Re: [Bug-gnubg] GnuBG faring poorly in back games

[Øystein O Johansen]

> ParlorBot is a bot on FIBS that uses a recent build of GnuBG.
> It seems that some players have figured out how to do pretty
> well against it, by deliberately playing severe back games.

Backgames can be really powerfull if you know how to play them. I believe
you is you say there are some weaknesses with GNU Backgammon and backgames.
The thing is to know how to defend yourself against someone who's playing a
backgame. The two best ways.

- Destroy the timing
- Build a prime

(These two items are a bit connected since building a prime is the way to
destroy the timing.)

Then playing agains a backgame you will leave a shot. It's really really
seldom you won't leave a shot agains a backgame. The key is to leave the
shot while you opponent has already crached his board or not built his
board yet.

 parlorbot : 0                 logistics : 0
  1)                            51: 8/3 6/5

logistics already has a plan..

  2) 23: 24/22* 24/21

Maybe 24/22* 13/10 is better, but tell me it's wrong to hit!

                                31: 25/22 8/7
  3) 55: 22/17* 13/8 8/3* 8/3   12: 25/24 25/23
  4)  Doubles => 2               Takes
  5) 24: 8/4 6/4                 Doubles => 4
  6)  Takes                     61: 24/23 13/7
  7) 14: 21/20* 17/13

This may be the crusial move: 13/8 defends better against the backgame, but
do you really think it is right?

                                54: 25/20 20/16
  8) 42: 13/9* 9/7

Or is this the crusial move? I don't know, but maybe it should not hit? It
looks a bit weird not hitting?

                                42: 25/23 13/9
  9) 46: 20/16* 16/10

I believe this may be the worst one. Making the barpoint is the key to
making the important prime. The prime can hopefully make logistic crash his
board before getting a shot. I believe that 13/7 13/9 is better. At this
moment parlorbot should concern about the timing.

                                55: 25/20 13/8 13/8 13/8

See? Why didn't you the barpoint, my friend?

 10) 14: 13/9 6/5*

This is also interesting. This is a new oportunity to make the barpoint. I
believe this is a major blunder and the move that dicides this match.

                                33:
 11)  Doubles => 8               Takes

Should drop this, but I didn't expect a drop. The fact that GNU Backgammon
evaluates this as a drop may confuse the evaluation of move 10.

 12) 56: 13/7 10/5

13/7 13/8 is also a strong candidate. As I write above, it's not dangerous
to be hit when the backgames borad is not yet built. 13/7 13/8 is also
planing to kill sixes. If the full prime is completed it will stay for a
couple of roll since sixes are killed. Ideally he will also but spare
checkers on the 8-point so he can play fives with comfort.

                                 Doubles => 16
 13)  Takes                     56:
 14) 32: 13/10 13/11

Make the 8-point and hope he enters, so his timing exhausts. logistics'
sixes becomes very bad for him when he enters. Not a big mistake though.

                                32: 25/23 8/5
 15) 45: 11/7 10/5              64: 23/17* 17/13
 16) 36: 25/22 22/16            56: 23/17 13/8
 17) 53: 16/11 11/8*            34:
 18) 13: 8/7 7/4

This doesn't look natural to me! Here he has to choose betwwen two plans.
a) make a flexible position which dosn't leave much shots og b) keep the
prime so logistics' board crunshes. This move does neither. 9/6 8/7 looks
like plan a move and 9/8 7/4 looks like plan b move. Not a burger, not a
lemmon but a little detail.

                                25: 25/23 8/3
 19) 23: 9/7 7/4                31: 7/4 6/5
 20) 55:                        51: 8/3 6/5
 21) 33: 7/4 7/4 7/4 6/3        45: 23/18 8/4
 22) 53: 5/0 3/0                33: 18/15 15/12 12/9 5/2
 23) 45: 5/0 4/0                35: 23/20* 7/2
 24) 33:                        12: 20/18 18/17
 25) 12: 25/24 4/2*             44:
 26) 15: 4/3                    43:
 27) 32: 6/3 4/2                35: 25/20 9/6
 28) 25: 6/4                    16: 20/19 19/13
 29) 64: 24/18 18/14            62: 17/11* 11/9
 30) 33:                        65: 24/18 18/13
 31) 21: 25/24 4/2              22: 13/11 11/9 9/7 9/7
 32) 33: 4/1* 4/1 4/1           43: 25/21 21/18
 33) 43:                        46: 18/14 14/8
 34) 14: 3/2                    66: 13/7 8/2 7/1* 7/1
 35) 42:                        64: 7/3 6/0
 36) 24:                        64: 6/0 6/2
 37) 23:                        13: 3/0 2/1
 38) 54:                        24: 4/2 4/0
 39) 12:                        22: 5/3 5/3 2/0 2/0
 40) 66: 25/19 19/13 13/7 7/1   11: 2/1 1/0 1/0 1/0
 41) 35: 3/0 3/0                61: 3/0 1/0
 42) 33: 3/0 2/0 2/0 2/0        56: 3/0 3/0
 43) 13: 2/1 1/0                51: 3/0 2/1
 44) 45: 1/0 1/0
                                 Wins 16 points

