Backgames
Laury Chizlett
noticed a strong reluctance for it to get into a backgame. (I mean here
a strict backgame, rather than a holding one: holding two or more
points, keeping good timing, while building up a strong forward
position, without making inner-board points, and waiting until your
opponent is home before hitting.) The general advice, which I try to
follow, is to only accept such a strategy as a last resort when there is
no other defence. They are so much fun, however, that I suspect I am not
alone in accepting a full backgame rather sooner than I should. But the
acceptance point of Jellyfish seems vanishingly low.
Magriel says of a position demonstrating the backgame that (I'm quoting
from memory) " ... at this point it would be a mistake to adopt any
other strategy". One of Robertie's books is almost totally made up of
backgames. I remember one game on FIBS where I had a backgame going and
we ended up collaborating, discussing each move, helping each other out,
totally fascinated by the varying problems each throw gave us. It was
the most enjoyable game I've had there! All this seems to make backgames
an enjoyable and acceptable strategy to humans - and yet the bots hate
them. Indeed their play against them plays into the backgame player's
hands, hitting men regularly and allowing recycling for no reason that I
can see. Is it the way they are trained?
Given that we all trust Jellyfish and Snowie rollouts, does this mean
the backgame is dead; a faulty strategy only kept alive in human v human
games by unconcious collusion between human players?
--
Laury Chizlett
TRP, 35 Colvestone Crescent, London, E8 2LG
Tel: 020 7923 0244 Fax: 020 7923 1471 la...@trpdata.demon.co.uk
Bob Meixner
Some players enjoy them so much that they steer to them at every chance.
When they win it is pretty; but the number of gammons(and backgammons) they
lose more than offset the occasional glorious win. As Robertie puts it...
"Back games are the goal-line stands of backgammon. A well-executed
goal-line stand may indeed save a touchdown. But a team that plays a game
from the shadow of its goalpost is bound to be a losing team." As you point
out in your post they are a last resort.
BTW, bots(especially Jellyfish) are not especially strong at backgames. I
suspect the complexity of the positions does not lend itself well to neural
nets. In fact, I believe the "use caution" option on JF is there because
some experts were able to run the cube to extremely high values against JF
and win large point swings because JF mis-evaluated back game equities.
Don't get me wrong. There are good backgame situations but they are
relatively rare, Imo.
Bob(5K_Guy).
"Laury Chizlett" <la...@trpdata.demon.co.uk> wrote in message
news:UfeTHRAg7z25Ew$4...@trpdata.demon.co.uk...
Daniel Murphy
<la...@trpdata.demon.co.uk> wrote:
>Given that we all trust Jellyfish and Snowie rollouts, does this mean
>the backgame is dead; a faulty strategy only kept alive in human v human
>games by unconcious collusion between human players?
Your "given" is disputable: we don't unquestioningly trust computer
rollouts, especially early game rollouts, prime vs. prime positions,
positions with many checkers back for one or both sides and deep
holding games because we know that JellyFish and Snowie tend to
undervalue the underdog's position and often misplay both sides.
The bots' reluctance to play backgames is well founded. Backgames
should be a last resort -- they lose gammons and backgammons -- and a
commmon human error is to reduce one's options by prematurely choosing
a backgame strategy. But the ability to play both sides of a backgame
should be in every player's bag of tricks. Rarely enough is a backgame
strategy appealing because it wins the most, but often enough in the
course of a game it becomes the strategy that loses the least ...
... or better, even, since many players today do not know how to
proceed against a backgame. Indeed, I have in mind an experienced
player who enjoys success against markedly inferior opposition by
intentionally steering -- not necessarily for a backgame -- but for
positions in which one or both players have many checkers back (he
doesn't much care which) -- in which his superior skill makes up for
the horrendous "blunders" he has to make to get to those positions.
Backgames >are< fun, and sometimes profitable ... What do you think of
this position, and computer evaluations?
O on roll, cube action
+-1--2--3--4--5--6--------7--8--9-10-11-12-+
| X X X O O | | O X |
| X X X O O | | O X |
| X X O O | | O X | S
| | | | n
| | | | o
| |BAR| | w
| | | | i
| | | O | e
| | | O |
| O X | | X O |
| O X | | X O |
+24-23-22-21-20-19-------18-17-16-15-14-13-+
Pipcount X: 251 O: 149 X-O: 0-0/Money (1)
CubeValue: 1
Evaluations below:
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
v
Snowie says:
3-Ply Money equity: 0.602
2.5% 32.1% 65.3% 34.7% 4.8% 0.1%
1. Double, take 0.951
2. No double 0.875 (-0.076)
3. Double, pass 1.000 (+0.049)
Proper cube action: Double, take
JellyFish says:
Level 7 Money equity: 0.548
3.7% 33.1% 61.3% 38.7% 4.3% 0.2%
Proper cube action: No double, take
Douglas Zare
> I've played hundreds of matches with Jellyfish and many times I've
> noticed a strong reluctance for it to get into a backgame. (I mean here
> a strict backgame, rather than a holding one: holding two or more
> points, keeping good timing, while building up a strong forward
> position, without making inner-board points, and waiting until your
> opponent is home before hitting.) The general advice, which I try to
> follow, is to only accept such a strategy as a last resort when there is
> no other defence. They are so much fun, however, that I suspect I am not
> alone in accepting a full backgame rather sooner than I should. But the
> acceptance point of Jellyfish seems vanishingly low.
First, of course you might want to play for the enjoyment rather than for
the win, while practicing. Second, it may be that avoiding a backgame is
JF's best strategy, but that you should aim to play better than JF, since
you can appreciate timing, the importance of shaking loose a second checker,
and the strength of an outfield prime.
