Arimaa Forum (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
Team Games >> 2007 One vs TheMob >> Move 43
(Message started by: Janzert on Oct 14th, 2008, 10:09am)

Title: Move 43
Post by Janzert on Oct 14th, 2008, 10:09am
Chessandgo responds with:

43g Ee3n rf4n Ee4e Rg1w

Interestingly this is the move OpFor expected. :)

Janzert

Title: Re: Move 43
Post by The_Jeh on Oct 14th, 2008, 10:42am
I'm getting "Error=Could not validate the Session ID provided". And coincidentally, there are 26 open games in the gameroom. Hopefully, it will clear up.

Title: Re: Move 43
Post by Fritzlein on Oct 14th, 2008, 10:46am
Also the move I expected.  In my brief look last move, 43s re5s cd6s cd5e re4s looked best.

Title: Re: Move 43
Post by Soter on Oct 14th, 2008, 11:06am
The Jeh, I get this message too.

Fritz, I agree with your proposal - at the moment I can't find anything better.

Title: Re: Move 43
Post by Soter on Oct 14th, 2008, 11:54am
If this 43s lands on the board, how do we counter 44g Rf1n Ef4s re3n Ef3w?

Title: Re: Move 43
Post by warren on Oct 14th, 2008, 6:29pm
Up to 10 ply bomb wanted to kill the horse, but at ply 10 it switched to Fritz's 43s. Bomb also suggests Soter's 44g. Here's Bomb's continutation:

43b re5s cd6s cd5e re4s
44w Rf1n Ef4s re3n Ef3w
44b Hg5n eg4n ce5w rf5s
45w Db6n cc7s cc6x Db7e Rc2w
45b Hg6w Hf6x eg5n ra8s ra7s
46w Dc7w Db7s rb5s Db6s

The resulting position leaves us each with a dog and a cat, us with 7-8 rabbits and him with 6. He has emerging goal threats in the northwest, so it's not as wonderful of a position as we expected. I don't see how to get us an immediate goal threat, so I think it's a mistake to abandon the north-west trap. How about playing a little more defensive and materially. Here's a new 43b and Bomb's continuation:

43b rh7w rg7w Ch3s dg3e
44w rf5n Ef4n Ef5s Rb1n
44b re5s rf6s Hg5n eg4n
45w Ra5n Rf1e Rc2e Rd2e
45b Hg6w Hf6x eg5n eg6w ef6w

and we win the horse for a rabbit or so. If he tries hard enough he can still capture our dog, but with a nice position for us:

43b rh7w rg7w Ch3s dg3e
44w Rf1n Ef4s Ef3e Rd4e
44b re5w rd5s rd4s rd3s
45w Rb1e Rc1e Rf2w Re4n
45b Hg5n eg4n Hg6w Hf6x eg5n
46w Eg3s dh3w dg3w df3x Eg2n
46b eg6s eg5s eg4w ef4w

My current favorite move is this 43b.

Title: Re: Move 43
Post by Fritzlein on Oct 14th, 2008, 9:02pm

on 10/14/08 at 11:54:02, Soter wrote:
If this 43s lands on the board, how do we counter 44g Rf1n Ef4s re3n Ef3w?

Bomb continues to overvalue the horse. Playing it out against Bomb thinking to a depth of 12 steps (i.e. BombP3) the game continues

43s re5s cd6s cd5e re4s
44g Rf1n Ef4s re3n Ef3w
44s re4e rf4s rf5s ce5w
45g Rc2e Rd2e Rf2e Re2e
45s Rg2e dg3s Rf2w dg2w
46g Rh2w Rh1w Rg1w Re2w(?)
46s df2w rf3s eg4s Hg5s
47g de2s Ee3s rf2n Ee2e
47s rf4w re4s re3s rf3w
48g Rd2n re2w Ef2w Rb1n
48s Rd4w cd5s Rd3w cd4s
49g rd2w Ee2w Rb2n Ra5n
49s de1n de2e re3s re2s

Forcing Bomb to take our dog hostage on 46g produced the continuation

46g Re2w Ee3s df2e Ee2e
46s eg4s rf3w rf4s cd5w, with goal to follow

Surely there are improvements for both sides, but the point about using Bomb to analyze is this: At first Bomb keeps wanting Silver to respond by taking the horse instead of continuing the attack, which may make for sub-optimal Gold defense as well as sub-optimal suggestions for Silver. Only when forced goal creeps within its horizon does Bomb finally understand the power of the Silver attack.

We should make very sure Gold has a sound goal defense before we reject 43s re5s cd6s cd5e re4s.

