|
||||||
|
Title: Comments for Game 103997 Fritzlein vs bot_clueless Post by camelback on Apr 26th, 2009, 8:56am Comments from Game 103997 (http://arimaa.com/arimaa/gameroom/comments.cgi?gid=103997 ) My apologies to the spectators for a slow, boring game. I played as carefully as I could, and even then I was just lucky to reach my goal of not losing a piece. Puzzle for the spectators: What forced goal did clueless see on 57g that I didn't? Fritzlein Sat 11:11 pm GMT 57g. He he! You do not expect that rabbit! Otherwise, the blockade at the start was impressive! All the while keeping elephant and camel free; that's something to remember! Hannoskaj Sat 11:43 pm GMT Fritzlein, the forced goal candidate I suggested does not appear to be a forced goal on further examination...clueless can pull one rabbit with the horse leaving room to pull the other with the cat, and there is then no one move goal for you. I guess you are right to generally only trust goals checked by computer... Simon Sun 12:54 am GMT However, after further further examination, it seems that if this is the only immediate defence, there is a goal in two moves at that point (final win on 59g). Simon Sun 1:19 am GMT However...this goal does not involve any losses...so if it is correct, that means that you could have won in three moves where you actually won in four, and still have had no losses...unless clueless decided to take up the sacrifice instead on 57s, but that would enable goal on 59g. Simon Sun 1:24 am GMT sorry, taking the sacrifice would enable goal on 58g. Simon Sun 1:25 am GMT There's another one move defence on 58s, but that one also seems to lead to a forced goal for you with no losses on 59g. Simon Sun 1:36 am GMT grr...one move defence on 57s, I mean. Simon Sun 1:37 am GMT I've found another move for 57g that seems to lead to a forced goal on 59g. Is there any way to win faster? Simon Sun 2:27 am GMT What about advancing e3 rabbit? Eltripas Sun 2:49 am GMT That is the other move that leads to goal on 59g. Simon Sun 4:00 am GMT Simon, thanks for all the analysis. Instead of thinking for myself, I cheated and fed the position to Bomb. In 55 seconds Bomb finds that your move 57g Cf^DM<Df> is goal in three. As you say, Silver can't take the camel on 57s without allowing goal on 58g. To delay goal for one more turn, Bomb prefers 57s RHgvRChv. But then even I could find 58g DM<*Rg^< to win on 59g. Bomb initially prefers the seemingly suicidal move 57g CfvDe>Re^^. Silver can't quite goal through the lane that Gold has opened, because 57s Rd4>Rd5vRevvv is five steps. Silver's pieces are mostly frozen. Only 57s Ca>Rb>>> or 57s E>CE^ even delay the goal another turn, and neither one spares Silver a step to use on offense. So it is also a goal in three without even giving clueless the opportunity for a spite capture. I wouldn't have needed to rely on clueless to delay goal as long as possible. In other words all of your analysis is confirmed by Bomb. But the "safe" winning move looks so crazy I don't know how I ever could have played it in a live game. I keep saying how some day I might get good at goal attack and defense, but I can't imagine my calculating skills will ever be that good. Fritzlein Sun 4:07 am GMT Nice! But now I am wondering if there may have been a three move forced goal available to you at 28g (final goal at 30g) Here's my plan: 28g Eb5w rb6s Ra6e Rb6n 28s looks forced either c-> b8 d->e8 or just d->b8; doesn't matter which move occurs for later analysis 29g Ea5n Ea6e ra7s 39s I don't see any defense 30g goal? Simon Sun 4:40 am GMT "or 57s E>CE^" I must admit I didn't see that one... Simon Sun 5:03 am GMT You are right, Simon. Your 28g would have been a goal in three. For more fun goal practice, let me note that on 58g Bomb finds a different goal in three than I found. Any guesses? Also it looks like a missed a goal in three on move 41g, but Bomb is somehow unable to confirm or refute my suspicions, so I'll need some help from human brains. I'm pretty sure I missed a goal in four on 40g, but that's definitely beyond Bomb's range. And come to think of it, if there is a goal in three for me on 41g, perhaps the move I actually played was goal in four and I just followed up badly. But the goal in four from 40g seems much more forcing than my in-game move. On move 35g, when I played my lame two-stepper, Bomb finds a crushing move that is apparently also goal in three. On move 32g, when I went after Bomb's dog for the first capture of the game, I could also have had goal in three. This one I actually found myself, although I know it is hard to believe that I can do anything without Bomb. :-) That makes AT LEAST FIVE goal in three opportunities that I missed. I said in chat that my lack of goal-finding ability doesn't matter much, but this game was frankly ridiculous. I can't just keep on saying that "some day" I will get better at it. All my cunning strategy that I'm so proud of is being invalidated by my goal incompetence. Enough! From now on goal study is my priority. The only good thing about this game is that it