• Announcements

    • khawk

      Download the Game Design and Indie Game Marketing Freebook   07/19/17

      GameDev.net and CRC Press have teamed up to bring a free ebook of content curated from top titles published by CRC Press. The freebook, Practices of Game Design & Indie Game Marketing, includes chapters from The Art of Game Design: A Book of Lenses, A Practical Guide to Indie Game Marketing, and An Architectural Approach to Level Design. The GameDev.net FreeBook is relevant to game designers, developers, and those interested in learning more about the challenges in game development. We know game development can be a tough discipline and business, so we picked several chapters from CRC Press titles that we thought would be of interest to you, the GameDev.net audience, in your journey to design, develop, and market your next game. The free ebook is available through CRC Press by clicking here. The Curated Books The Art of Game Design: A Book of Lenses, Second Edition, by Jesse Schell Presents 100+ sets of questions, or different lenses, for viewing a game’s design, encompassing diverse fields such as psychology, architecture, music, film, software engineering, theme park design, mathematics, anthropology, and more. Written by one of the world's top game designers, this book describes the deepest and most fundamental principles of game design, demonstrating how tactics used in board, card, and athletic games also work in video games. It provides practical instruction on creating world-class games that will be played again and again. View it here. A Practical Guide to Indie Game Marketing, by Joel Dreskin Marketing is an essential but too frequently overlooked or minimized component of the release plan for indie games. A Practical Guide to Indie Game Marketing provides you with the tools needed to build visibility and sell your indie games. With special focus on those developers with small budgets and limited staff and resources, this book is packed with tangible recommendations and techniques that you can put to use immediately. As a seasoned professional of the indie game arena, author Joel Dreskin gives you insight into practical, real-world experiences of marketing numerous successful games and also provides stories of the failures. View it here. An Architectural Approach to Level Design This is one of the first books to integrate architectural and spatial design theory with the field of level design. The book presents architectural techniques and theories for level designers to use in their own work. It connects architecture and level design in different ways that address the practical elements of how designers construct space and the experiential elements of how and why humans interact with this space. Throughout the text, readers learn skills for spatial layout, evoking emotion through gamespaces, and creating better levels through architectural theory. View it here. Learn more and download the ebook by clicking here. Did you know? GameDev.net and CRC Press also recently teamed up to bring GDNet+ Members up to a 20% discount on all CRC Press books. Learn more about this and other benefits here.
Sign in to follow this  
Followers 0
whitenose

minimax with alpha beta pruning implementation stuck.

28 posts in this topic

Hello,
Im new to this so please escuse my mistakes.
I'm trying to implement a 9Men's Morris game and i'm using minimax with alpha-beta pruning finding BestMove for "computer".

The problem is that if depth >= 3 then getBestMove() returns invalid move (board) , meaning that
it's returning a move (board) from other team sometimes - and I don't know where i go wrong.

Is my algorithm implementation done right ?! What i'm missing ?
Please Help !!


Here is my Java code , and getBestMove() function :

[code]

private int getBestMove(board board, int depth, int alpha, int beta, boolean bIsMax)
{
// is depth reached ? or do we have a winner ? (checkWinner() returns 0 if nobody is winning so far!)
if (depth == 0 || board.checkWinner()!=0)
{
// if depth reached, or we DO have a winner.
return board.evaluateBoard();
}

// is "computer" turn
if (bIsMax)
{
// generate all possible boards for "computer" (MAX).
ArrayList<board> boardsArray = board.generateNextPossibleBoards(bIsMax);

// loop thru all possible previously generated boards
for(board nextBoard: boardsArray)
{
// recurse
int score = Math.max(alpha, getBestMove(nextBoard, depth - 1, alpha, beta, !bIsMax));
nextBoard.showBoard("next-max[" + boardsArray.size() + "]"); // display to user current board.(debbuging purpose)

// found a better board ?
if ( score > alpha )
{
alpha = score;
chosenBoard = nextBoard; // save best board so far to "chosenBoard"
}

if(beta<=score) break; // prune
}

// clear boards array when done.
boardsArray.clear();

