2a 4a 32 to play - warning QF ahead

[Walt]

XGID=--AABbCBC---bC----accbb---:0:0:1:32:3:1:0:5:3

X:X O:O
Score is X:3 O:1 5 pt.(s) match.
+13-14-15-16-17-18------19-20-21-22-23-24-+
| X O | | O O O O |
| X | | O O O O |
| X | | O O |
| | | |
| | | |
| |BAR| |
| | | |
| | | |
| X | | X |
| O X X | | X O X |
| O X X | | X O X X X |
+12-11-10--9--8--7-------6--5--4--3--2--1-+
Pip count X: 108 O: 120 X-O: 3-1/5
Cube: 1
X to play 32

I don't usually so QF, but this position is interesting, and if I just
post it as-is I'm fairly sure that most people here will just opt for
the prosaic 13/8.

But there's a better move (according to XG rollout). Can you find it?




--
//Walt

[Bradley K. Sherman]

Walt <walt_...@yahoo.com> wrote:
>Score is X:3 O:1 5 pt.(s) match.
> +13-14-15-16-17-18------19-20-21-22-23-24-+
> | X O | | O O O O |
> | X | | O O O O |
> | X | | O O |
> | | | |
> | | | |
> | |BAR| |
> | | | |
> | | | |
> | X | | X |
> | O X X | | X O X |
> | O X X | | X O X X X |
> +12-11-10--9--8--7-------6--5--4--3--2--1-+

> X:108 O:120, Cube:1, X to play 32, (not 13/8)

I certainly would have played 13/8, but given
the hint I will try:
8/3

--bks

p.s What's QF?

[Walt]

On 5/10/2013 12:05 PM, Bradley K. Sherman wrote:


>
> p.s What's QF?
>

Quiz factor. It's where you choose something other than what you would
do OTB (Over the board in an actual game) because it's presented as a
problem.

If there's an 'obvious' move, QF makes you think that it must be wrong,
otherwise why would anyone post the position?


--
//Walt

[crf]

Normally, 13/8 or 8/3. But my super quiz factor answer: 4/1, 3/1. 5s and 6s are great already, this makes 2s, 3s and 4s good, too.

[peps...@gmail.com]

On Friday, May 10, 2013 8:22:03 PM UTC+1, crf wrote:
> Normally, 13/8 or 8/3. But my super quiz factor answer: 4/1, 3/1. 5s and 6s are great already, this makes 2s, 3s and 4s good, too.

I think you're right to max the QF factor. However, I don't think your play's good because X's pip count isn't large enough to waste pips by moving to the ace point. I think donating goods to charity is a fine thing. However, the desired gifts are generally books or clothing, rather than pips.

Avoid getting stripped by playing 6/3 4/2.

Paul Epstein

[badgolferman]

Well, I would have played 8/3 from the get go.

[peps...@gmail.com]

I think 8/3 would have been my choice too, even if Walt hadn't introduced the problem with a song about Quentin Farley. Like crf, I looked for something a bit more special. I'd be very surprised if there's much equity at stake there. Surely, we're just talking about improvements which gain 0.02 equity or so, at the most?

Paul Epstein

[Freeven]

It seems that knowing 13/8 isn't best should be helpful in figuring out
why something else is better, but I'm not having much luck reverse
engineering this.

After 13/8, 3s seem to be the most awkward number, since the spares on
the 8pt will be blocked on them. As far as I can see, all of X's 3s can
be played comfortably next turn after 13/8, so maybe the problem XG is
trying to solve is caused by consecutive rolls with 3s in them. If
that's the case, X should be looking to reach a position that allows him
to absorb a lot of 3s. That suggests playing 8/6 with the 2, since
that's the only move that "creates" another 3. But that doesn't seem to
help, since X must now find a 3 to play, and either 6/3 or 4/1 "destroy"
a 3, for no net gain.

Either there's an aspect of this "3 problem" I'm not considering, or
maybe I'm chasing phantoms and something else is the issue. In either
case, I wasn't able to reason my way to daylight on this.

