1. Forum
  2. Usenet
  3. REC.GAMES.BACKGAMMON

  • A marginal race

    From pepstein5@gmail.com@21:1/5 to All on Tue Sep 13 13:39:33 2022
    Obviously, this should be treated as a race but I've learned not to Axelise these positions. The Axelisation algo has been trained only on pure
    races not near-races, so I don't think the Axelisations are reliable here.
    It turns out to be extremely close exactly marginal. So far, XG agrees
    with my take, but that can easily change if I roll it out.
    Since I chose not to Axelise, I looked at 10% + 2 and also 8/9/12.
    Both give a 1-pip drop but I've learned by experience that these tend
    to underestimate the underdog in this type of long-distance near-race
    so I adjusted to a take. As it's so marginal, dropping would have been
    just as good.

    Paul


    XGID=-----CD-D---cD--acad-a--b-:0:0:-1:00:4:6:3:0:10
    X:eXtremeGammon O:Daniel

    Score is X:6 O:4. Unlimited Game, Jacoby Beaver
    +13-14-15-16-17-18------19-20-21-22-23-24-+
    | X O | | O O |
    | X O | | O O |
    | X O | | O O |
    | O | | O |
    | | | |
    | |BAR| |
    | | | |
    | O | | X |
    | O X | | X |
    | O X | | X X |
    | O X X X | | X X X |
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    Pip count X: 109 O: 123 X-O: 6-4
    Cube: 1
    X on roll, cube action

    Analyzed in 4-ply
    Player Winning Chances: 78.67% (G:0.03% B:0.00%)
    Opponent Winning Chances: 21.33% (G:0.00% B:0.00%)

    Cubeless Equities: No Double=+0.574, Double=+1.147

    Cubeful Equities:
    No double: +0.880 (-0.117)
    Double/Take: +0.997
    Double/Pass: +1.000 (+0.003)

    Best Cube action: Double / Take

    eXtreme Gammon Version: 2.10

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Timothy Chow@21:1/5 to peps...@gmail.com on Tue Sep 13 21:53:07 2022
    On 9/13/2022 4:39 PM, peps...@gmail.com wrote:
    Obviously, this should be treated as a race but I've learned not to Axelise these positions. The Axelisation algo has been trained only on pure
    races not near-races, so I don't think the Axelisations are reliable here.
    It turns out to be extremely close exactly marginal. So far, XG agrees
    with my take, but that can easily change if I roll it out.
    Since I chose not to Axelise, I looked at 10% + 2 and also 8/9/12.
    Both give a 1-pip drop but I've learned by experience that these tend
    to underestimate the underdog in this type of long-distance near-race
    so I adjusted to a take. As it's so marginal, dropping would have been
    just as good.

    Another thing to be aware of is that even though this sort of position
    looks very simple, it's one where XG 3-ply routinely makes mistakes.
    I commented on this phenomenon back in 2013.

    http://www.bgonline.org/forums/webbbs_config.pl?read=136252

    For this reason, when I roll out races or near-races, I usually use the strongest rollout settings that I have the patience for.

    ---
    Tim Chow

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pepstein5@gmail.com@21:1/5 to Tim Chow on Wed Sep 14 00:24:17 2022
    On Wednesday, September 14, 2022 at 2:53:10 AM UTC+1, Tim Chow wrote:
    On 9/13/2022 4:39 PM, peps...@gmail.com wrote:
    Obviously, this should be treated as a race but I've learned not to Axelise these positions. The Axelisation algo has been trained only on pure
    races not near-races, so I don't think the Axelisations are reliable here. It turns out to be extremely close exactly marginal. So far, XG agrees
    with my take, but that can easily change if I roll it out.
    Since I chose not to Axelise, I looked at 10% + 2 and also 8/9/12.
    Both give a 1-pip drop but I've learned by experience that these tend
    to underestimate the underdog in this type of long-distance near-race
    so I adjusted to a take. As it's so marginal, dropping would have been
    just as good.
    Another thing to be aware of is that even though this sort of position
    looks very simple, it's one where XG 3-ply routinely makes mistakes.
    I commented on this phenomenon back in 2013.

    http://www.bgonline.org/forums/webbbs_config.pl?read=136252

    For this reason, when I roll out races or near-races, I usually use the strongest rollout settings that I have the patience for.