Title: Re: Move 43
Post by Soter on Oct 15th, 2008, 12:54am
How about letting the silicon brain show its endgame prowess and analysing the position 28 steps deep? ( I thought about doing it myself but at present I don't have access to my Bomb >:() . I am fully aware that it will take a while, but if the game tree obscures some nasty tactical surprises, then we simply must find them. I'd hate to see this game in the "big reversals" category...

Title: Re: Move 43
Post by 99of9 on Oct 15th, 2008, 1:34am
at 12 ply Gnobot recommends
43s Hg5n eg4n re5s dg3s
which results in a few trades, but seems worth considering.

I'll try to run a deeper search overnight.

Title: Re: Move 43
Post by RonWeasley on Oct 15th, 2008, 5:21am

on 10/14/08 at 11:54:02, Soter wrote:
If this 43s lands on the board, how do we counter 44g Rf1n Ef4s re3n Ef3w?


So far, Soter's 44g is the best response, but still not good enough.
Here's my offering.

43s cd6s re5s cd5e re4s
44g Rf1n Ef4s re3n Ef3w
44s rf5s rf4s re4e ce5w
45g Rc2e Rd2e Rf2e Re2e
45s Hg5w eg4n Hf5n Hf6x eg5w
46g Db6w rb5n rb6n Da6e
46s Rg2e dg3s Rf2w dg2w
47g Re2w Ee3s df2e Ee2e
47s Rd4s cd5s Rd3w Rc3x cd4s
48g Rb1e Rc1e Rd2e Ra5e
48s rf3w rf4w re4w rd4w
49g Rb5e Db6s Db5s Ch3w
49s ef5w ee5w Rc5n Rc6x ed5w
50g Db4s Rd1w Rc1n Rh2n
50s rc4s ec5s cd3s rb7s
51g Ef2s Ef1w Ee1w Rh1w
51s Cg3w Cf3x dg2n re3e rf3s
52g Ed1e Re2n Ee1n Rg1w
52s dg3w rf2e df3x rg2s

The point is that if gold defends, we can take the horse on move 45 without penalty.  Gold can't take the c7 cat because of our b5 rabbit.  After we take the horse, we have a free elephant to use.

Title: Re: Move 43
Post by warren on Oct 15th, 2008, 6:13am
Fritz and RW have convinced me that bomb is misleading in this position so I'll do future analyses without bomb. I tentatively support Fritz's 43s.

Title: Re: Move 43
Post by Fritzlein on Oct 15th, 2008, 7:05am

on 10/15/08 at 06:13:41, warren wrote:
Fritz and RW have convinced me that bomb is misleading in this position so I'll do future analyses without bomb. I tentatively support Fritz's 43s.

Sorry, I didn't mean to say you need to turn off Bomb entirely, just that it isn't completely trustworthy in balancing immediate material gain against strategic goal attack.  I've been glad for all your analysis with Bomb because for the most part it has allowed me to turn off Bomb and just think, knowing that you are covering the tactical blunder angle.

I feel it is more interesting and fun to try to analyze unassisted, but I have to admit that cyborg (human+bot) analysis is probably stronger.

Title: Re: Move 43
Post by mistre on Oct 15th, 2008, 9:10am
I support 43s re5s cd6s cd5e re4s


Title: Re: Move 43
Post by Janzert on Oct 15th, 2008, 9:19am

on 10/15/08 at 00:54:40, Soter wrote:
How about letting the silicon brain show its endgame prowess and analysing the position 28 steps deep? ( I thought about doing it myself but at present I don't have access to my Bomb >:() . I am fully aware that it will take a while, but if the game tree obscures some nasty tactical surprises, then we simply must find them. I'd hate to see this game in the "big reversals" category...


The problem is "a while" is probably longer than we have time to move in. It's the curse of the exponential. :)

Here is a list of times for opfor searching from the current position.

depth 4
time 1
depth 5
time 1
depth 6
time 1
depth 7
time 1
depth 8
time 2
depth 9
time 6
depth 10
time 11
depth 11
time 16
depth 12
time 59
depth 13
time 144
depth 14
time 1299
depth 15
time 2849
...
depth 17
time 29805
losing_moves 1336
score 424
pv re5s Hg5n eg4n dg3s w rf5n rf6x Ef4n Ef5s Rh1n b Hg6w Hf6x eg5n Rh2s dg2e w re4s Ef4w re3e rf3x
depth 17+
time 43383
losing_moves 1336
depth_searched 1
score 453
pv re5s Hg5n eg4n dg3s w rf5n rf6x Ef4n Ef5s Rf1w b Hg6w Hf6x eg5n cd6s dg2n w rb5e Db6s rc5s

Depth 17 and 17+ are from a previous run I did but didn't save the early results from. Times are usually within a percent or two of each other though.

