How not to blitz
tc...@lsa.umich.edu
because I really should have known better. See if you can find the
right play. Ignore the score; this is a money game.
GNU Backgammon Position ID: zM3gAEXDNuIBKA
Match ID : cAkOAGAAEAAA
+24-23-22-21-20-19------18-17-16-15-14-13-+ O: gnubg
| X X O O O | O | O X | 6 points
| O O O | | O X |
| O | | X |
| | | X |
| | | |
| |BAR| |v (Cube: 1)
| | | |
| | | |
| | | O |
| X X X | | X O | Rolled 43
| X O O X X | | X X O | 2 points
+-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: tchow
--
Tim Chow tchow-at-alum-dot-mit-dot-edu
The range of our projectiles---even ... the artillery---however great, will
never exceed four of those miles of which as many thousand separate us from
the center of the earth. ---Galileo, Dialogues Concerning Two New Sciences
Bradley K. Sherman
<tc...@lsa.umich.edu> wrote:
> +24-23-22-21-20-19------18-17-16-15-14-13-+ O: gnubg
> | X X O O O | O | O X | 6 points
> | O O O | | O X |
> | O | | X |
> | | | X |
> | | | |
> | |BAR| |v (Cube: 1)
> | | | |
> | | | |
> | | | O |
> | X X X | | X O | Rolled 43
> | X O O X X | | X X O | 2 points
> +-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: tchow
Money game == Jacoby rule == 7/4,13/9
--bks
tc...@lsa.umich.edu
When you mention the Jacboy rule here, are you saying that the Jacoby rule is
relevant to your choice of move? How so?
Bradley K. Sherman
<tc...@lsa.umich.edu> wrote:
>In article <hnugcl$j33$1...@reader1.panix.com>,
>Bradley K. Sherman <b...@panix.com> wrote:
>>Money game == Jacoby rule == 7/4,13/9
>
>When you mention the Jacboy rule here, are you saying that the Jacoby rule is
>relevant to your choice of move? How so?
If I get hit, no matter how ugly the resulting position is, I can't get gammoned.
Thus I can be more aggressive with my play.
--bks
Grunty
Having made the ace-point you're (regrettably) committed to a blitz.
You'd be at a disadvantage in a positional game, since you lack those
two checkers for successfully building a block upon O's checkers,
whereas O has a strong front position against yours. This scenario is
likely to come up with a quiet play like 10/6 24/21, where O would be
a favorite to make an anchor.
So why I'd continue attacking with 13/9 7/4* trying to deny O his best
anchor.
And with some luck (O not rolling a 3 or a 4), I could end up well
positioned to try an initial double, mostly based on the gammon threat.
Grunty
> So why I'd continue attacking with 13/9 7/4*
> trying to deny O his best anchor.
I meant: So xxx I'd continue attacking...
tc...@lsa.umich.edu
Bradley K. Sherman <b...@panix.com> wrote:
>If I get hit, no matter how ugly the resulting position is, I can't get
>gammoned. Thus I can be more aggressive with my play.
Would your play be different if the Jacoby rule were not in force?
Bradley K. Sherman
<tc...@lsa.umich.edu> wrote:
>In article <hnuqdj$47m$1...@reader1.panix.com>,
>Bradley K. Sherman <b...@panix.com> wrote:
>>If I get hit, no matter how ugly the resulting position is, I can't get
>>gammoned. Thus I can be more aggressive with my play.
>
>Would your play be different if the Jacoby rule were not in force?
Probably not. If you're implying I give too much weight to
Jacoby, you might be right. I'm much more familiar with
tournament play than money play (there used to be five or
more tournaments per week in the SF Bay Area).
--bks
Walt
> I made a huge blunder in the position below, all the more embarrassing
> because I really should have known better. See if you can find the
> right play. Ignore the score; this is a money game.
> GNU Backgammon Position ID: zM3gAEXDNuIBKA
> Match ID : cAkOAGAAEAAA
> +24-23-22-21-20-19------18-17-16-15-14-13-+ O: gnubg
> | X X O O O | O | O X | 6 points
> | O O O | | O X |
> | O | | X |
> | | | X |
> | | | |
> | |BAR| |v (Cube: 1)
> | | | |
> | | | |
> | | | O |
> | X X X | | X O | Rolled 43
> | X O O X X | | X X O | 2 points
> +-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: tchow
>
The options here seem to be:
hit 'em both with 7/3* 7/4*
hit loose with 10/3*
hit loose with 7/4* 10/6
bring in two more builders and make a valuable blocking point 6 away
from the lead runner with 13/9 13/10
play it safe with 24/21 10/6
With one checker on the bar already, hitting both loose is reckless.