    I didn't have the patience to wait to the end of the rollout when I did it last night.
    However, looking at the numbers, the results were remarkably similar to the analysis stat. It consistently wobbled in the 0.99 -- 1.008 range.

    Paul

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Stick Rice@21:1/5 to peps...@gmail.com on Wed Sep 14 14:55:38 2022
    On Wednesday, September 14, 2022 at 3:24:18 AM UTC-4, peps...@gmail.com wrote:
    On Wednesday, September 14, 2022 at 2:53:10 AM UTC+1, Tim Chow wrote:
    On 9/13/2022 4:39 PM, peps...@gmail.com wrote:
    Obviously, this should be treated as a race but I've learned not to Axelise
    these positions. The Axelisation algo has been trained only on pure races not near-races, so I don't think the Axelisations are reliable here.
    It turns out to be extremely close exactly marginal. So far, XG agrees with my take, but that can easily change if I roll it out.
    Since I chose not to Axelise, I looked at 10% + 2 and also 8/9/12.
    Both give a 1-pip drop but I've learned by experience that these tend
    to underestimate the underdog in this type of long-distance near-race
    so I adjusted to a take. As it's so marginal, dropping would have been just as good.
    Another thing to be aware of is that even though this sort of position looks very simple, it's one where XG 3-ply routinely makes mistakes.
    I commented on this phenomenon back in 2013.

    http://www.bgonline.org/forums/webbbs_config.pl?read=136252

    For this reason, when I roll out races or near-races, I usually use the strongest rollout settings that I have the patience for.
    I didn't have the patience to wait to the end of the rollout when I did it last night.
    However, looking at the numbers, the results were remarkably similar to the analysis stat. It consistently wobbled in the 0.99 -- 1.008 range.

    Paul

    It's not just that 3 ply makes mistakes, it's that the move filter is also not large enough to handle the many legal options (esp w/smaller rolls) that comes with such positions. In this position for example I'm sure you did a standard 3ply/XGR rollout (
    or whatever the default is) and determined good enough, it's borderline right? Well I did a rollout with ++/++ and a gigantic move filter and the position is worth about +0.915 aka a monster take.

    Stick

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pepstein5@gmail.com@21:1/5 to Stick Rice on Thu Sep 15 01:53:08 2022
    On Wednesday, September 14, 2022 at 10:55:40 PM UTC+1, Stick Rice wrote:
    On Wednesday, September 14, 2022 at 3:24:18 AM UTC-4, peps...@gmail.com wrote:
    On Wednesday, September 14, 2022 at 2:53:10 AM UTC+1, Tim Chow wrote:
    On 9/13/2022 4:39 PM, peps...@gmail.com wrote:
    Obviously, this should be treated as a race but I've learned not to Axelise
    these positions. The Axelisation algo has been trained only on pure races not near-races, so I don't think the Axelisations are reliable here.
    It turns out to be extremely close exactly marginal. So far, XG agrees with my take, but that can easily change if I roll it out.
    Since I chose not to Axelise, I looked at 10% + 2 and also 8/9/12. Both give a 1-pip drop but I've learned by experience that these tend to underestimate the underdog in this type of long-distance near-race so I adjusted to a take. As it's so marginal, dropping would have been just as good.
    Another thing to be aware of is that even though this sort of position looks very simple, it's one where XG 3-ply routinely makes mistakes.
    I commented on this phenomenon back in 2013.

    http://www.bgonline.org/forums/webbbs_config.pl?read=136252

    For this reason, when I roll out races or near-races, I usually use the strongest rollout settings that I have the patience for.
    I didn't have the patience to wait to the end of the rollout when I did it last night.
    However, looking at the numbers, the results were remarkably similar to the
    analysis stat. It consistently wobbled in the 0.99 -- 1.008 range.