The depth_searched value is a badly named reference to how many root moves have been searched at the current depth. As you can see it takes over 30% of the total time just to search the first move at depth 18.

Btw, the move OpFor likes here (43s re5s Hg5n eg4n dg3s) it sticks with from depth 14 through 17 at least.

Janzert

Title: Re: Move 43
Post by Fritzlein on Oct 15th, 2008, 9:48am
Let's see, if it takes 1 minute to search 3 ply and the branching factor is about 2^14, and searching an extra ply increases nodes searched by an average factor of sqrt(b.f.), then searching 7 ply as Soter suggests would take about 2^28 minutes, or a bit more than 510 years. How are we doing on reserve time? ;D


on 10/15/08 at 01:34:35, 99of9 wrote:
at 12 ply Gnobot recommends
43s Hg5n eg4n re5s dg3s

It seems GnoBot and OpFor are in agreement.  If Gold answers with rf5n rf6x Ef4n Ef5s Ch3w, is there a devastating continuation, or are the bots pleased to trade two rabbits for the horse? If OpFor likes the H for RR trade, this may be a case of FAME under-valuing rabbits, but I don't know that GnoBot would think the same way.

Title: Re: Move 43
Post by Janzert on Oct 15th, 2008, 11:21am
Yes, basically OpFor is happy with the H for rr trade.

depth 13
time 2062
losing_moves 2236
score 551
pv Hg6w Hf6x eg5n dg2e dh2n w re4s Ef4w re3e rf3x Ee4s b Cg3n dh3w Cg4n dg3n w Rf1e

It's not just FAME that likes the H for rr trade though, FAME, DAPE and DAPE(eo) all like it. FAME gives gold +0.72 before and -3.67 after, DAPE goes from +1.02 to -5.07 and DAPE(eo) from -1.18 to -2.23. So FAME is in the middle in how much it likes it.

Janzert

Title: Re: Move 43
Post by aaaa on Oct 15th, 2008, 11:39am

on 10/15/08 at 09:48:20, Fritzlein wrote:
If OpFor likes the H for RR trade, this may be a case of FAME under-valuing rabbits, but I don't know that GnoBot would think the same way.

I don't think it's reasonable to expect a purely material evaluator not to prefer trading two rabbits for the horse here. It's up to the non-material part of the evaluation to lower the net worth of capturing the horse on account of it having been taken hostage.

Title: Re: Move 43
Post by Fritzlein on Oct 15th, 2008, 12:08pm
Right, a horse is worth more than two rabbits in general, so the material eval can't be faulted for taking the trade. The key is knowing that delaying the trade will make it instead a horse for only one rabbit, or a horse for free, or an outright goal. These things are too deep to search exhaustively in this position, and rely instead on positional judgment which is dodgy even in humans. For example, two moves ago I was willing to trade our cat to take the horse, until camelback's rabbit advance persuaded me otherwise.

Title: Re: Move 43
Post by Fritzlein on Oct 15th, 2008, 12:15pm

on 10/15/08 at 11:21:33, Janzert wrote:
It's not just FAME that likes the H for rr trade though, FAME, DAPE and DAPE(eo) all like it. FAME gives gold +0.72 before and -3.67 after, DAPE goes from +1.02 to -5.07 and DAPE(eo) from -1.18 to -2.23. So FAME is in the middle in how much it likes it.

Very interesting. Recalling how DAPE values the deputy (second-strongest piece) even more than FAME (e.g. M > HD as first trade), I'm no longer surprised that OpFor and GnoBot agree. Bomb wants to trade H for D instead of for RR, because Bomb's dog value is constant, i.e. it doesn't view the dog as the deputy when the horse is gone, whereas Bomb does think the rabbit value goes up for later rabbits.

Despite their differences, then, all three bots want a material win, because they don't see forced goal, and their respective static eval functions don't give the goal attack as much weight as the material gain. So the question for us voters is whether the goal attack really is worth more than the bots think, or whether the unanimous opinion of silicon is right in this case.

Title: Re: Move 43
Post by Soter on Oct 15th, 2008, 1:30pm
I should have made myself clearer, gentlemen: I just wanted to say that maybe we shouldn't move too quickly this time - and instead focus on squeezing the deepest possible analysis from our bots. Speaking fits my communication style a tad better than writing :). Anyway, some other possible reactions to 44g Rf1n Ef4s re3n Ef3w: 44s dg3s Rf2w dg2w rf5s and 44s dg3s dg2e Rf2e rf5s (the last one slightly wacky).