I did end up peaking at XG. I'd still play 13/8.

[Tim Chow]

I would play 13/8. XG sometimes does funny things in holding games
with blotty boards, so if the rollout favors something other than 13/8
I'd want to investigate carefully before taking the rollout result at
face value.

---
Tim Chow

[peps...@gmail.com]

Perhaps 13/8 is best but there are 7 billion people on this planet (perhaps even slightly more) and a few of them are desperately eager to find out the reasons why this move is your preferred choice.
When I learned the statement of Pythagoras's theorem at school, the teacher and textbook didn't give any proof whatsoever, and I remained ignorant of the proof until I was about fifteen, several years after I knew the statement.
Backgammon is exactly like Pythagoras's theorem -- the assessments should have accompanying explanations.

Paul Epstein

[Stick]

13/8 is correct, period. Show me a rollout that says otherwise and
I'll show you a rollout that is wrong.

Stick

[peps...@gmail.com]

On Saturday, May 11, 2013 7:59:09 PM UTC+1, Stick wrote:
> 13/8 is correct, period. Show me a rollout that says otherwise and
>
> I'll show you a rollout that is wrong.
>
>

Stick,

If I get this position in a match at 2a 4a, I will play 8/3. Could you explain how I would suffer thereby? (Of course, I'm unlikely to be 2a 4a in a match since I'm generally bad enough to lose matches without reaching 2a.)

Paul Epstein

[Walt]

Paul implied that he was eager to find out the reasons why 13/8 is my
preferred choice, so here's what I said at the time, before seeing any
bot results:

"Prepare to clear the midpoint with 13/8. 65 or doubles will then allow
us to clear it safely, and we should have several rolls in which to get
"lucky".

Covering the blot on then three is probably not far behind equity wise,
but I think playing off the midpoint should be best."

In hindsight, the reason 8/3 is inferior is that it doesn't prepare to
clear the midpoint and it strips the 8. It looks good because it fills
in the three point (the next point we want to make) but that's illusory
- we probably won't have a blot to shoot at for several moves, so making
homeboard points is not worth that much.

Which neatly segues into the bot move: XG likes 8/6 4/1 - leaving 4
blots in the homeboard!

This is a bot move if ever there was a bot move. I don't usually like
these kinds of problems, where a perfectly solid and reasonable move is
edged out by .01 by an outre and strange move by the bot. All I can say
it that at least I warned everyone.

While I'll still play 13/8 in this position, the bot choice is
instructive in that it points out that making homeboard points in
positions like this is not worth as much as usual. See the related
position below for more.

XGID=--AABbCBC---bC----accbb---:0:0:1:32:3:1:0:5:3

X:X O:O
Score is X:3 O:1 5 pt.(s) match.
+13-14-15-16-17-18------19-20-21-22-23-24-+
| X O | | O O O O |
| X | | O O O O |
| X | | O O |
| | | |
| | | |
| |BAR| |
| | | |
| | | |
| X | | X |
| O X X | | X O X |
| O X X | | X O X X X |
+12-11-10--9--8--7-------6--5--4--3--2--1-+
Pip count X: 108 O: 120 X-O: 3-1/5
Cube: 1
X to play 32

1. Rollout� 8/6 4/1 eq:+0.522
Player: 68.76% (G:3.41% B:0.21%)
Opponent: 31.24% (G:1.33% B:0.04%)
Confidence: �0.008 (+0.514..+0.530) - [97.5%]
Duration: 9 minutes 58 seconds

2. Rollout� 13/8 eq:+0.511 (-0.011)
Player: 68.38% (G:1.67% B:0.07%)
Opponent: 31.62% (G:0.91% B:0.01%)
Confidence: �0.008 (+0.503..+0.518) - [2.4%]
Duration: 9 minutes 19 seconds

3. Rollout� 6/1 eq:+0.503 (-0.019)
Player: 68.08% (G:3.74% B:0.23%)
Opponent: 31.92% (G:1.48% B:0.03%)
Confidence: �0.009 (+0.494..+0.512) - [0.1%]
Duration: 9 minutes 25 seconds