Of the loose hits, 10/6 7/4* looks better since it slots a better point
and doubles the number of builders in range.
The safe play doesn't do it for me - too passive.
Over the board, I'm pretty sure I'd play 13/9 13/10. This brings two
more builders in range making a blitz reasonable. With the current
position, we don't have enough ammo in range to follow through with the
blitz. If he fails to anchor, ship the cube.
But perhaps the real blunder here is failing to double - there are lots
of market losers here, although you didn't roll one.
//Walt
Walt
> Of the loose hits, 10/6 7/4* looks better since it slots a better point
> and doubles the number of builders in range.
Correction. It does nothing of the sort.
13/9 7/4* does though. This seems to be the best loose hit, and may be
best overall for money/Jacoby. I wouldn't touch it where gammons counted.
//Walt
Grunty
> But perhaps the real blunder here is failing to double - > there are lots of market losers here, although you
> didn't roll one.
Buddy, either you got up on a allbrighty mood this morning, or I got
up on a very dark one...
tc...@lsa.umich.edu
>
> GNU Backgammon Position ID: zM3gAEXDNuIBKA
> Match ID : cAkOAGAAEAAA
> +24-23-22-21-20-19------18-17-16-15-14-13-+ O: gnubg
> | X X O O O | O | O X | 6 points
> | O O O | | O X |
> | O | | X |
> | | | X |
> | | | |
> | |BAR| |v (Cube: 1)
> | | | |
> | | | |
> | | | O |
> | X X X | | X O | Rolled 43
> | X O O X X | | X X O | 2 points
> +-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: tchow
The play that tops the rollout, 13/9 7/4*, seems perfectly obvious to me now,
and I'm having trouble remembering what possessed me to play 13/9 13/10. I
think I felt that the blitz was not likely to succeed, and so I decided to
switch to priming as my primary plan. Perhaps I also thought that bringing
more ammunition down could help with the blitz if O danced again, and I also
was concerned about leaving too many blots around when O had a 3-point board.
But all those concerns are phantoms. The blitz is chancy, but with
the 1-point made, nine checkers in the zone with more reinforcements
available on the midpoint, and essentially equal board strength, there's
no reason to give up on the blitz until O actually anchors. And O has not
anchored yet. Therefore it is right to hit; the top five plays all hit.
If O had no board and if X had a stronger attacking formation, then the
double hit might be right since it does the best job of preventing O from
anchoring. But here the double hit is too loose. The thematic thing to
do with the non-hitting move is to bring another checker into the zone.
Now we're down to 13/9 7/4* or 13/10 7/3*, and it's pretty clear that
13/9 7/4* is better, since it knocks O off the more important point and
gives more flexibility and coverage.
Walt mentioned the possibility that X should have doubled. This is very
optimistic. When assessing a blitz, it is crucial to examine the defender's
home board, even though that might seem to be a sector that is far from the
scene of action. The point is that carrying out a blitz often requires
hitting loose, and if the defender hits back from the bar then suddenly the
defender's home board becomes highly relevant. O's board is even stronger
than X's here and so O has a trivial take. X is nowhere near a double.
In fact X is only about even money to win the game, with just a slight
advantage because of the gammon threat.
Since Bradley mentioned the Jacoby rule, I started another rollout running
with the Jacoby rule turned off. I don't expect it to show much difference
though. The Jacoby rule makes less difference to checker play than many
people believe. It does indeed make it impossible for a player to become
too good to double with the cube centered, but here there is not much chance
that O will suddenly become too good to double. The Jacoby rule does not
usually "prevent you from being gammoned." If things go poorly then you
may face a cube, but you may have to take it, and then you can get gammoned
of course. If things go very poorly then you may have to drop, but a drop
is a drop whether the Jacoby rule is in effect or not. Only when things go
outrageously poorly and your opponent would otherwise be too good to double
does the Jacoby rule offer you any protection.
I've often wondered how much difference the Jacoby rule has made in
practice. Clearly the intent was to reduce "downtime"---i.e., boredom
caused by sitting around idly while your opponent has you closed out
and is gammoning you. Given that in expert play, the Jacoby rule makes
relatively little difference, the amount of downtime reduction might
not be all that large. On the other hand, your run-of-the-mill player
probably plays on for the gammon a lot of the time when objectively
he should just cash. So for such players, the Jacoby rule may indeed
reduce downtime.