// return alpha.
return alpha;
}
else
{
// generate all possible boards for "player" (MIN).
ArrayList<board> boardsArray = board.generateNextPossibleBoards(bIsMax);

// loop thru all these possible boards.
for(board nextBoard: boardsArray)
{
// recurse
beta = Math.min(beta, getBestMove(nextBoard, depth - 1, alpha, beta, !bIsMax));
nextBoard.showBoard("next-min[" + boardsArray.size() + "]");

if(beta <= alpha) break; // prune.
}

// clear boards array when done.
boardsArray.clear();

// return beta.
return beta;
}

}


[/code]

and the calling of this method is:

[code]
short depth = 2;

// get best move for "computer" on this phase
getBestMove(simulationBoard, depth , Integer.MIN_VALUE+1, Integer.MAX_VALUE-1, true);
[/code]

and this is the evaluation function :

[code]
// evaluate current board.
public int evaluateBoard()
{
int score = 0;

score = score + (checkWinner()); // checkWinner() returns -1000000 if player wins, +1000000 if computer wins.
score = score + (checkMills()*1000); // checkMills() returns -10000*mills_count if player has some mills formed, or +10000*mills_count otherwise.
score = score + (checkBlockers()*500); // checkBlockers() returns -1000*block_count if player has bloked some tokens, or +1000*block_count otherwise
score = score + (checkFormers()*50); // etc.
score = score + (checkTokenDifference()*10);
return score;
}

[/code]


and the generateNextPossibleBoards() function wich chooses "default" switch case here :

[code]

public ArrayList<board> generateNextPossibleBoards(boolean bIsMax)
{
// copy current board
board currentBoard = new board(this);

// instantiate a empty array.
ArrayList<board> boardsArray = new ArrayList<board>();

// current team is unknown.
short team = -1;
short game_status = -1;

// update current team (can be "computer" or "human" player).
if(bIsMax)
{
team = renderer.TEAM_OPPONENT;
game_status = (short) renderer.gOpponent.gGameStatus;
}
else
{
team = renderer.TEAM_PLAYER;
game_status = (short) renderer.gPlayer.gGameStatus;
}

// according to current player's (computer/human) game-status:
switch(game_status)
{
...

default:

// generate a move (a board) and save this new board to boardsArray.
// if no moves left, return all boards so far!
for(short i=0;i<24;i++)
{
// go thru all tokens in this board.
if(currentBoard.gNode[i].gOccupiedByTeam==team)
{
// check if any empty neighbours for this token
for(short n=0;n<4;n++)
{
// get neighborur node id
short id = currentBoard.gNode[i].gNeighbourId[n];

// check if neighbour is empty spot (empty node)
if(id!=-1 && currentBoard.gNode[id].gOccupiedByTeam==renderer.TEAM_NOTEAM)
{
// make move for this token to the empty neighbour found above.
currentBoard.makeMoveToken( i , id );
boardsArray.add(new board(currentBoard));

// undo previous move
currentBoard.undoMoveToken( id , i );
}
}


}

}

break;
}//end switch.

// return all possible boards so far.
return boardsArray;
}
[/code]


thak you.

P.S: sorry for the bad view of some of the code.
just could not succeded making it "java" (using [source lang="java"]) because it was "eating" my code and the functions were displayed
partially . Edited by whitenose
0

Share this post


Link to post
Share on other sites
I don't really know Java, but I don't see any obvious bugs in the code you posted.

It's easier to get things right by adopting the negamax convention (in every node positive score means good for the player to move, so you flip the sign of the result of the recursive call and treat every node as a max node).
0

Share this post


Link to post
Share on other sites
[quote name='alvaro' timestamp='1349727261' post='4988097']
I don't really know Java, but I don't see any obvious bugs in the code you posted.

It's easier to get things right by adopting the negamax convention (in every node positive score means good for the player to move, so you flip the sign of the result of the recursive call and treat every node as a max node).
[/quote]

Thank you for replying mr. alvaro.
The algorithm should be the same, and it looks ok to me so far also ... but i still dont get it [img]http://public.gamedev.net//public/style_emoticons/default/blink.png[/img]
Why it returns some invalid moves when [b]depth [/b]is greater or equal than 3 (i.e depth >= 3) ??

