I have problem with my own Chess Engine using minimax algorithm to search for chess moves I use a 5 plies depth search and with only material/bonus/mobility evaluation , but it also make dumb moves and sacrifices valuable pieces even when I give to them infinity (which is sure a search problem), I'm not using any types of pruning and gives a 5 depth search result in few seconds.
I'm stuck in this problem for a week, I am sure the Problem is with the Backtracking not the Chess Logic (so anyone with no chess background would solve this :)) and I searched a lot this is my first Question in Stack Overflow and I hope you guys won't Disappoint me :)
Here is the simple search code
int GameControl::Evaluate(ChessBoard _B)
{
int material=0,bonus=0,mobility=0;
for(int i=0;i<8;i++)
for(int j=0;j<8;j++)
{
if(_B.Board[i][j]!=EMPTY)
{
if(_B.Board[i][j]->pieceColor==WHITE){
material+=-_B.Board[i][j]->Weight;
bonus+=-_B.Board[i][j]->bonusPosition[i][j];
mobility+=-_B.Board[i][j]->getPossibleMovesList(i,j,B).size();
}
else {
material+=_B.Board[i][j]->Weight;
bonus+=_B.Board[i][j]->bonusPosition[i][j];
mobility+=_B.Board[i][j]->getPossibleMovesList(i,j,B).size();
}
}
}
return material+bonus/10+mobility/20;
}
pair<pair<int,int>,pair<int,int>> GameControl::minimax( int depth , ChessBoard _B )
{
short int i,j;
int bestValue = -INFINITY;
pair<pair<int,int>,pair<int,int>> bestMove;
vector< pair<int,int> > ::iterator it;
vector< pair<int,int> > Z;
for( i = 0; i < 8; i++ )
for( j = 0; j < 8; j++ )
{
if(_B.Board[i][j]!=EMPTY && _B.Board[i][j]->pieceColor==BLACK )
{
Z=_B.Board[i][j]->getPossibleMovesList(i,j,_B);
for(it=Z.begin();it!=Z.end();it++)
{
pair<int,int> temp;
temp.first=i,temp.second=j;
ChessPieces* DestinationPiece;
DestinationPiece=_B.Board[(*it).first][(*it).second];
//Moving
_B.Board[(*it).first][(*it).second]=_B.Board[i][j];
_B.Board[i][j]=EMPTY;
//
int value = minSearch( depth-1 , _B );
if( value > bestValue )
{
bestValue = value;
bestMove.first.first = i;
bestMove.first.second = j;
bestMove.second.first = (*it).first;
bestMove.second.second = (*it).second;
}
//Undo Move
_B.Board[i][j]=_B.Board[((*it).first)][(*it).second];
_B.Board[(*it).first][(*it).second]=DestinationPiece;
}
}
}
return bestMove;
}
int GameControl::minSearch( int depth , ChessBoard _B )
{
short int i;
short int j;
if(depth==0)
return Evaluate(_B);
int bestValue = INFINITY;
for( i = 0; i < 8; i++ )
for( j = 0; j < 8; j++ )
{
vector< pair<int,int> > ::iterator it;
vector< pair<int,int> > Z;
if(_B.Board[i][j]!=EMPTY && _B.Board[i][j]->pieceColor==WHITE && !_B.Board[i][j]->V.empty())
{
Z=_B.Board[i][j]->getPossibleMovesList(i,j,_B);
for(it=Z.begin();it!=Z.end();it++)
{
pair<int,int> temp;
temp.first=i;
temp.second=j;
ChessPieces* DestinationPiece;
DestinationPiece=_B.Board[(*it).first][(*it).second];
// Moving
_B.Board[(*it).first][(*it).second]=_B.Board[i][j];
_B.Board[i][j]=EMPTY;
//
int value = maxSearch( depth-1 , _B );
if( value < bestValue )
bestValue = value;
//Undo Move
_B.Board[i][j]=_B.Board[(*it).first][(*it).second];
_B.Board[(*it).first][(*it).second]=DestinationPiece;
//
}
}
}
return bestValue;
}
int GameControl::maxSearch( int depth , ChessBoard _B )
{
short int i;
short int j;
if(depth==0)
return Evaluate(_B);
vector< pair<int,int> > ::iterator it;
vector< pair<int,int> > Z;
int bestValue = -INFINITY;
for( i = 0; i < 8; i++ )
for( j = 0; j < 8; j++ )
{
if(_B.Board[i][j]!=EMPTY && _B.Board[i][j]->pieceColor==BLACK )
{
Z=_B.Board[i][j]->getPossibleMovesList(i,j,_B);
for(it=Z.begin();it!=Z.end();it++)
{
pair<int,int> temp;
temp.first=i,temp.second=j;
ChessPieces* DestinationPiece;
DestinationPiece=_B.Board[(*it).first][(*it).second];
//Moving
_B.Board[(*it).first][(*it).second]=_B.Board[i][j];
_B.Board[i][j]=EMPTY;
//
int value = minSearch( depth-1 , _B );
if( value > bestValue )
bestValue = value;
//Undo Move
_B.Board[i][j]=_B.Board[((*it).first)][(*it).second];
_B.Board[(*it).first][(*it).second]=DestinationPiece;
}
}
}
return bestValue;
}
You're not doing quiescence search, so the dumb moves are likely due to the well-known horizon effect that fixed depth minimax searches are susceptible to. At a minimum you should extend search for any forced moves, checks or captures where a piece captures one of equal or greater value.