1. Rollout 13/9 7/4* Eq.: +0.128
0.508 0.238 0.011 - 0.492 0.195 0.024 CL +0.047 CF +0.128
[0.001 0.001 0.000 - 0.001 0.002 0.001 CL 0.004 CF 0.010]
2. Rollout 13/10 7/3* Eq.: +0.047 ( -0.081)
0.497 0.219 0.009 - 0.503 0.186 0.013 CL +0.024 CF +0.047
[0.002 0.002 0.000 - 0.002 0.002 0.001 CL 0.004 CF 0.012]
3. Rollout 10/6 7/4* Eq.: +0.027 ( -0.102)
0.491 0.217 0.008 - 0.509 0.185 0.013 CL +0.009 CF +0.027
[0.001 0.002 0.001 - 0.001 0.002 0.001 CL 0.004 CF 0.012]
4. Rollout 7/4* 7/3* Eq.: +0.019 ( -0.110)
0.488 0.239 0.011 - 0.512 0.204 0.019 CL +0.003 CF +0.019
[0.001 0.001 0.000 - 0.001 0.002 0.001 CL 0.004 CF 0.011]
5. Rollout 24/21 7/3* Eq.: -0.023 ( -0.151)
0.482 0.203 0.008 - 0.518 0.192 0.016 CL -0.033 CF -0.023
[0.001 0.002 0.000 - 0.001 0.002 0.001 CL 0.004 CF 0.012]
6. Rollout 24/21 10/6 Eq.: -0.050 ( -0.178)
0.485 0.171 0.007 - 0.515 0.166 0.007 CL -0.026 CF -0.050
[0.001 0.002 0.000 - 0.001 0.002 0.000 CL 0.004 CF 0.011]
7. Rollout 10/3* Eq.: -0.087 ( -0.215)
0.472 0.190 0.007 - 0.528 0.180 0.010 CL -0.050 CF -0.087
[0.001 0.002 0.000 - 0.001 0.002 0.000 CL 0.004 CF 0.010]
8. Rollout 13/6 Eq.: -0.150 ( -0.278)
0.466 0.181 0.008 - 0.534 0.186 0.010 CL -0.075 CF -0.150
[0.001 0.002 0.000 - 0.001 0.002 0.000 CL 0.004 CF 0.012]
9. Rollout 13/10 13/9 Eq.: -0.180 ( -0.308)
0.465 0.184 0.007 - 0.535 0.196 0.012 CL -0.088 CF -0.180
[0.002 0.002 0.000 - 0.002 0.002 0.001 CL 0.005 CF 0.012]
Full cubeful rollout with var.redn.
1296 games, Mersenne Twister dice gen. with seed 850551878 and
quasi-random dice
Play: supremo 2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 16 more moves within equity 0.32
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]
Bradley K. Sherman
<tc...@lsa.umich.edu> wrote:
>
>Since Bradley mentioned the Jacoby rule, I started another rollout running
>with the Jacoby rule turned off. I don't expect it to show much difference
>though.
It would certainly be interesting to see a few early game situations
where Jacoby really does change the checker play. I freely admit
I have played too few money matches to calibrate my internal
Jacoby-o-meter.
BTW, when gnubg increases the ply-level, is the code written in such
a way that additional processors can be incorporated?
--bks
tc...@lsa.umich.edu
>a way that additional processors can be incorporated?
I don't know, but this would be a nice feature. You could try asking your
question on the bug-gnubg mailing list.
tc...@lsa.umich.edu
>where Jacoby really does change the checker play.
I thought about this a little and it seems pretty tough to come up with
a plausible example. Maybe something where you have a choice between
making a six-prime that guarantees a cash next time, and a blitzing play
which usually makes you too good but also loses your cash sometimes. Then
the blitz might be right without Jacoby, but wrong with Jacoby.
If on the other hand you're looking for an example where the Jacoby
rule indicates that you should play more aggressively because "you don't
have to be afraid of your opponent's gammons," then that seems even more
difficult to come up with (especially in the early game), as I explained
previously. Typically you'd have to be losing so badly that you're
opponent is bordering on being too good. A move which gains something
when it works, but leaves your opponent too good when it doesn't, might
be right with Jacoby and wrong without it. But it's not so easy to come
up with a position like that where the cube is in the center and your
opponent doesn't have an immediate cube turn that renders Jacoby moot.