> Magriel says of a position demonstrating the backgame that (I'm quoting
> from memory) " ... at this point it would be a mistake to adopt any
> other strategy". One of Robertie's books is almost totally made up of
> backgames. I remember one game on FIBS where I had a backgame going and
> we ended up collaborating, discussing each move, helping each other out,
> totally fascinated by the varying problems each throw gave us. It was
> the most enjoyable game I've had there! All this seems to make backgames
> an enjoyable and acceptable strategy to humans - and yet the bots hate
> them. Indeed their play against them plays into the backgame player's
> hands, hitting men regularly and allowing recycling for no reason that I
> can see. Is it the way they are trained?
Their evaluations are far from the rollouts. There was a very interesting
contest in Kit Woolsey's GammOnLine in which one tried to construct the
worst possible move in cubeless backgammon. I think the best construction
was evaluated as a 3.8 point blunder, but the Snowie rollouts of similar
positions were about a point worse. I believe that part of this discrepancy
was because Snowie was not acting to save the backgammon properly in the
rollout. I don't know whether this indicates a flaw in its training.
> Given that we all trust Jellyfish and Snowie rollouts,
I don't. They need to play equally well on both sides of a backgame for the
rollouts to be reliable.
> does this mean
> the backgame is dead; a faulty strategy only kept alive in human v human
> games by unconcious collusion between human players?
Match play is much different from money play in that one often does not care
about being gammoned or backgammoned. Backgames are much more reasonable
then. (I was surprised to see JF play backgames against me in 1-point
matches.) In money play, backgames should arise but less frequently than one
might like.
Douglas Zare
Christian Decroze
the men are :-(
Daniel Murphy <rac...@best.com> a écrit dans le message :
39db5a69...@news.cybercity.dk...> >Given that we all trust Jellyfish and Snowie rollouts, does this mean
> >the backgame is dead; a faulty strategy only kept alive in human v human
> >games by unconcious collusion between human players?
>
Laury Chizlett
>>the backgame is dead; a faulty strategy only kept alive in human v human
>>games by unconcious collusion between human players?
>
>Your "given" is disputable: we don't unquestioningly trust computer
>rollouts, especially early game rollouts, prime vs. prime positions,
>positions with many checkers back for one or both sides and deep
>holding games because we know that JellyFish and Snowie tend to
>undervalue the underdog's position and often misplay both sides.
>
It is good to read this. I do not have the full version of Jellyfish and
am not able to do rollouts, but am interested to know how inclusive the
"we" in the above is: often in this newsgroup the rollout is taken as
the final word, like a stern father ruling over the antics of children
who should know better.
<snip>
>Backgames >are< fun, and sometimes profitable ... What do you think of
>this position, and computer evaluations?
>
> O on roll, cube action
> +-1--2--3--4--5--6--------7--8--9-10-11-12-+
> | X X X O O | | O X |
> | X X X O O | | O X |
> | X X O O | | O X | S
> | | | | n
> | | | | o
> | |BAR| | w
> | | | | i
> | | | O | e
> | | | O |
> | O X | | X O |
> | O X | | X O |
> +24-23-22-21-20-19-------18-17-16-15-14-13-+
> Pipcount X: 251 O: 149 X-O: 0-0/Money (1)
> CubeValue: 1
>
>Evaluations below:
>Snowie says:
>
> 3-Ply Money equity: 0.602
> 2.5% 32.1% 65.3% 34.7% 4.8% 0.1%
> 1. Double, take 0.951
> 2. No double 0.875 (-0.076)
> 3. Double, pass 1.000 (+0.049)
> Proper cube action: Double, take
>
>JellyFish says:
>
> Level 7 Money equity: 0.548
> 3.7% 33.1% 61.3% 38.7% 4.3% 0.2%
> Proper cube action: No double, take
As I say I am unused to rollouts, but is the bots consensus that roughly
2/3 of the time they will win and 1/3 of those games will be gammons? I
am an intermediate player, but I suspect these are the wrong way round!
Aons of timing, 3 anchors! I've played the game for 40-odd years but
for 38 of those have been unfamiliar with doubling and am really finding
it difficult to adjust, but if someone doubled me in the above position,
I would beaver like a shot. If otters, racoons and the like were being
played I imagine we would soon run out of small furry rodents. Would I
be wrong? And who's to tell me I'd be wrong?
André Nicoulin
those types of positions, where rollout settings might have to be tuned
(3ply, cubeless, not truncated, etc), and more CPU time might have to be
devoted to the rollout, such that you get a meaningful answer.
I did already post on this aspect (how to set up the rollout settings in
Snowie) some weeks ago. www.deja.com ?
André Nicoulin, Oasya SA
"Laury Chizlett" <la...@trpdata.demon.co.uk> a écrit dans le message news:
D4mmOrAF...@trpdata.demon.co.uk...
JP White
> Sory Daniel, the graphic is so ugly that it's impossible to understand where
> the men are :-(
Everything looks very pretty from here. I believe you may have a problem
displaying the board graphics correctly. It is necessary to use a non
proportional (fixed width) font to view diagrams here.
JP
--
JP White
mailto:jpwh...@bellsouth.net
Daniel Murphy
<chde...@access.ch> wrote:
>Sory Daniel, the graphic is so ugly that it's impossible to understand where
>the men are :-(
Most Frequently Asked Question: How can I get the boards in the
messages in this newgroup to look right?
Use a fixed width font, such as Courier, for displaying messages.