With depth>2 the algorithm should "see" ahead further moves and return them.
I only get moves from "player"'s team (not the "computer" team as it should).

But with [b]depth = 1 or 2[/b] all is just [b]Fine[/b] ! (it returns best board/move from all of possible boards'/moves)
But this has a disadvantage: it cannot anticipate further moves of the "player" . Only the "power" of the evaluateBoard() function is used then (a local board evaluation). Edited by whitenose
0

Share this post


Link to post
Share on other sites
You have a bug, but it is not obvious to me what it is, from the code posted here and my understanding of Java. It is perfectly possible that your bug is in some other part of the code. I'm afraid you are going to have to debug it yourself. Find a position where the problem occurs and try to step through the code with a debugger. Unfortunately, recursive code is notoriously confusing to debug unless you have experience.
0

Share this post


Link to post
Share on other sites
when adding features to a minimax search I've found it useful to compare the output of the newly enhanced search with the output of the basic search (in this case the pure minimax). For most new features the results should be the same (not always, in rare cases of transposition tables, obviously things like razoring will change the output).

If that checks out then you need to step through (depth three shouldn't be too hard for this) maybe with a contrived starting position.Remember that your fist branch should return at least some value (not an invalid board), so your step iteration will only be 2 + branching factor to at least see the first leaf branch return something.
0

Share this post


Link to post
Share on other sites
I think I see the problem in the code. The chosenBoard looks like a global variable, and it will be overwritten at a lower level from a future cutoff from the current 'best board'. Avoid global variables in recursive code, rather return the best board and let the parent decide what to do with it.
0

Share this post


Link to post
Share on other sites
Wow, I just noticed something crazy in your code. What are you calling showBoard in the middle of the search?

EDIT: If you expect that code to only be executed at the root, but it turns out it's getting executed in internal nodes as well, the output will be very confusing. Perhaps that's what you are experiencing...

I recommend separating the search of the root of the tree into its own function. There is a little bit of code duplication, but there are enough differences to justify it (returning a move, reporting progress, iterative deepening, time control, looking up the opening book, clearing the hash tables or at least making them obsolete...). Edited by alvaro
0

Share this post


Link to post
Share on other sites
[quote name='Druzil' timestamp='1349732608' post='4988136']
I think I see the problem in the code. The chosenBoard looks like a global variable, and it will be overwritten at a lower level from a future cutoff from the current 'best board'. Avoid global variables in recursive code, rather return the best board and let the parent decide what to do with it.
[/quote]

That is correct mr Druzil, and i think you are right !
I must return chosenBoard recursively (as a reference parameter) to the previously function call and decide what to do with it.
So far the chosenBoard is overwritten and the last value is always the value of "last best board" found at lower levels.
0

Share this post


Link to post
Share on other sites
[quote name='alvaro' timestamp='1349732686' post='4988137']
Wow, I just noticed something crazy in your code. What are you calling showBoard in the middle of the search?

EDIT: If you expect that code to only be executed at the root, but it turns out it's getting executed in internal nodes as well, the output will be very confusing. Perhaps that's what you are experiencing...

I recommend separating the search of the root of the tree into its own function. There is a little bit of code duplication, but there are enough differences to justify it (returning a move, reporting progress, iterative deepening, time control, looking up the opening book, clearing the hash tables or at least making them obsolete...).
[/quote]

showBoard() function only displays the current board to the debug window, it's just a debbugging function - i dont care about performance now :).
I was trying to see what happens when getBestMove() calls itself recursively - so i've made this function (wich doesn't modify the content of the board)

I will keep this in mind - separating the search of the root (level 0) into its own function.
But also must say that i'm thinking using negamax algorithm if i will not figure where i go wrong.... [img]http://public.gamedev.net//public/style_emoticons/default/unsure.png[/img]
0