In practice you will very rarely go wrong if you simply ignore the Jacoby
rule for money-game checker-play decisions. On the other hand the Jacoby
rule does sometimes mean more aggressive cube-turning, as in the example
I posted recently.
Peter Schneider
"Bradley K. Sherman" wrote
> BTW, when gnubg increases the ply-level, is the code written in such
> a way that additional processors can be incorporated?
I have heard that recent versions do. But I didn't look into the code, nor
do I have a multicore to test it with. That test should be simple.
Best,
Peter aka the juggler
tc...@lsa.umich.edu
Peter Schneider <schneiderp...@gmx.net> wrote:
>"Bradley K. Sherman" wrote
>
>> BTW, when gnubg increases the ply-level, is the code written in such
>> a way that additional processors can be incorporated?
>
>I have heard that recent versions do. But I didn't look into the code, nor
>do I have a multicore to test it with. That test should be simple.
Just to be clear, you're saying that you heard that something like GNU
4-ply evaluation can be run in parallel? As opposed to rollouts? Both
should be parallelizable but doing rollouts in parallel sounds a bit
easier to program than doing evaluations in parallel.
Peter Schneider
<tc...@lsa.umich.edu> wrote
> Just to be clear, you're saying that you heard that something like GNU
> 4-ply evaluation can be run in parallel? As opposed to rollouts?
Err... now that you mention it... I may have heard it about rollouts.
> Both
> should be parallelizable but doing rollouts in parallel sounds a bit
> easier to program than doing evaluations in parallel.
Why? Whether you generate rolls or just inspect all possible rolls should
not make much of a diff.
tc...@lsa.umich.edu
Peter Schneider <schneiderp...@gmx.net> wrote:
>Why? Whether you generate rolls or just inspect all possible rolls should
>not make much of a diff.
Well, 4-ply evaluation comes with a few more bells and whistles, like move
filters, which might currently be programmed in a serial manner. You're
right that it shouldn't be hard in principle, but I'm just guessing that
rollouts will be slightly easier to parallelize in practice.
tc...@lsa.umich.edu
In article <4ba43758$0$513$b45e...@senator-bedfellow.mit.edu>, I wrote:
>> GNU Backgammon Position ID: zM3gAEXDNuIBKA
>> Match ID : cAkOAGAAEAAA
>> +24-23-22-21-20-19------18-17-16-15-14-13-+ O: gnubg
>> | X X O O O | O | O X | 6 points
>> | O O O | | O X |
>> | O | | X |
>> | | | X |
>> | | | |
>> | |BAR| |v (Cube: 1)
>> | | | |
>> | | | |
>> | | | O |
>> | X X X | | X O | Rolled 43
>> | X O O X X | | X X O | 2 points
>> +-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: tchow
[...]
>Since Bradley mentioned the Jacoby rule, I started another rollout running
>with the Jacoby rule turned off. I don't expect it to show much difference
>though.
As I expected, it doesn't make a big difference. Hitting is still a clear
winner. I didn't roll out as many different plays here but the overall
message is clear.
1. Rollout 13/9 7/4* Eq.: +0.097
0.507 0.237 0.012 - 0.493 0.198 0.022 CL +0.044 CF +0.097
[0.002 0.001 0.000 - 0.002 0.002 0.001 CL 0.005 CF 0.010]
2. Rollout 13/10 7/3* Eq.: +0.022 ( -0.075)
0.494 0.214 0.008 - 0.506 0.182 0.013 CL +0.015 CF +0.022
[0.001 0.002 0.000 - 0.001 0.002 0.001 CL 0.004 CF 0.011]
3. Rollout 10/6 7/4* Eq.: -0.017 ( -0.114)
0.489 0.215 0.008 - 0.511 0.191 0.014 CL -0.004 CF -0.017
[0.002 0.002 0.000 - 0.002 0.002 0.001 CL 0.004 CF 0.012]
4. Rollout 24/21 10/6 Eq.: -0.035 ( -0.131)
0.488 0.173 0.006 - 0.512 0.168 0.007 CL -0.020 CF -0.035
[0.001 0.002 0.000 - 0.001 0.002 0.000 CL 0.004 CF 0.011]
Full cubeful rollout with var.redn.
1296 games, Mersenne Twister dice gen. with seed 847426592 and
Bradley K. Sherman
<tc...@lsa.umich.edu> wrote:
>
>As I expected, it doesn't make a big difference. Hitting is still a clear
Noted. Thanks.
--bks