provides plenty of unintended goal in three puzzles. I hope the community can profit from my exercise in futility. Fritzlein Sun 5:42 am GMT Incidentally, for that 57s the appropriate 58g seems to be 58g Mg6w dg7s Mf6s Df7e when I saw that (took me a little while) I wondered if it could be a third forced goal move for 57g as well...but it looks like after the double rabbit pull on 57s, gold needs to kill the dog to get the rabbit available again, and that takes 4 steps in this case, and on the next move 2 steps are needed to unfreeze the rabbit and it can't reach goal in the next 2 steps so gold can't score on 59s (if that move is used on 57g instead of 58g) Simon Sun 5:43 am GMT "For more fun goal practice, let me note that on 58g Bomb finds a different goal in three than I found. Any guesses?" push cat south with camel? I haven't proven it is goal in three though "Also it looks like a missed a goal in three on move 41g, but Bomb is somehow unable to confirm or refute my suspicions, so I'll need some help from human brains." Could you be more specific? I don't see any 3 move goals...the most promising move I see is rD< dMv, and c> d> dR< seems to stop that well enough to not have goal by 43g "I'm pretty sure I missed a goal in four on 40g" could be...after the same moves as above you can now in this case push rabbit with dog to unfreeze rabbit which unfreezes other rabbit which moves forward... eventually scoring on 43g "On move 35g, when I played my lame two-stepper, Bomb finds a crushing move that is apparently also goal in three." you mean 36g? let me guess...pushing dog north, stepping both rabbits west? or from chat: 05:20:24 Simon how about pushing dog west? 05:21:21 Simon maybe move each of the two eastern rabbit 1 step west as well I think that was the move I was commenting on... "On move 32g, when I went after Bomb's dog for the first capture of the game, I could also have had goal in three." I had some difficulty seeing this...but my best guess would be hM< R< ? Well, thanks for the goal exercises...I should have gone to bed long before now... Simon Sun 6:55 am GMT Congrats Karl! I would like to comment on your last comment about goal attacks. Missing 5 goal oppportunities in a game definitely happens to all of us all the time. I can't say I know why, but here are a few thoughts on that. I would think it's about 10 times harder to find a goal during a game than off the board with the instruction: "gold plays and goals in 3" or just "gold has a forced goal, find it". I guess it's pretty normal, as failing to find the puzzle in the required time has no consequence while using one's whole time bank to fail to find a goal definitely reduces the winning chances by an unbearable amount. This said, I think that it might be interesting to force oneself into the right state of mind to look for a goal. Whenever we smell goal opportunities, maybe the right thing is to convince ourselves: "there must be a forced goal here", and search a little bit with that thing in mind, say during 30 or 60 seconds in 2mn games for instance. This is something I think I do more or less unconsciously sometimes, but I will try to do it regularly. The second point I can see is that most of the goals we miss are actually moves we didn't even consider, or just barely. When someone after the game tells us: "that move is a goal in 3", it usually takes us very little time to get convinced of it, compared to the time we would have needed to find it in front of the board. Sometimes we actually had considered it, but failed to work it out properly, like the goal in 3 you and Simon told me about in my first challenge game, and we cannot do anything to avoid that, but I don't think it's the most common case. Admittedly we cannot compete with the exhaustive or even quiescence move generators of bots, but maybe we should try to review first all the possibly interesting moves with respect to goal. I know I usually consider rather few candidate moves for each move, and I think that's a serious leak, probably much bigger at arimaa than at other lesser branching factor strategy games. In my last challenge game, I think I thought for about 6 minutes on the first move of the goal sequence, and I actually just concentrated on 2 or 3 moves during maybe the first 2 or 3 minutes, and then maybe 2 others in the next 1 or 2 minutes before considering the correct 37s which is definitely the most logical move with respect to goal attack, one I think I would have envisioned much much more quickly if someone had told me: "hey, watch out, you have a forced goal here!". Just trying to make the first move(s) I consider work out (strategically speaking) would definitely have prevented me from finding the correct move had I not had a full time reserve, or just if that first line had made me win a rabbit or whatever, as I would have played it then. Trying to enumerate the most logical candidates for goal attack is something I need to try as well in long games. I am not a strategy games psychologist, so maybe all the above is trash, but anyway I'm looking to