Share this post


Link to post
Share on other sites
[quote name='whitenose' timestamp='1349755934' post='4988213']
But also must say that i'm thinking using negamax algorithm if i will not figure where i go wrong.... [img]http://public.gamedev.net//public/style_emoticons/default/unsure.png[/img]
[/quote]
certainly negamax is a simpler and cleaner algorithm. One thing to keep in mind is if you intend to reuse your algorithm for >2 player games or games with multisteps per play you won't be able to use negamax as there is an assumption built into the call that the next level down is your single opponent. if this is not the case then I would suggest moving to negamax will save you potential headaches later on
0

Share this post


Link to post
Share on other sites
[quote name='Druzil' timestamp='1349818928' post='4988504']
[quote name='whitenose' timestamp='1349755934' post='4988213']
But also must say that i'm thinking using negamax algorithm if i will not figure where i go wrong.... [img]http://public.gamedev.net//public/style_emoticons/default/unsure.png[/img]
[/quote]
certainly negamax is a simpler and cleaner algorithm. One thing to keep in mind is if you intend to reuse your algorithm for >2 player games or games with multisteps per play you won't be able to use negamax as there is an assumption built into the call that the next level down is your single opponent. if this is not the case then I would suggest moving to negamax will save you potential headaches later on
[/quote]
It's not obvious how to use minimax in games with more than 2 players anyway. I would look into Monte Carlo Tree Search if you want to write engines for a larger class of games. Edited by alvaro
0

Share this post


Link to post
Share on other sites
[quote name='alvaro' timestamp='1349821621' post='4988512']
It's not obvious how to use minimax in games with more than 2 players anyway.
[/quote]

It stops being minimax, but it is clear what the generalization is, right? You drop the zero-sum requirement, you give each player his own reward function (rather than having one be the negative of the other), and everybody does a MAX on his own function.

(Granted, this gets impractical for large numbers of players.)

How alpha-beta search generalizes is less clear to me. You can do a generic branch-and-bound, with upper and lower bounds for each player's reward (so you'd keep cutting chunks off a bounding cube), but there are other options; I don't know if the game structure lets you keep just one number instead of two per player, or if non--axis-aligned planes (I haven't thought this through) are useful. This has [i]got[/i] to be standard, but I don't know.
0

Share this post


Link to post
Share on other sites
I have thought multiple-player minimax a bit myself, and also how to extend minimax and alpha-beta to games with randomness (e.g., games with dice). You can think of games with randomness as having an extra player that doesn't have any preference for one result or another, so will pick a "move" at random.

I have never actually tried to implement any of it, but I think alpha-beta is going to be of very limited value in those more complicated scenarios.

MCTS on the other hand extends very naturally to multiplayer games and to games with randomness. You still need to evaluate the result of the game as one number (the utility) per player. The rest of the MCTS scheme works basically without modification from the 2-player deterministic version of the technique.
0

Share this post


Link to post
Share on other sites
I have implementations of multiplayer minimax and there certainly are papers out there going into detail of the issues. The easiest algoriths are paranoid search which basically treats all opponents essentially as one (they are all out to get you) which is the most pessimistic view as apposed to maxn which is the most optmistic method (everyone just tried to max their own score).
0

Share this post


Link to post
Share on other sites
Hello,

I'm so sorry for the late reply. I had a car accident, all it's ok...
Thanks for all your answers !

As i said first, i'm trying to do a simple 9Men's Morris game (for Android).
This game has 3 phases: PlacingTokens, PlayGame, EndGame as you well know.

I have switched to [b][i]negaMax()[/i][/b] but the problem persists, and i think I know where/when it happens (i will explain) but dont know how to fix it. [img]http://public.gamedev.net//public/style_emoticons/default/wacko.png[/img]
The problem is returning final "bestMove" to the main calling function .

