on Mon Jan 1 13:19:25 2024
    From Newsgroup: rec.games.backgammon

    On Monday, January 1, 2024 at 7:49:02rC>PM UTC, Stick Rice wrote:
    On Friday, December 29, 2023 at 1:18:14rC>PM UTC-5, peps...@gmail.com wrote:
    2A 2A play has been discussed ad nauseam and it would be tedious to
    rehash the details.
    Since, barring any checker blunders, doubling at the first opportunity is a version of optimal play, I was slightly surprised at XG's hold here.
    I would not be surprised if the double was optional but I thought that the overriding goal of XG is to play as accurately as possible so, for the sake
    of coding simplicity, I was expecting a double here.
    However, what I suspect is actually a bug is XG's determination that doubling
    is actually wrong. Surely doubling is fine even if holding is just as good?
    Any comments? I suppose occasionally XG will do something that a human will see as clearly non-optimal (or at least
    questionable). One of XG's evaluations is clearly slightly wrong but that might just be the price of the coding design.

    It reminds me of a position where XG had all its checkers and I had a single
    checker. My checker was on my deuce point, XG held my ace point and XG
    was on roll. If XG can't hit me and can't run off the backgammon then any player who isn't a rank beginner knows to hold the ace point hoping for the one percent parlay where I roll 11 and the opponent hits. 1% really isn't so small. But XG just ran off my acepoint with one checker and conceded the backgammon.
    This evaluation seems kind of like that.

    Having said all this, I now clearly remember that at 2A, 2A with my 52, holding is actually fine because XG's best sequence is 55, dance, which
    is a clear take.
    However, the determination that doubling is wrong seems suspect?

    Paul

    XGID=-b----E-D---eD---c-e--A-A-:0:0:-1:00:9:9:0:11:10

    X:XG Roller+ O:Daniel
    Score is X:9 O:9 11 pt.(s) match. +13-14-15-16-17-18------19-20-21-22-23-24-+
    | X O | | O X |
    | X O | | O X |
    | X O | | O |
    | X O | | O |
    | X | | O |
    | |BAR| |
    | | | X |
    | O | | X |
    | O X | | X |
    | O X | | X |
    | O X | | X O O |
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    Pip count X: 167 O: 160 X-O: 9-9/11
    Cube: 1
    X on roll, cube action

    Analyzed in XG Roller+
    Player Winning Chances: 49.72% (G:14.79% B:1.11%)
    Opponent Winning Chances: 50.28% (G:15.26% B:1.91%)

    Cubeless Equities: No Double=-0.014, Double=-0.022

    Cubeful Equities:
    No double: -0.015
    Double/Take: -0.022 (-0.007)
    Double/Pass: +1.000 (+1.015)

    Best Cube action: No double / Take
    Percentage of wrong pass needed to make the double decision right: 0.7%

    eXtreme Gammon Version: 2.10, MET: Kazaross XG2
    The bot shouldn't double at 2a 2a unless it sees at least one market losing sequence. Are there any MLS here?
    No, there aren't. I learned that quite early on FIBS in the nineties.
    I criticised my opponent for not doubling immediately and he replied that doubling is unnecessary because 55/dance
    is still a take because gammons aren't in play.
    Paul
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From Timothy Chow@tchow12000@yahoo.com to rec.games.backgammon on Mon Jan 1 17:45:37 2024
    From Newsgroup: rec.games.backgammon

    On 1/1/2024 2:49 PM, Stick Rice wrote:
    The bot shouldn't double at 2a 2a unless it sees at least one market losing sequence. Are there any MLS here?

    I'm not sure exactly what you're saying here, but just to be clear,
    bots don't explicitly consider market-losing sequences when deciding
    whether to double.

    https://www.bgonline.org/forums/webbbs_config.pl?read=57030

    ---
    Tim Chow
    --- Synchronet 3.21b-Linux NewsLink 1.2
  • From Stick Rice@bananaboater315@gmail.com to rec.games.backgammon on Mon Jan 1 17:56:14 2024
    From Newsgroup: rec.games.backgammon

    On Monday, January 1, 2024 at 5:45:40rC>PM UTC-5, Timothy Chow wrote:
    On 1/1/2024 2:49 PM, Stick Rice wrote:
    The bot shouldn't double at 2a 2a unless it sees at least one market losing sequence. Are there any MLS here?
    I'm not sure exactly what you're saying here, but just to be clear,
    bots don't explicitly consider market-losing sequences when deciding
    whether to double.

    https://www.bgonline.org/forums/webbbs_config.pl?read=57030

    ---
    Tim Chow
    They don't, but they do.
    Stick
    --- Synchronet 3.21b-Linux NewsLink 1.2