4. Rollout� 8/3 eq:+0.492 (-0.030)
Player: 67.84% (G:1.66% B:0.04%)
Opponent: 32.16% (G:1.04% B:0.03%)
Confidence: �0.008 (+0.485..+0.500) - [0.0%]
Duration: 8 minutes 56 seconds


� 1296 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG Roller


eXtreme Gammon Version: 2.10, MET: Kazaross XG2



Here's the previous play. If you are tempted by the the homeboard-point
making 13/10(3) 6/3 play, please look at the 2nd and 3rd best moves,
and note how much better they are than 13/10(3) 6/3.


XGID=--AABbC-C---cE----acbbb---:0:0:1:33:3:1:0:5:10

X:Player 1 O:Player 2

Score is X:3 O:1 5 pt.(s) match.
+13-14-15-16-17-18------19-20-21-22-23-24-+
| X O | | O O O O |
| X | | O O O O |
| X | | O |

| X | | |
| X | | |
| |BAR| |
| | | |
| | | |
| O X | | X |
| O X | | X O X |
| O X | | X O X X X |
+12-11-10--9--8--7-------6--5--4--3--2--1-+
Pip count X: 120 O: 128 X-O: 3-1/5
Cube: 1
X to play 33

1. XG Roller++ 13/7(2) eq:+0.581
Player: 70.65% (G:2.04% B:0.05%)
Opponent: 29.35% (G:1.21% B:0.02%)

2. XG Roller++ 13/1 eq:+0.503 (-0.078)
Player: 68.16% (G:2.80% B:0.04%)
Opponent: 31.84% (G:1.89% B:0.03%)

3. XG Roller++ 13/10(3) 4/1 eq:+0.486 (-0.095)
Player: 67.72% (G:2.39% B:0.04%)
Opponent: 32.28% (G:2.44% B:0.05%)

4. XG Roller++ 13/4 6/3 eq:+0.464 (-0.117)
Player: 66.85% (G:1.57% B:0.02%)
Opponent: 33.15% (G:1.58% B:0.02%)

5. XG Roller++ 13/10(3) 6/3 eq:+0.456 (-0.125)
Player: 66.52% (G:1.78% B:0.03%)
Opponent: 33.48% (G:2.33% B:0.03%)


eXtreme Gammon Version: 2.10, MET: Kazaross XG2

[peps...@gmail.com]

> 1. Rollout� 8/6 4/1 eq:+0.522

>
> Player: 68.76% (G:3.41% B:0.21%)
>
> Opponent: 31.24% (G:1.33% B:0.04%)
>

> Confidence: �0.008 (+0.514..+0.530) - [97.5%]

>
> Duration: 9 minutes 58 seconds
>
>
>

> 2. Rollout� 13/8 eq:+0.511 (-0.011)

>
> Player: 68.38% (G:1.67% B:0.07%)
>
> Opponent: 31.62% (G:0.91% B:0.01%)
>

> Confidence: �0.008 (+0.503..+0.518) - [2.4%]

>
> Duration: 9 minutes 19 seconds
>
>
>

> 3. Rollout� 6/1 eq:+0.503 (-0.019)

>
> Player: 68.08% (G:3.74% B:0.23%)
>
> Opponent: 31.92% (G:1.48% B:0.03%)
>

> Confidence: �0.009 (+0.494..+0.512) - [0.1%]

>
> Duration: 9 minutes 25 seconds
>
>
>

> 4. Rollout� 8/3 eq:+0.492 (-0.030)

>
> Player: 67.84% (G:1.66% B:0.04%)
>
> Opponent: 32.16% (G:1.04% B:0.03%)
>

> Confidence: �0.008 (+0.485..+0.500) - [0.0%]

>
> Duration: 8 minutes 56 seconds
>
>

Walt, you're not considering the confidence levels. The real message is that all 3 of the top choices are roughly equal. I think you're wrong to say that the bot is deciding on 13/8. Under the standard confidence levels, the equity of 13/8 could be as high as 0.518, and the equity of the rollout winner could be as low as 0.514.
Stick is not wrong, and the rollout is not wrong either. But your interpretation of the rollout seems wrong.