Had some time today to implemeent [i][b]negaMax()[/b][/i] and i hope didn't made mistakes - and it seems that at "depth=1" when PlacingTokens phase is active -> all it's OK. (meaning that the "computer" plays good enough).
The calling code for this looks like this now for [i][b]negaMax()[/b][/i]:



[code]

// as you can see "depth=1" (MAX = "computer" turn !)
case renderer.GAMESTATUS_PLACETOKEN:

//PARAMETERS: simulationBoard, depth, alpha, beta, sign (or color wich wil alternate between +1 or -1)
negaMax(simulationBoard, 1, Integer.MIN_VALUE+1, Integer.MAX_VALUE-1, 1);
[/code]

My "negaMax()" function looks like this:

[code]

private int negaMax(board board, int depth, int alpha, int beta, int sign)
{
// if depth reached or we DO have a winner on "this" board
if(depth<=0 || board.checkWinner(sign))
{
// evaluateBoard() evaluates board from "sign"'s point of view and returns a positive or negative score.
return sign*board.evaluateBoard(sign);
}

int max = Integer.MIN_VALUE+1; // -VERY SMALL value.

// generate all available possible moves
ArrayList<board> boardsArray = board.generateNextPossibleBoards(sign);

// loop thru all generated boards
for(board nextBoard: boardsArray)
{
// recurse and store result to "x" .
int x = -negaMax(nextBoard, depth-1, -beta, -alpha, -sign); // sign alterantes between +1 & -1 (computer & human)

// got a "better/best" move ?
if (x>max)
{
// save the "best" move so far.
max = x;
chosenBoard = nextBoard;
chosenBoard.showBoard("FOUND BEST MOVE");
}

// alpha-beta pruning
if (x>alpha) alpha = x;
if (alpha>=beta) return alpha;
}

// clear boards when done.
boardsArray.clear();

// return best move so far.
return max;
}

[/code]

allright, and now the checkWinner(sign) method (wich returns true or false[u] [i]if a team is Winning[/i][/u])

[code]

public boolean checkWinner(int sign)
{
board board = this;

// ---------------------------------------------------------------------------------
// 0. get current team
// ---------------------------------------------------------------------------------
int team = renderer.TEAM_NOTEAM;
int opponent = renderer.TEAM_NOTEAM;

if(sign==1)
{
// if "computer" turn
team = renderer.TEAM_OPPONENT;
opponent = renderer.TEAM_PLAYER;
}
else
{
// if "human" turn
team = renderer.TEAM_PLAYER;
opponent = renderer.TEAM_OPPONENT;
}


// ---------------------------------------------------------------------------------
// 1. check if only 2 tokens (from opponent team!) are left on current board.
// ---------------------------------------------------------------------------------
int counter = 0;

for(short i=0;i<24;i++)
{
if(board.gNode[i].gOccupiedByTeam!=renderer.TEAM_NOTEAM && board.gNode[i].gOccupiedByTeam==opponent) counter++; // count opponent tokens
}

switch(team)
{
case renderer.TEAM_OPPONENT:
if(counter<=2 && renderer.gPlayer.gTokensToPlaceOnTable==0)
{
return true; // opponent has won.
}
break;
case renderer.TEAM_PLAYER:
if(counter<=2 && renderer.gOpponent.gTokensToPlaceOnTable==0)
{
return true; // player has won.
}
break;
}


// ---------------------------------------------------------------------------------
// 2. all opponent tokens are blocked! (no more possible moves, all tokens are blocked)
// ---------------------------------------------------------------------------------
for(short i=0;i<24;i++)
{
// go thru all tokens in this board, if opponent is not blocked -> return false.
if(board.gNode[i].gOccupiedByTeam==opponent)
{
// check if any empty neighbours for this token
for(short n=0;n<4;n++)
{
// get neighborur id
int id = board.gNode[i].gNeighbourId[n];

// check if neighbour is empty space.
if(id!=-1 && board.gNode[id].gOccupiedByTeam==renderer.TEAM_NOTEAM)
{
// still having empty spaces to move. (NOT BLOCKED!)
return false;
}
}
}
}

// opponent is BLOCKED! so this team is winning -> return true!
return true;
}

[/code]


and this is the evaluateBoard(sign) function wich evaluates current board from --> "[b]sign[/b]" ("computer" or "human") point of view :

[code]

// evaluate current board.
public int evaluateBoard(int sign)
{
int score = 0;
int formers = checkFormers(sign); // if 2 tokens on same row/column has been found, then a former has been found.
int blockers = checkBlockers(sign); // if 1 token blocks a token form other team, then we have a blocker.
int mills = checkMills(sign); // if a mill has been formed (3 tokens in a row/column), then we have a mill.
int tkcount = checkTokenCount(sign);// count token (team - opponent) difference on this board.