Title: Re: Move 43
Post by camelback on Oct 15th, 2008, 2:14pm
I support Fritz' move 43s re5s cd6s cd5e re4s

This is better than other choices mentioned for this move.

Title: Re: Move 43
Post by warren on Oct 15th, 2008, 8:44pm

on 10/15/08 at 07:05:15, Fritzlein wrote:
Sorry, I didn't mean to say you need to turn off Bomb entirely, just that it isn't completely trustworthy in balancing immediate material gain against strategic goal attack.  I've been glad for all your analysis with Bomb because for the most part it has allowed me to turn off Bomb and just think, knowing that you are covering the tactical blunder angle.

I feel it is more interesting and fun to try to analyze unassisted, but I have to admit that cyborg (human+bot) analysis is probably stronger.

Don't worry, I didn't interpret your comments as a request that I extinguish bomb. I simply decided I'd like to try making my own mistakes rather than bomb's.

Title: Re: Move 43
Post by warren on Oct 15th, 2008, 9:02pm

on 10/15/08 at 12:15:02, Fritzlein wrote:
Despite their differences, then, all three bots want a material win, because they don't see forced goal, and their respective static eval functions don't give the goal attack as much weight as the material gain.  So the question for us voters is whether the goal attack really is worth more than the bots think, or whether the unanimous opinion of silicon is right in this case.

If this were our last chance to take the horse I would say we should play it safe and take it. However unless we're missing a hail-mary 44g or 45g, we can always capture his horse a few turns from now at the cost of at most a rabbit if we later decide that the immediate goal threat won't work. So why not give the goal threat a try?

Title: Re: Move 43
Post by 99of9 on Oct 16th, 2008, 1:56am

on 10/15/08 at 01:34:35, 99of9 wrote:
at 12 ply Gnobot recommends
43s Hg5n eg4n re5s dg3s
which results in a few trades, but seems worth considering.

I'll try to run a deeper search overnight.

Gnobot sticks with this move from 8 steps until 16(+) steps.

Title: Re: Move 43
Post by 99of9 on Oct 16th, 2008, 2:16am
Out of interest, in Fritz's line, what do we play if his response is:

43s cd6e re5s ce6s re4s
44g Ef4w re3w Ee4s Rf1n

Title: Re: Move 43
Post by RonWeasley on Oct 16th, 2008, 4:13am

on 10/16/08 at 02:16:17, 99of9 wrote:
Out of interest, in Fritz's line, what do we play if his response is:

43s cd6e re5s ce6s re4s
44g Ef4w re3w Ee4s Rf1n


Here's one way to respond.  It's looking more and more like we will take the horse to free up our e.  The immediate goal threats allow us to trade H for r.

43s cd6s re5s cd5e re4s
44g Ef4w re3w Ee4s Rf1n
44s dg3s dg2e Rf2e rf5s
45g Rc2e Rd2e Re2e Ra5n (gold can't take the d3 rabbit)
45s Hg5w eg4n Hf5n Hf6x eg5w
46g Ee3e Ef3w rf4s rf3x Rg2n
46s ce5w cd5w ef5w ee5w
47g Rg3n Rg4n Rg5n Rg6n
47s cc7e cd7e rb5s rb4s
48g Db6s Db5s Ee3s rd3e
48s rb8s Rd4e ed5s ed4w
49g Ee2w Rf2w Ed2w Ec2w
49s re3w rd3s rd2s

Title: Re: Move 43
Post by warren on Oct 16th, 2008, 3:56pm
Voice vote for 43s cd6s re5s cd5e re4s?

Title: Re: Move 43
Post by UruramTururam on Oct 17th, 2008, 12:23am
Looks fine.

Title: Re: Move 43
Post by Soter on Oct 17th, 2008, 12:46am
I examined it once again and feel rather convinced; count me in.

Title: Re: Move 43
Post by 99of9 on Oct 17th, 2008, 1:17am
Ok, let's do it, I can't see any big danger.

Title: Re: Move 43
Post by arimaa_master on Oct 17th, 2008, 4:42am

My vote for that move too!

Title: Re: Move 43
Post by RonWeasley on Oct 17th, 2008, 4:52am
That seems like enough votes.

43s cd6s re5s cd5e re4s



Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1!
YaBB 2000-2003. All Rights Reserved.