The big plus of 13/8 is that it gives an extra great clearing roll -- namely 65. I didn't consider this point. This dwarfs any argument for any alternative play. I think it muddies the analysis to point out that doubles also clear the midpoint because doubles clear the midpoint anyway, no matter what X does. The glory of 13/8 is that 65 provides a further road to salvation, even when the two dice persist in showing different faces to the light.

Paul Epstein

[Stick]

In this position and esp. many similar type positions it is far more
important to prepare to clear the midpoint and keep your spares on the
high points (typically 8pt and 6pt) than it is to make points behind
the anchor. The reason? You have a race lead and your opponent is
the one playing the holding game. How often do you need to have a
board? Basically never. What you need is flexibility and timing in
order for you to clear the hardest points.

Here 13/8 is better not only because it immediately prepares for the
bonus roll of 65 but also because it allows small doublets to play
better. This is one reason why you try as quickly as possible to
strip the midpoint down to two checkers when playing against a holding
game. Your double 2s and 3s are a lot stronger with only 2 checkers
there instead of 3 or 4.

Now, to your rollout and XG making plays like 6, 1. It is a donkey.
While my rollout is for money with the cube (properly) already turned,
I believe it would yield the same results as if I used your score
specific position. With your position, 3 ply thinks 13/8 is better
initially and XGR+ thinks the bizarre 6, 1 is correct. The rollout
I'm posting has much higher settings so it will keep the bot from
making more strange plays down the road. It may only be 108 games but
the confidence level is already higher than yours. I'm not willing to
waste any more time on it since I know 13/8 is right.

XGID=--AABbCBC---bC----accbb---:1:-1:1:32:0:0:3:0:3

X:Player 1 O:Player 2

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 O O | +---+
| X | | O O O O | | 2 |
| X | | O O | +---+

| | | |
| | | |
| |BAR| |
| | | |
| | | |
| X | | X |
| O X X | | X O X |
| O X X | | X O X X X |
+12-11-10--9--8--7-------6--5--4--3--2--1-+

Pip count X: 108 O: 120 X-O: 0-0
Cube: 2, O own cube
X to play 32

1. Rollout¹ 13/8 eq:+0.269
Player: 68.08% (G:0.90% B:0.02%)
Opponent: 31.92% (G:0.64% B:0.01%)
Confidence: ±0.010 (+0.259..+0.279) - [99.7%]
Duration: 54 minutes 08 seconds

2. Rollout¹ 8/6 4/1 eq:+0.248 (-0.021)
Player: 67.15% (G:1.23% B:0.03%)
Opponent: 32.85% (G:1.03% B:0.04%)
Confidence: ±0.011 (+0.237..+0.260) - [0.3%]
Duration: 1 hour 00 minute


¹ 108 Games rolled with Variance Reduction.
Dice Seed: 12215931
Moves and cube decisions: XG Roller+
Search interval: Large


eXtreme Gammon Version: 2.10

[Tim Chow]

On May 13, 11:58 am, Stick <checkmug...@yahoo.com> wrote:
> With your position, 3 ply thinks 13/8 is better
> initially and XGR+ thinks the bizarre 6, 1 is correct.  The rollout
> I'm posting has much higher settings so it will keep the bot from
> making more strange plays down the road.  It may only be 108 games but
> the confidence level is already higher than yours.

I basically agree with Stick but I have a couple of quibbles here. If
XGR+ thinks the bizarre 6, 1 is correct then there's no guarantee that
the rollout settings will *prevent* the computer from making more
strange plays down the road. One expects that strange plays will be
*reduced*, though.

Also, one should not make too much of the confidence levels when you
choose to stop the rollout so soon. You remember this example of
course:

http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=70733

http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=87889
---
Tim Chow