 Game 2
 parlorbot : 0                 logistics : 0
  1) 25: 24/22 13/8             52: 8/3* 6/4
  2) 36: 25/22* 22/16           11: 25/24 24/23 23/22 6/5
  3) 46: 24/20* 16/10           44: 25/21 13/9 13/9 9/5*
  4) 56: 25/20* 10/4*           63: 25/22
  5) 21: 8/7 6/4                32: 25/22 22/20
  6) 62: 20/14 7/5*             36: 25/22 22/16
  7) 51: 14/9* 6/5              35: 25/22 13/8
  8)  Doubles => 2               Takes
  9) 31: 13/10 10/9

Naaaaa.... The barpoint is the important point. If you can't make it, at
least slot it! The shot isn't dangerous since it no dangerous board yet.
13/10 8/7 is the good move.

                                 Doubles => 4
 10)  Takes                     46: 13/9 13/7
 11) 22: 13/11 11/9 9/7 9/7     46: 22/16* 8/4
 12)  Doubles => 8               Takes
 13) 35: 25/20 8/5              26: 16/10 7/5*
 14) 55: 25/20* 20/15* 15/10 7/2

What!?!??? This looks unlogical to me. I just can't see any reason to break
the prime to slot the 2-point. I really can't understand this. It doen't
look like a bug either.

                                 Doubles => 16
 15)  Takes                     32: 25/22 25/23*
 16) 24: 25/23 6/2*             15: 25/24 8/3
 17) 54: 23/18 18/14

Running... really interesting it maight be correct, but who knows?

                                44: 22/18* 22/18 9/5 9/5
 18) 15: 25/24 10/5             64: 24/18 22/18
 19) 31: 24/23 5/2              44: 18/14 18/14 8/4 6/2*
 20) 66:                        36: 14/11* 11/5
 21) 56:                        35: 14/9 5/2
 22) 36: 25/22*                 44:
 23) 15: 25/24 22/17            42:
 24) 54: 17/12 12/8             14: 25/24 9/5
 25) 44: 13/9 13/9 9/5 9/5      35: 18/15 18/13
 26) 54: 8/3 8/4                65: 24/18 18/13
 27) 66: 24/18 8/2              64: 15/11 13/7*
 28) 53: 25/22 22/17            64: 11/5 7/3
 29) 55: 17/12* 12/7 7/2 5/0    12: 25/24 24/22*
 30) 62:                        63: 22/16 16/13
 31) 33: 25/22* 6/3 6/3 5/2     26: 25/19 5/3*
 32) 51: 25/24                  54: 19/14 5/1*
 33) 66:                        66: 24/18 14/8 13/7 7/1
 34) 42:                        44: 18/14 14/10 10/6 8/4
 35) 25:                        51: 24/19 4/3
 36) 63:                        26: 19/17 17/11
 37) 24:                        32: 11/8 8/6
 38) 44:                        43: 6/2 6/3
 39) 36:                        26: 6/4 6/0
 40) 62: 25/19 19/17            55: 5/0 5/0 4/0 4/0
 41) 21: 17/15 2/1
                                 Wins 16 points

It's not the worst things I've seen from GNU Backgammon. There are actually
positions it misplays even worse. This is more like positional mistakes in
a backgame defence. Move 10 in the first game is the biggest mistake I
believe, and proves that GNU Backgammon still have something to learn.
These are position it should have some knowledge about. Joseph? Do you have
an idea how to improve this?

> GnuBG is run in TTY mode and is initialized as follows:
>
> To gnubg: set player 1 name opp
> To gnubg: set confirm new off
> To gnubg: set display off
> To gnubg: set output rawboard on
> To gnubg: set evaluation chequer eval plies 2
> To gnubg: set evaluation chequer eval cubeful on
> To gnubg: set evaluation cubedecision eval plies 2
> To gnubg: set evaluation cubedecision eval cubeful on
> To gnubg: set analysis chequer eval plies 2
> To gnubg: set analysis chequer eval cubeful on
> To gnubg: set analysis cubedecision eval plies 2
> To gnubg: set analysis cubedecision eval cubeful on
>
> Any thoughts?

Yes! Shouldn't it be a setting like:

set player 0 evaluation chequer eval plies 2
set player 0 evaluation cubedecision eval plies 2

???

Which match equity table are you using? How old is your version og GNU
Backgammon. I find some mismatches between my analysis and the moves done.

-Øystein


-------------------------------------------------------------------
The information contained in this message may be CONFIDENTIAL and is
intended for the addressee only. Any unauthorised use, dissemination of the
information or copying of this message is prohibited. If you are not the
addressee, please notify the sender immediately by return e-mail and delete
this message.
Thank you.