qchess/src/agent_bishop.py

   1 # A sample agent
   2
   3
   4 class AgentBishop(AgentRandom): # Inherits from AgentRandom (in qchess)
   5         def __init__(self, name, colour,value={"pawn" : 1, "bishop" : 3, "knight" : 3, "rook" : 5, "queen" : 9, "king" : 100, "unknown" : 2}):
   6                 InternalAgent.__init__(self, name, colour)
   7                 self.value = value
   8                 self.aggression = 2.0 # Multiplier for scoring due to aggressive actions
   9                 self.defence = 1.0 # Multiplier for scoring due to defensive actions
  10
  11                 self.depth = 0 # Current depth
  12                 self.max_depth = 2 # Recurse this many times (for some reason, makes more mistakes when this is increased???)
  13                 self.recurse_for = -1 # Recurse for the best few moves each times (less than 0 = all moves)
  14
  15                 for p in self.board.pieces["white"] + self.board.pieces["black"]:
  16                         p.last_moves = None
  17                         p.selected_moves = None
  18
  19
  20
  21         def get_value(self, piece):
  22                 if piece == None:
  23                         return 0.0
  24                 return float(self.value[piece.types[0]] + self.value[piece.types[1]]) / 2.0
  25
  26         # Score possible moves for the piece
  27
  28         def prioritise_moves(self, piece):
  29
  30                 #sys.stderr.write(sys.argv[0] + " : " + str(self) + " prioritise called for " + str(piece) + "\n")
  31
  32
  33
  34                 grid = self.board.probability_grid(piece)
  35                 #sys.stderr.write("\t Probability grid " + str(grid) + "\n")
  36                 moves = []
  37                 for x in range(w):
  38                         for y in range(h):
  39                                 if grid[x][y] < 0.3: # Throw out moves with < 30% probability
  40                                         #sys.stderr.write("\tReject " + str(x) + "," + str(y) + " (" + str(grid[x][y]) + ")\n")
  41                                         continue
  42
  43                                 target = self.board.grid[x][y]
  44
  45
  46
  47
  48                                 # Get total probability that the move is protected
  49                                 self.board.push_move(piece, x, y)
  50
  51
  52
  53                                 defenders = self.board.coverage(x, y, piece.colour, reject_allied = False)
  54                                 d_prob = 0.0
  55                                 for d in defenders.keys():
  56                                         d_prob += defenders[d]
  57                                 if len(defenders.keys()) > 0:
  58                                         d_prob /= float(len(defenders.keys()))
  59
  60                                 if (d_prob > 1.0):
  61                                         d_prob = 1.0
  62
  63                                 # Get total probability that the move is threatened
  64                                 attackers = self.board.coverage(x, y, opponent(piece.colour), reject_allied = False)
  65                                 a_prob = 0.0
  66                                 for a in attackers.keys():
  67                                         a_prob += attackers[a]
  68                                 if len(attackers.keys()) > 0:
  69                                         a_prob /= float(len(attackers.keys()))
  70
  71                                 if (a_prob > 1.0):
  72                                         a_prob = 1.0
  73
  74                                 self.board.pop_move()
  75
  76
  77
  78                                 # Score of the move
  79                                 value = self.aggression * (1.0 + d_prob) * self.get_value(target) - self.defence * (1.0 - d_prob) * a_prob * self.get_value(piece)
  80
  81                                 # Adjust score based on movement of piece out of danger
  82                                 attackers = self.board.coverage(piece.x, piece.y, opponent(piece.colour))
  83                                 s_prob = 0.0
  84                                 for a in attackers.keys():
  85                                         s_prob += attackers[a]
  86                                 if len(attackers.keys()) > 0:
  87                                         s_prob /= float(len(attackers.keys()))
  88
  89                                 if (s_prob > 1.0):
  90                                         s_prob = 1.0
  91                                 value += self.defence * s_prob * self.get_value(piece)
  92
  93                                 # Adjust score based on probability that the move is actually possible
  94                                 moves.append([[x, y], grid[x][y] * value])
  95
  96                 moves.sort(key = lambda e : e[1], reverse = True)
  97                 #sys.stderr.write(sys.argv[0] + ": Moves for " + str(piece) + " are " + str(moves) + "\n")
  98
  99                 piece.last_moves = moves
 100                 piece.selected_moves = None
 101
 102
 103
 104
 105                 return moves
 106
 107         def select_best(self, colour):
 108
 109                 self.depth += 1
 110                 all_moves = {}
 111                 for p in self.board.pieces[colour]:
 112                         self.choice = p # Temporarily pick that piece
 113                         m = self.prioritise_moves(p)
 114                         if len(m) > 0:
 115                                 all_moves.update({p : m[0]})
 116
 117                 if len(all_moves.items()) <= 0:
 118                         return None
 119
 120
 121                 opts = all_moves.items()
 122                 opts.sort(key = lambda e : e[1][1], reverse = True)
 123
 124                 if self.depth >= self.max_depth:
 125                         self.depth -= 1
 126                         return list(opts[0])
 127
 128                 if self.recurse_for >= 0:
 129                         opts = opts[0:self.recurse_for]
 130                 #sys.stderr.write(sys.argv[0] + " : Before recurse, options are " + str(opts) + "\n")
 131
 132                 # Take the best few moves, and recurse
 133                 for choice in opts[0:self.recurse_for]:
 134                         [xx,yy] = [choice[0].x, choice[0].y] # Remember position
 135                         [nx,ny] = choice[1][0] # Target
 136                         [choice[0].x, choice[0].y] = [nx, ny] # Set position
 137                         target = self.board.grid[nx][ny] # Remember piece in spot
 138                         self.board.grid[xx][yy] = None # Remove piece
 139                         self.board.grid[nx][ny] = choice[0] # Replace with moving piece
 140
 141                         # Recurse
 142                         best_enemy_move = self.select_best(opponent(choice[0].colour))
 143                         choice[1][1] -= best_enemy_move[1][1] / float(self.depth + 1.0)
 144
 145                         [choice[0].x, choice[0].y] = [xx, yy] # Restore position
 146                         self.board.grid[nx][ny] = target # Restore taken piece
 147                         self.board.grid[xx][yy] = choice[0] # Restore moved piece
 148
 149
 150
 151                 opts.sort(key = lambda e : e[1][1], reverse = True)
 152                 #sys.stderr.write(sys.argv[0] + " : After recurse, options are " + str(opts) + "\n")
 153
 154                 self.depth -= 1
 155                 return list(opts[0])
 156
 157
 158
 159         # Returns [x,y] of selected piece
 160         def select(self):
 161                 #sys.stderr.write("Getting choice...")
 162                 self.choice = self.select_best(self.colour)[0]
 163
 164                 #sys.stderr.write(" Done " + str(self.choice)+"\n")
 165                 return [self.choice.x, self.choice.y]
 166
 167         # Returns [x,y] of square to move selected piece into
 168         def get_move(self):
 169                 #sys.stderr.write("Choice is " + str(self.choice) + "\n")
 170                 self.choice.selected_moves = self.choice.last_moves
 171                 moves = self.prioritise_moves(self.choice)
 172                 if len(moves) > 0:
 173                         return moves[0][0]
 174                 else:
 175                         return AgentRandom.get_move(self)
 176