The Shark -- "Ruining the game" or "Helping extend the analysis"?

I'm trying to figure out what my role should be here.

There are two things I am sure of:

1.  I should not comment on games in progress.  Neither should anyone else.

2.  I should not comment on the results of specific positions in private messages.  Only in the forum.

After those two rules, I've just been commenting on other people's comments, figuring that isn't going to ruin the 5x5 game too badly.

However, now I've gotten a request to comment more.  The upper limit of what I could do is for any given game, I could walk through each of the positions and list, for each legal move in that position, the number of replies which are wins, the number which are losses, and the number for which I do not know the outcome.

What do people think?  Would that be good for the development of Dots and Boxes?

Here is a typical sample (just one position, from the recently analyzed game Tobias Lang vs romei, and from late enough in the game that I don't think anyone would mind):

|_. Move |_. Wins |_. Losses |_. Unknown |

| E2 | 17 | 24 | |

| C4 | | 42 | |

| E4 | 21 | 20 | |

| I4 | 34 | 4 | |

| C6 | | 42 | |

| E6 | 4 | 38 | |

| G6 | 1 | 40 | |

| I6 | 1 | 40 | |

| C8 | 34 | 5 | |

| G8 | 1 | 40 | |

| D3 | 1 | 41 | |

| D5 | | 42 | |

| D9 | 34 | 5 | |

| F3 | 21 | 20 | |

| F7 | 2 | 40 | |

| F9 | 1 | 41 | |

| H3 | 34 | 4 | |

| H9 | 1 | 40 | |

| A2 | | 42 | |

| B1 | | 42 | |

| C2 | | 42 | |

| B3 | | 42 | |

| B11 | 21 | 20 | |

| A10 | 21 | 20 | |

| K10 | | 42 | |

| J11 | | 42 | |

| I10 | 1 | 41 | |

| J1 | 34 | 4 | |

| J3 | 34 | 4 | |

| A4 | | 42 | |

| A6 | | 42 | |

| A8 | 34 | 5 | |

| B5 | | 42 | |

| D11 | 34 | 5 | |

| F11 | | 42 | |

| H11 | 2 | 40 | |

| G10 | 4 | 38 | |

| K8 | 7 | 34 | |

| K6 | 3 | 39 | |

| J7 | 7 | 34 | |

| H1 | 34 | 4 | |

| F1 | 17 | 24 | |

| D1 | | 42 | |

| E2 | 17 | 24 | |

Gosh, I wish there were a preview button.  Well, here goes....

I think the first thing to bear in mind should be what the actual topic is about and if sharks insight is even wanted. Carroll created that puzzle a couple days ago and I was actually still trying to figure out if there was a winning line when you spoiled there wasn't so I was a bit sad about that one. I think puzzles should be treated differently from analysis since puzzles are there to puzzle with the main point being that human can have some fun with getting surprised by how weird the position is etc, and analysis is some kind of research.

As for analysis, where I think The_Sharks knowledge is definitely something that should be utilised, I'd say it very much depends on:

a) what's the actual move number

b) how good was the play before the move in question

c) how frequent does the position in question appear

and d) is the data given a response on a question?

a) what's the actual move number:

- I think it is not necessary to give any information if the movenumber is 18 or higher since these can be analysed within minutes

- I think you should not give any information on specific moves if the position in question is very early with like 5 or less lines played

- I think it's okay to give data if the movenumber is 13 or higher since these can be solved with the right software spending only a couple hours (so this is a good time saver).

- I think that if the movenumber is in between 5 and 13 you should decide depending on b), c) and d)

b) how good was the play before the move in question

- I think that if the play before the move in question was bad it doesn't really harm to give information since it's not a good idea to repeat the play anyways

- I think that if the play was good (and that includes play involving mistakes that are hard to see) it is better not to give too much information (the strength of the players may be a good guideline to evaluate this)

c) how frequent does the position in question appear

I think that if the position is one that appears very very frequently there shouldn't be any information given. There is certain opening patterns that lead to the same position every time e.g. http://www.trmph.com/dnb/board#5,f5f3h5g6i4g8d3 is probably the most frequently appearing position and no information should be given here.

d) is the data given a response on a question

I think that if somebody who you know you can trust asks for it, then it should probably be okay to give away some information.

Here's what I learned from playing chess. A position would come up at a chess club, or a tournament, and my peers and I would copy the game and go home and come up with our analysis of the game without program assistance. We would then meet up at the next chess club date, and go over our analysis, and go over the game with each other, as well as our notes during the game, and or about certain positions of the game. Then after we went home, we would then computer analyze the game over the questions we came up with together and the positions we were most curious about, or uncertain about. Then the ensuing week, discuss the computer analysis, and understand each of the programs moves and trees it came up with. That's an order in which I would suggest a forum work. We start with a game, and go over the game the first time, with our personal analysis of the game, and the positions brought up during the game that grab our attention. Then, we can do a cross analysis, answering questions as a group, and come up with positions that we agree we are most uncertain of, and then once we've finished our human analysis, we can have an appropriate computer analysis to analyze the positions we have questions over. This not only allows us to use our brains for problem solving, we also get more familiar with a certain position. I'm sure anyone knows what its like to play a game to lets say move 16, and then while at work, or a restaurant, play the game perfectly to the position you are at, and analyze the game from there. These types of in depth analysis' help you memorize games, and positions. The shark should be asked for its wisdom, not try to give it so quickly!:D

Sounds good.  I don't mind waiting, and I don't mind sharing.  However, I think I'd like to hear from more people before I decide exactly what my policy should be.