score = score + formers*250; // multiplier is 10 here for the "formers"
score = score + blockers*500; // multiplier is 500 here for the "blockers"
score = score + mills*1000; // etc.
score = score + tkcount*10;

if(sign==1) // is it "computer" turn?
{
return -score; // return score for "computer"
}
else
{
return +score; // return score for "human"
}

}

[/code]


OK, now generating boards in 1st phase (PlacingTokens) is done like so:

[code]

public ArrayList<board> generateNextPossibleBoards(int sign)
{
// copy current board
board currentBoard = new board(this);

// instantiate a empty array.
ArrayList<board> boardsArray = new ArrayList<board>();

// current team is unknown.
int team = -1;
int opponent = -1;
int game_status = -1;

// update current team (can be "computer" or "human" player).
if(sign==1)
{
team = renderer.TEAM_OPPONENT;
opponent = renderer.TEAM_PLAYER;
game_status = renderer.gOpponent.gGameStatus;
}
else
{
team = renderer.TEAM_PLAYER;
opponent = renderer.TEAM_OPPONENT;
game_status = renderer.gPlayer.gGameStatus;
}

// according to current player's (computer/human) game-status:
switch(game_status)
{
case renderer.GAMESTATUS_PLACETOKEN:
// simulate a single place token (on this board) and save this new board to boardsArray.
for(short i=0;i<24;i++)
{
// go thru all empty spaces on this board , and if not occupied by any token
if(currentBoard.gNode[i].gOccupiedByTeam==renderer.TEAM_NOTEAM)
{
// place (a opponent) token to the empty neighbour found
currentBoard.makePlaceToken( i , opponent ); // place a opponent token !!!

boardsArray.add(new board(currentBoard));

// undo previous move
currentBoard.undoPlaceToken( i , renderer.TEAM_NOTEAM);
}
}

break;
...
... return boardsArray; // return all generated boards.
[/code]

OK , all it's good so far, the "computer" finds best moves and plays good against "human" .
--------------------------------------------------------------------------------------------------------------------------------------------------


The problem is when swithcing to phase2 -> i.e the "PlayGame" phase !
The code doesn't change EXCEPT generating "moves" when in "PlayGame" phase.
Now, generating "moves" is done like so:

[code]
public ArrayList<board> generateNextPossibleBoards(int sign)
{
// copy current board
board currentBoard = new board(this);

// instantiate a empty array.
ArrayList<board> boardsArray = new ArrayList<board>();

// current team is unknown.
int team = -1;
int opponent = -1;
int game_status = -1;

// update current team (can be "computer" or "human" player).
if(sign==1)
{
team = renderer.TEAM_OPPONENT;
opponent = renderer.TEAM_PLAYER;
game_status = renderer.gOpponent.gGameStatus;
}
else
{
team = renderer.TEAM_PLAYER;
opponent = renderer.TEAM_OPPONENT;
game_status = renderer.gPlayer.gGameStatus;
}

// according to current player's (computer/human) game-status:
switch(game_status)
{
....
....
case renderer.GAMESTATUS_PLAYNORMAL:

// generate a move (a board) and save this new board to boardsArray.
// if no moves left, return all boards so far!
for(short i=0;i<24;i++)
{
// go thru all tokens in this board.
if(currentBoard.gNode[i].gOccupiedByTeam==team)
{
// check if any empty neighbours for this token
for(short n=0;n<4;n++)
{
// get neighborur id
int id = currentBoard.gNode[i].gNeighbourId[n];

// check if neighbour is empty spot.
if(id!=-1 && currentBoard.gNode[id].gOccupiedByTeam==renderer.TEAM_NOTEAM)
{
// make move for this token to the empty neighbour found above.
currentBoard.makeMoveToken( i , id );
boardsArray.add(new board(currentBoard));

// undo previous move
currentBoard.undoMoveToken( id , i );
}
}


}

}

break; // end PLAYNORMAL.