hear how others concretely organize their reflexion in front of the board. It looks like a very complicated matter, the same player in front of the same position with same state of mind would probably come up with different moves almost all the time if we could repeat twice the same experiment. That's something a bit frightening at a start of the game, we know that all our strategical knowledge is here, but the reason why during the game we will find bad or good moves are very elusives. I guess that remembering that the key word in "strategy game" is "game" helps not to be too concerned with how our brain will handle the game at the start of it even though it's a mysterious matter :-) On a different matter, now the challenge is over. It was a bit tougher than expected after the qualifiers I guess, maybe just as tough than expected after the CC. In any case jdb and other programmers have done good, bots have cerainly become stronger. I like how clueless fights for trap control when one of its piece gets hostaged, I think trap control fighting is the most important part of the game, and clueless has definitely a lot of good in it with that respect. This sais, it still has a lot of weaknesses. I think giving hostages for free is one of them, if clueless fought against them, making its opponent pay for the hostage, say at least a rabbit for any hostage, it would get much tougher to get out it in good shape. I think I will make a post in the forum as for several moves where I think clueless showed weakness, even though I guess it won't make jdb's job of improving clueless much easier. Jean chessandgo Sun 9:34 am GMT Thanks for the comments, Jean. I was upset with my play last night, but this morning I realize that whatever I don't know is a learning opportunity. Learning is fun. If I already knew everything I was going to know about Arimaa, it would soon get boring. I should be glad that my game has obvious holes and major possible improvements. Like you, I tend to consider only a very small set of moves on each turn. In that context, isn't it remarkable that we can play relatively well? Something about our training must guide us in selecting moves to consider, so that we usually look at strong moves instead of weak ones. That is something we have learned to do from experience and study. Like you, I focus a great deal on trap control. But even when I restrict my attention to trap control, there would be hundreds of potential moves, and I don't even look at most of them. Something about our trap control training helps us find the relevant moves, and tells us who is going to win or lose most trap control fights without even calculating ahead. Therefore, the question is whether goal-finding training could do the same thing. Is goal attack and defense fundamentally harder than trap control, or is it merely something I have paid less attention to so far, so that my instincts are not yet developed? Will I some day be able to just see "there's a forced goal there somewhere", as I now can see "I ought to be able to take over that trap"? Or will there be no markers, so that I can't instantly guess what move forces goal and what move doesn't, and I can't tell what position has a defense and what position doesn't? Some evidence that I can train myself comes from being able to name the relevant factors. Obviously one must consider the number of defenders, number of attackers, and how close the rabbit is to goal. Less obviously, one can consider how much space there is for pushing/pulling, whether defenders can be blocked out to preserve space, whether the defenders are mere rabbits that can't push back or freeze the threatening rabbit but which can be pulled out to make space, whether a piece can stand directly behind an attacking rabbit to prevent it from being frozen or pushed back, etc. Perhaps there are enough comprehensible factors that we can look for the presence/absence of the factors without directly calculating. Probably there are more factors than I have even named, so I can get better at goal attack by learning the factors. But even if there aren't more signposts I might still train myself to better understand the relationships between existing signposts, and save computation that way. Some evidence that I can't train myself is that goal attack seems to involve a positional concession almost all of the time. It seems that to play for goal I have to weaken my own goal defense, or weaken my trap control, or give up a hostage, or create a goal threat for the opponent by pulling out their rabbits, etc. Many times in control Arimaa I play a move that doesn't do exactly what I thought it would, but which is a strong move anyway. For goal attack, in contrast, it might be that if a move doesn't win, it is weak and shouldn't be played. That would mean calculation is king and human fuzzy logic is no good. The one area where I might disagree with you Jean is that I don't think it is a leak to consider only a few moves per turn. I mean, I am not sure that you would play better if you would train yourself to consider