I think purgency's suggestion is pretty elaborate. Does this mean I can always ask for a specific position in the forum and Shark will tell me the truths immediately (if it's not an ongoing game of course)?

This wouldn't interfere with Sean's appeal as it's only meant for analyzing existing games.

No game has ever managed anything more clever than "honor system for correspondence, do your best to catch cheats in tournaments". Anything else is too complicated, and will be ignored.

I would prefer if sharks analysis about games less than 10 moves in is not shared. I am certainly not angry if it is, but it would be my preferred scenario.

I don't mind what you proposed to give the number of winning/losing/drawing moves for any position.

But giving the winning moves for moves earlier than 14 might spoil the fun.

Okay Carroll, then I'll ask about the number of winning/losing (there are no drawing moves and 4x4 database is accessible for everyone) moves for every single follow-up position of that specific

Bump.  Let's talk about this.  I'd hate to see this resource completely unused, but I also don't want to ruin the game for everyone.

Thoughts?

I'm way too curious now. Let's consider publishing a good part of the database and move on to 7x7 as standard size?

It might be only a matter of time until others solve 5x5 too.

The game might not even be ruined as there are too many openings to memorize them all (obviously if we're allowed to consult a public database in turn-base play is a different question).

In some positions I played, I was really interested in knowing if I'd chosen a winning move, or the optimal one, or even if there was a winning move at all (who hasn't been there, huh?). Once a game is over, I think there's no harm in giving answers unless they are about the first few moves (purgency made a good dissection about how to proceed in this case).

I assume that everyone is familiar with David Wilson's 4x4 opening book:

http://wilson.engr.wisc.edu/boxes/play.php?game=4x4

http://wilson.engr.wisc.edu/boxes/index.shtml

One possibility that has been suggested is to do the same thing for 5x5 up to 5 or 6 lines played.

What do people think?

Okay, here is a proposed policy:

1.  No comments on Little Golem unless requested specifically (and obviously never on games in progress).  I may ask whether my input is desired and would then wait for confirmation before posting.  The one exception to this is that I may try to find examples of perfect games and post that a given game was perfect.

2.  The Shark will be available online 24/7 to play against in an interactive mode (i.e. you can reference it using a UI which allows you to play both sides and thus get to any position you want, then ask it for its opinion).  This URL will be available until the end of the World Computer Championship, at which point it can be taken down if people feel that it ruins the game:

http://ec2-52-10-226-219.us-west-2.compute.amazonaws.com:8080/TheShark/SharkClient/index.html

3.  I will be sending David Wilson some data such that he can create an interactive 5x5 opening resource like his 4x4 opening resource:

http://wilson.engr.wisc.edu/boxes/play.php?game=4x4&opt=YC

4.  I will be sending David Wilson a complete analysis of the 22 games he already has analyzed:

http://wilson.engr.wisc.edu/boxes/games/

Hm, if i use the .jar to play the_shark then in some positions i'm getting an error message:

Exception in thread "Thread-145" java.lang.RuntimeException: All moves must be in the database!

        at TheShark.DotsIncredible.evaluateMove(DotsIncredible.java:5456)

        at TheShark.DotsIncredible.evaluateMoves(DotsIncredible.java:5560)

        at TheShark.DotsIncredible.findMove(DotsIncredible.java:5608)

        at TheShark.DotsIncredible.handleRequest(DotsIncredible.java:6972)

        at TheShark.DotsIncredible.server(DotsIncredible.java:8205)

        at TheShark.DotsIncredible.main(DotsIncredible.java:335)

here are some of the positions that cause the error if I let the_shark make a move:

http://www.trmph.com/dnb/board#5,h7g8e10g10h11i6j11d9

http://www.trmph.com/dnb/board#5,f1f9e6d9b5g10d7b9b7d11g8b11h7h11j9k8e4k6

http://www.trmph.com/dnb/board#5,f3f9e4g8

The last one is particularly interesting because its a mirrored position. For some opening moves like 1.f3 the error would trigger every single time I would try to mirror the_sharks moves.

But in this game it was able to play without issues and if I just enter each position and click the make-a-computer-move-button then no error actually accurs with this particular game.

Oh the error actually doesn't occur anymore with the .jar if it starts with 1.f3 and I just mirror. But it does still occur if I enter one of the 3 positions that i posted in previous post and then click on make computer move.

Thanks, Purgency.  I fixed that problem and it should be all better now.  (It's such a pain trying to juggle 12TB of files....)

Thanks for putting up this applet on amazon servers William!

I have some problems too to play against the applet, with "all moves must be in the database!" error message even when TheShark choses its moves, for example in the game:

1: EF (or f7)

2: ED (or d7)

3: GF (or f5)

also TheShark tells me he has analyzed about 82188581000000 on a total of 1152921500000000000 positions, does it mean that in a random game, a position has a chance 1 in 14028 to be into the database?

Ack!  Not again!  Anyway, I rechecked everything and now it should be working.

@Carroll Well, not really.

For example, considering only those positions with exactly one move played, I analyzed 9 positions, covering all 60 positions.