....
} // end function generateNextPossibleBoards()

[/code]

I have done [i][b][u]a picture[/u][/b][/i] for you to see what happens when [b][i]depth>1[/i][/b] and [b][i]why[/i][/b] the chosenBoard is [u][b]not [/b][/u]correctly chosen.

[url="http://imageshack.us/photo/my-images/600/mistaker.jpg/"]http://imageshack.us...0/mistaker.jpg/[/url]

step1) As you can see, token from node=3 (depth=3) generates a new "move" to node=4 (depth=2).
step2) After that negaMax() recurses itself again and the token from node=4 generates again a new move to node=7(depth=1).
step3) Similar, from node=7 a token is generated again and set to node=8 (when negaMax() recurses again).

Now token from node=8 is selected as the "[b]chosen[/b]" best move --> and return it to main function, right ? (because we have Win or good score)
But...when moving from starting node=3 to node=8 there's a problem (because node=8 token is "far" away from token=3)
[u][b]This is what i must solve, how do i return best move when depth > 1 ?[/b][/u]



Im sorry for the long post.
I hope you accept my appologies !

LE:
@alvaro : OK, ill try that and get back to you. THANKS! Edited by whitenose
0

Share this post


Link to post
Share on other sites
Please re-read this thing I wrote:
[quote name='alvaro' timestamp='1349732686' post='4988137']
I recommend separating the search of the root of the tree into its own function. There is a little bit of code duplication, but there are enough differences to justify it (returning a move, reporting progress, iterative deepening, time control, looking up the opening book, clearing the hash tables or at least making them obsolete...).
[/quote]

NegaMax should return a score and a score only. Then write another function that looks at each move available at the current position on the board and calls NegaMax for each one of them. This function is really very similar to NegaMax but, among other differences, it returns a move.
0

Share this post


Link to post
Share on other sites
[quote name='Druzil' timestamp='1349906815' post='4988884']
I have implementations of multiplayer minimax and there certainly are papers out there going into detail of the issues. The easiest algoriths are paranoid search which basically treats all opponents essentially as one (they are all out to get you) which is the most pessimistic view as apposed to maxn which is the most optmistic method (everyone just tried to max their own score).
[/quote]

This is interesting. "Paranoid search" hadn't occurred to me.

The "paranoid" concept is cool for a couple reasons.

The first is that it turns an N-player game into N separate two-player games. And if your game is zero-sum across all the players, then each of these two-player games will also be zero-sum, and so easy to solve (in mixed strategies, when turns are simultaneous).

The second is that, with this solution concept, we can understand "how we got there." With Nash equilibrium, it's unclear how you actually arrive at an equilibrium, or which equilibrium you arrive it. You've got things like Condorcet dynamics and "fictitious play," but these are kind of unsatisfying; there's no great reason to expect players to use these mental processes.

One question to ask about "paranoid" solutions is how "consistent" they are. Are the solutions you get even Nash equilibria? I expect not. Whereas I think(?) that maxn solutions are Nash equilibria. (But if not -- do we care? Why do we want Nash equilibria in the first place?)
1

Share this post


Link to post
Share on other sites
It seems to work now , final functions here:

negamax (returing a best score):
[source lang="java"]public int negamax(board board, int depth, int alpha, int beta, int sign)
{
if (depth == 0 || board.checkWinner(sign))
{
return board.evaluateBoard(sign);
}

int max = Integer.MIN_VALUE+1;

ArrayList<board> boardsArray = board.generateNextPossibleBoards(sign);


for(board nextBoard: boardsArray)
{
int score = -negamax(nextBoard, depth-1, -beta, -alpha, -sign);

if (score>max) max = score;
if (score > alpha) alpha = score;
if (score >= beta) return score;
}

return max;
}
[/source]

and function returing a move (helped by negamax() function) :

[source lang="java"] public board returnBestMove(board board, int depth, int sign)
{
board bestMove = null;
int best = Integer.MIN_VALUE+1;

ArrayList<board> boardsArray = board.generateNextPossibleBoards(sign);

for(board nextBoard: boardsArray)
{
if (bestMove == null) bestMove = nextBoard;

int score = -negamax(nextBoard, depth, Integer.MIN_VALUE+1, Integer.MAX_VALUE-1, sign);

if (score > best)
{
bestMove = nextBoard;
best = score;
}
}

return bestMove;
}
[/source]


bestMove = it's a board also, when function returnBestMove() finishes I just compare [i][b]bestMove [/b][/i]board with[i][b] initial Board[/b][/i] and
identify wich [i][b]token [/b][/i]has been moved (from wich pos, to wich destination pos).