    Paul
    It's not just that 3 ply makes mistakes, it's that the move filter is also not large enough to handle the many legal options (esp w/smaller rolls) that comes with such positions. In this position for example I'm sure you did a standard 3ply/XGR rollout
    (or whatever the default is) and determined good enough, it's borderline right? Well I did a rollout with ++/++ and a gigantic move filter and the position is worth about +0.915 aka a monster take.

    Your understanding of what I did is exactly correct (unfortunately). I think Tim would have approached the question more thoroughly and scientifically.
    I also agree that 0.915 can be described as a "monster take".
    This accords with my understanding of backgammon. For these long pseudo-races, 8/9/12 and 10% + 2 generally underrate the taker's chances quite significantly, so you have to do
    a considerable adjustment, not just the one pip deficit here. That's why I took somewhat confidently OTB, and was (a bit) surprised about the (wrong) marginal verdict.

    Very helpful post (by you) here!!

    Thanks.

    BTW, my post which (jokingly) referred to you, was an observation of mine on how weak players lose money to pros or semi-pros.
    Of course, I have no idea if it applies to you personally. It's perfectly plausible (to me) that you only ever play strong players at backgammon,
    and all of them know that you shouldn't always cash positions even when you have significant losing sequences.

    Paul

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pepstein5@gmail.com@21:1/5 to Stick Rice on Thu Sep 15 09:47:37 2022
    On Wednesday, September 14, 2022 at 10:55:40 PM UTC+1, Stick Rice wrote:
    On Wednesday, September 14, 2022 at 3:24:18 AM UTC-4, peps...@gmail.com wrote:
    On Wednesday, September 14, 2022 at 2:53:10 AM UTC+1, Tim Chow wrote:
    On 9/13/2022 4:39 PM, peps...@gmail.com wrote:
    Obviously, this should be treated as a race but I've learned not to Axelise
    these positions. The Axelisation algo has been trained only on pure races not near-races, so I don't think the Axelisations are reliable here.
    It turns out to be extremely close exactly marginal. So far, XG agrees with my take, but that can easily change if I roll it out.
    Since I chose not to Axelise, I looked at 10% + 2 and also 8/9/12. Both give a 1-pip drop but I've learned by experience that these tend to underestimate the underdog in this type of long-distance near-race so I adjusted to a take. As it's so marginal, dropping would have been just as good.
    Another thing to be aware of is that even though this sort of position looks very simple, it's one where XG 3-ply routinely makes mistakes.
    I commented on this phenomenon back in 2013.

    http://www.bgonline.org/forums/webbbs_config.pl?read=136252

    For this reason, when I roll out races or near-races, I usually use the strongest rollout settings that I have the patience for.
    I didn't have the patience to wait to the end of the rollout when I did it last night.
    However, looking at the numbers, the results were remarkably similar to the
    analysis stat. It consistently wobbled in the 0.99 -- 1.008 range.

    Paul
    It's not just that 3 ply makes mistakes, it's that the move filter is also not large enough to handle the many legal options (esp w/smaller rolls) that comes with such positions. In this position for example I'm sure you did a standard 3ply/XGR rollout
    (or whatever the default is) and determined good enough, it's borderline right? Well I did a rollout with ++/++ and a gigantic move filter and the position is worth about +0.915 aka a monster take.

    Stick
    I think Axelisation finds a big take too, despite my scepticism about using it for non-pure races.
    The taker has 12 crossovers against the opponent's 11 so add 1 to the taker. Both sides have a gap on 4,5,6 which is filled in by the opponent.
    So add 1 to both.
    We then get 125 against 110. 110 * 7/6 - 125 = 3 1/3 > 2 -- monster take indeed!