more possibilities every move. I think the focus for training should be getting better at seeing which moves are worth looking at. Before and after your training you could look at only four moves, but after training the four you choose to look at would be a better four because of your improved instincts. (Of course, this relates to why it is easier to find a forced goal when someone tells you it is there: your instincts of which few moves to look at have been given a big helping hand that you don't have in a live game.) As for the state of the Challenge, I agree that a big reason clueless is tougher to beat this year is that it knows how to escalate the tension rather than just accepting a small loss and doing nothing about it. Even if we can still outpunch clueless, it is a step forward for the bot to force us to trade punches rather than to get it in a headlock and squeeze. Fritzlein Sun 4:00 pm GMT |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Fritzlein on Apr 26th, 2009, 9:30am Thanks for copying the discussion over here, camelback. Simon, On 58g pushing the cat back with the camel is indeed goal in three, but that's not what I had in mind. I was thinking of 58g Ra^RbE<M<. I guess the position on 58g makes a poor goal in three puzzle, because there are at least three different solutions. It reminds me of a riddle my Dad thought of as a child. It goes like this: "What's white?". For some reason the grownups he told his riddle to just couldn't get it, when obviously the answer was, "A refrigerator!" Your suggestion of 41g DgR<MDv is not goal in three due to the defense 41s Ce>D>RD<. It may be a forced goal in more than three, and it is probably stronger than the move I actually played, but it isn't the puzzle solution. The move that I think is goal in three (but which confuses Bomb for some reason) is 41g MDvEC>. After I play that move Bomb doesn't see a forced goal in two against itself at 12 steps search depth, although it should due to its goal extensions, and I am sure that the move it wants to play loses in two. Therefore I am not sure whether to trust Bomb or to trust my calculation that it is really goal in three. One move earlier, though, 40g DgR<MDv (as you suggest) does seem to work. That's the move I had in mind for goal in four, but in fact, after the forced 40s Ce>D>RD<, the move you suggest 41g D4Rf<Rg>Rh6^ actually wins in two. So apparently there was a goal in three on 40g. Yes, I meant 36g, not 35g. Your guess of 36g MD<R7<Rf< is indeed goal in three, although I had in mind 36g MD<R7<De>. I'm impressed that you found the winning move live, even though you neglected to chat the move numbers. :P Are you suggesting a three step move 32g MH<Rg<? I beleive that can be defended simply by pushing the rabbit back to the g-file with 32s De^DR>Cd>. But I think sliding both rabbits over works: 32g MH<Rg<Rh7<. If you are keeping score at home, I believe I missed goal in three on six occasions. 28g RbE<Ra>^ 32g MH<Rg<Rh7< 36g MD<R7<Rf< 40g DgR<MDv 41g MDvEC> 57g CfvDe>Re^^ The fact that the goal in three on some of those moves is not unique makes it all the more embarrassing. I should publish a goal-in-three collection based only on this one game. Before I do that, however, can anyone find a seventh move on which I missed goal in three? :-) |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on Apr 26th, 2009, 1:33pm Quote:
that's what I said...I mentioned that move because I couldn't see a better one... but Quote:
... but I currently don't see how to force goal by 43g after 41s cc6n cc7w ce6e dg6e...suggest a move for 42g if you disagree... Quote:
Yes, I missed that...my excuse is going to be that it's counterintuitive to move the rabbit when you're going to have to move it right back again the next move... |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Fritzlein on Apr 26th, 2009, 1:55pm on 04/26/09 at 13:33:41, Simon wrote:
You're right, I can't goal by 43g, but how about 42g M<RhDgvR7^ to goal right away? ;) |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on Apr 26th, 2009, 2:08pm yes...that's quite a blind spot I have...come to think of it, when we were in chat after the game I think at one point thought a very similar move would not work due to the rabbit being frozen, not realizing the dog unfreezes it at the same time it pulls the rabbit. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on Apr 26th, 2009, 2:43pm And I am now fairly confident that there is indeed a forced goal in 3 moves on 41g. If silver spends less than 2 steps in the west on 41s, pulling the a7 rabbit and stepping the a6 rabbit forward looks like it will lead to a win on 43g. So in order to prevent a goal by 43g, silver has a budget of only 2 steps in the east. and in order to prevent the pattern you just pointed out to me, that can only be moving the cat or the horse to f7. In the west it looks like it has to move cat to b7. Now, on move 42g, gold pushes rabbit west with dog and cat east with elephant. In order for silver to stop the threat in the west, it needs 2 steps, and in the east it needs at least 3 (if the horse is on f7) or 4 (if the cat is there). Of course, my analysis may not be very credible if I can miss a 1 move goal after seriously trying to analyze a position... |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Ice on May 14th, 2009, 12:54pm I figured I'd post this here too. If you ran the camel all the way over to the Elephant' side and froze the cat and two rabbits, isn't there a forced goal in 1 move there then? |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Ice on May 14th, 2009, 12:54pm Sorry, move 29W is the spot where I am talking. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Fritzlein on May 14th, 2009, 1:24pm on 05/14/09 at 12:54:11, Ice wrote:
For all the goals I missed in this game, your suggestion is actually one move I considered at the time. Unfortunately, the immediate goal threat is stopped by moving the silver dog to c7. If there is no empty square next to the gold camel, then the gold camel can't make a pull. Thus there is no goal in two from 29g, at least as far as I can see. It is possible that your 29g is the start of a forced goal in more than three moves, but I don't see it right off the bat. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by aaaa on May 14th, 2009, 1:29pm My bot discovered that 29g Me7w De4n Md7w Mc7w is goal in three. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Fritzlein on May 14th, 2009, 1:49pm on 05/14/09 at 13:29:48, aaaa wrote:
Wow, that auto-generated goal-finding code is really paying off! The De4n step (to prevent the f6-rabbit from crossing over and aiding the defense) is fiendishly clever. So, let me update the list of forced goal in three that I missed: 28g RbE<Ra>^ 29g M<<<De^ 32g MH<Rg<Rh7< 36g MD<R7<Rf< 40g DgR<MDv 41g MDvEC> 57g CfvDe>Re^^ |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on May 14th, 2009, 3:01pm Maybe I'm just obtuse, but I haven't seen a way to score a goal by 31g for that 29g. If the goal occurs after 31g, then it's not a missed three move forced goal according to the definition being used in the thread. If it is a goal by 31g, could you help me see it by explaining what 30g in response to 29s d->c7 x (for definiteness x = rf6w) leads to goal on 31g?. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on May 14th, 2009, 3:09pm ok, it has to be that r step, not just for definiteness, or 30g ra7n Mb7w Ma7e Ra6n would be such a move. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on May 14th, 2009, 3:11pm doh, no that would still take 5 steps to goal on 31g... |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by aaaa on May 14th, 2009, 3:22pm Would it be too much of a hint if I said that it would involve a sacrifice quite some distance away from the goal area? |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on May 14th, 2009, 3:29pm I considered that, but even after reconsidering I don't see goal by 31g, the horse stops it... I might add that it looks like with the rabbit step the horse doesn't even need to do anything. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on May 14th, 2009, 3:36pm Heh, now I see it. Thanks. edit: I should have thought more carefully about what could be achieved with fewer than four steps. I was lazy and overconfident in my earlier skepticism... (to avoid confusion - it does require all 4, but two of them establish the threat, the other two block two defences; but each of those defences would have cost all 4 steps even to stop the threat established by the first 2, so I should have looked more carefully at what could have been accomplished with those 2 steps plus some other 2 steps (in fact, I don't think I even noticed that those 2 steps could establish a threat, because I was so focused on 4 step moves. Oops.)) |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by aaaa on May 15th, 2009, 8:24am My bot discovered another goal-in-three on move 29g. Can anyone find it? |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by aaaa on May 15th, 2009, 1:24pm on 05/14/09 at 13:49:33, Fritzlein wrote:
About half a day of computer analysis proves that these are all of Gold's missed goal-in-three opportunities. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Fritzlein on May 15th, 2009, 2:55pm Whew, I'm glad that I only missed seven goal-in-three opportunities. Eight in one game would have been really embarrassing! :-[ |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by aaaa on May 17th, 2009, 11:45am on 05/15/09 at 08:24:08, aaaa wrote:
The answer is to trade a tempo by having the camel pull the cat in the process of going to b7. |
||||||
|
Title: Re: Comments for Game 103997 Fritzlein vs bot_clue Post by Simon on Aug 3rd, 2009, 12:09am I never did figure out how this works. 29g Me7w cf7w Md7w Mc7w 29s ce7e de6n de7w dd7w 29g? So it seems gold needs 2 steps to clear b6, then in the remaining steps can't block both the rabbit and camel from reoccupying it. What am i missing? edit: duh, now I see it... |
||||||
|
Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1! YaBB © 2000-2003. All Rights Reserved. |