If i've made some mistakes - your suggestions will be highly appreciated !
Thanks .
0

Share this post


Link to post
Share on other sites
[quote][code] int score = -negamax(nextBoard, depth, Integer.MIN_VALUE+1, Integer.MAX_VALUE-1, sign);[/code][/quote]

Following the general alpha-beta algorithm, that should actually be
[code] int score = -negamax(nextBoard, depth, Integer.MIN_VALUE+1, -best, sign);[/code]

Other than that, the only thing I would have done differently is not having a variable `max' at all: You can simply remove all mentions of it and return alpha at the end instead.

EDIT: Wait, why are you passing a sign around? Edited by alvaro
0

Share this post


Link to post
Share on other sites
@alvaro
[quote name='alvaro' timestamp='1350315133' post='4990405']
...
EDIT: Wait, why are you passing a sign around?
[/quote]

Sign value is: +1 or -1 (it alternates between MAX = +1 and MIN = -1 teams).
It's not correct ?
0

Share this post


Link to post
Share on other sites
What do you do with it? I don't see it in your code...

The whole point of NegaMax is that it makes all the nodes max nodes, with the resulting code being simpler.

EDIT: Oh, I see you are passing it to your evaluation function. One typically uses whose turn it is, which is part of the board class. Edited by alvaro
1

Share this post


Link to post
Share on other sites
Hello,
I'm sorry i keep coming back to this subject, but i have some issues wich i help you can give me an answer.
I have this EvalFunction :


[source lang="java"] public int evaluateBoard(int sign) // -1 = "computer" and +1 = "human"
{
int total = 0;

int computer_closedmills = this.checkMills(-1); // count how many closed mills AI has (3 tokens in a row/column).
int computer_openmills = this.checkFormers(-1); // count how many 2 in a row/column tokens AI has.
int computer_blockers = this.checkBlockers(-1); // count if one AI token -> blocks 2 player's tokens
int computer_tokensleft = this.checkTokens(-1); // count how many tokens AI has.
int computer_openspaces = this.checkOpenSpaces(-1); // count how many move possibilities AI has (in current board)


int player_closedmills = this.checkMills(1); // count how many mills "human" has
int player_openmills = this.checkFormers(1); // count how many 2 in a row/column "human" has.
int player_blockers = this.checkBlockers(1); // etc.
int player_tokensleft = this.checkTokens(1);
int player_openspaces = this.checkOpenSpaces(1);

// compute total score
total -= ((computer_closedmills * 20)+ (computer_openmills * 40) + (computer_blockers * 30) + (computer_tokensleft * 10) + (computer_openspaces*5));
total += ((player_closedmills * 20) + (player_openmills * 40) + (player_blockers * 30) + (player_tokensleft * 10) + (player_openspaces*5));

if(sign==-1) return total;
else return -total;
}
[/source]

Ok, now i have [u]this situation[/u] as you can see below represented in this [i][b]Picture [/b][/i]:

[url="http://imageshack.us/photo/my-images/688/evalfuncnotworking.jpg/"]http://imageshack.us...notworking.jpg/[/url]


Note that red tokens represent "computer".
Instead of moving token 22 to position 19 (wich will form a new mill) , my evalFunction() makes another decision
and moves something let's say like token 21 to position 9 , or 18 to position 10.

Can you help me improve my eval function so that token 22 moves to 19 ?
([i] so that mill 21-22-23 breaks and token 22 forms a new mill on 18-19-20[/i])

thanks ! Edited by whitenose
0

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now
Sign in to follow this  
Followers 0