    Paul

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Timothy Chow@21:1/5 to Stick Rice on Fri Sep 16 07:41:51 2022
    On 9/14/2022 5:55 PM, Stick Rice wrote:

    Well I did a rollout with ++/++ and a gigantic move filter and the
    position is worth about +0.915 aka a monster take.
    Wait, what position are we talking about? I just did an
    XGR+/XGR+/Gigantic rollout and it came out to a borderline
    pass. See below.

    I should mention that I just bought a new laptop and installed
    version 2.19.211.pre-release, which is not the version I've been
    using for the past few years, but I doubt that makes a difference.
    I did notice, however, that when I copied and pasted the position,
    it mangled the XGID in the same way that I've seen Paul's XGIDs
    mangled, and I had to correct it manually.

    XGID=-B--A-DACA--dC---d-dc-----:0:0:1:00:0:0:3:0:10
    X:Player 2 O:Player 1

    Score is X:0 O:0. Unlimited Game, Jacoby Beaver
    +13-14-15-16-17-18------19-20-21-22-23-24-+
    | X O | | O O |
    | X O | | O O |
    | X O | | O O |
    | O | | O |
    | | | |
    | |BAR| |
    | | | |
    | O | | X |
    | O X | | X |
    | O X | | X X |
    | O X X X | | X X X |
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    Pip count X: 109 O: 123 X-O: 0-0
    Cube: 1
    X on roll, cube action

    Analyzed in Rollout
    No double
    Player Winning Chances: 78.63% (G:0.01% B:0.00%)
    Opponent Winning Chances: 21.37% (G:0.00% B:0.00%)
    Double/Take
    Player Winning Chances: 78.71% (G:0.01% B:0.00%)
    Opponent Winning Chances: 21.29% (G:0.00% B:0.00%)

    Cubeless Equities: No Double=+0.573, Double=+1.149

    Cubeful Equities:
    No double: +0.890 (-0.110)
    Double/Take: +1.016 (+0.016)
    Double/Pass: +1.000

    Best Cube action: Double / Pass

    Rollout:
    842 Games rolled with Variance Reduction.
    Moves and cube decisions: XG Roller+
    Search interval: Gigantic
    Confidence No Double: ± 0.006 (+0.884..+0.896)
    Confidence Double: ± 0.008 (+1.007..+1.024)

    Double Decision confidence: 100.0%
    Take Decision confidence: 100.0%

    Duration: 9 hours 19 minutes

    eXtreme Gammon Version: 2.19.211.pre-release


    XGID=-B--A-DACA--dC---d-dc-----:0:0:1:00:0:0:3:0:1X:Player 2 O:Player 1
    0

    Score is X:0 O:0. Unlimited Game, Jacoby Beaver
    +13-14-15-16-17-18------19-20-21-22-23-24-+
    | X O | | O O |
    | X O | | O O |
    | X O | | O O |
    | O | | O |
    | | | |
    | |BAR| |
    | | | |
    | O | | X |
    | O X | | X |
    | O X | | X X |
    | O X X X | | X X X |
    +12-11-10--9--8--7-------6--5--4--3--2--1-+
    Pip count X: 109 O: 123 X-O: 0-0
    Cube: 1
    X on roll, cube action

    Analyzed in Rollout
    No double
    Player Winning Chances: 78.63% (G:0.01% B:0.00%)
    Opponent Winning Chances: 21.37% (G:0.00% B:0.00%)
    Double/Take
    Player Winning Chances: 78.71% (G:0.01% B:0.00%)
    Opponent Winning Chances: 21.29% (G:0.00% B:0.00%)

    Cubeful Equities:
    No double: +0.890 (-0.110)
    Double/Take: +1.016 (+0.016)
    Double/Pass: +1.000

    Best Cube action: Double / Pass

    Rollout:
    842 Games rolled with Variance Reduction.
    Moves and cube decisions: XG Roller+
    Search interval: Gigantic
    Confidence No Double: ± 0.006 (+0.884..+0.896)
    Confidence Double: ± 0.008 (+1.007..+1.024)

    eXtreme Gammon Version: 2.19.211.pre-release

    ---
    Tim Chow

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)