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):
   6                 InternalAgent.__init__(self, name, colour)
   7                 self.value = {"pawn" : 1, "bishop" : 3, "knight" : 3, "rook" : 5, "queen" : 9, "king" : 100, "unknown" : 4}
   8
   9                 self.aggression = 2.0 # Multiplier for scoring due to aggressive actions
  10                 self.defence = 1.0 # Multiplier for scoring due to defensive actions
  11
  12                 self.depth = 0 # Current depth
  13                 self.max_depth = 2 # Recurse this many times (for some reason, makes more mistakes when this is increased???)
  14                 self.recurse_for = -1 # Recurse for the best few moves each times (less than 0 = all moves)
  15
  16                 for p in self.board.pieces["white"] + self.board.pieces["black"]:
  17                         p.last_moves = None
  18                         p.selected_moves = None
  19
  20
  21
  22         def get_value(self, piece):
  23                 if piece == None:
  24                         return 0.0
  25                 return float(self.value[piece.types[0]] + self.value[piece.types[1]]) / 2.0
  26
  27         # Score possible moves for the piece
  28
  29         def prioritise_moves(self, piece):
  30
  31                 #sys.stderr.write(sys.argv[0] + " : " + str(self) + " prioritise called for " + str(piece) + "\n")
  32
  33
  34
  35                 grid = self.board.probability_grid(piece)
  36                 #sys.stderr.write("\t Probability grid " + str(grid) + "\n")
  37                 moves = []
  38                 for x in range(w):
  39                         for y in range(h):
  40                                 if grid[x][y] < 0.3: # Throw out moves with < 30% probability
  41                                         #sys.stderr.write("\tReject " + str(x) + "," + str(y) + " (" + str(grid[x][y]) + ")\n")
  42                                         continue
  43
  44                                 target = self.board.grid[x][y]
  45
  46
  47
  48
  49                                 # Get total probability that the move is protected
  50                                 self.board.push_move(piece, x, y)
  51
  52
  53
  54                                 defenders = self.board.coverage(x, y, piece.colour, reject_allied = False)
  55                                 d_prob = 0.0
  56                                 for d in defenders.keys():
  57                                         d_prob += defenders[d]
  58                                 if len(defenders.keys()) > 0:
  59                                         d_prob /= float(len(defenders.keys()))
  60
  61                                 if (d_prob > 1.0):
  62                                         d_prob = 1.0
  63
  64                                 # Get total probability that the move is threatened
  65                                 attackers = self.board.coverage(x, y, opponent(piece.colour), reject_allied = False)
  66                                 a_prob = 0.0
  67                                 for a in attackers.keys():
  68                                         a_prob += attackers[a]
  69                                 if len(attackers.keys()) > 0:
  70                                         a_prob /= float(len(attackers.keys()))
  71
  72                                 if (a_prob > 1.0):
  73                                         a_prob = 1.0
  74
  75                                 self.board.pop_move()
  76
  77
  78
  79                                 # Score of the move
  80                                 value = self.aggression * (1.0 + d_prob) * self.get_value(target) - self.defence * (1.0 - d_prob) * a_prob * self.get_value(piece)
  81
  82                                 # Adjust score based on movement of piece out of danger
  83                                 attackers = self.board.coverage(piece.x, piece.y, opponent(piece.colour))
  84                                 s_prob = 0.0
  85                                 for a in attackers.keys():
  86                                         s_prob += attackers[a]
  87                                 if len(attackers.keys()) > 0:
  88                                         s_prob /= float(len(attackers.keys()))
  89
  90                                 if (s_prob > 1.0):
  91                                         s_prob = 1.0
  92                                 value += self.defence * s_prob * self.get_value(piece)
  93
  94                                 # Adjust score based on probability that the move is actually possible
  95                                 moves.append([[x, y], grid[x][y] * value])
  96
  97                 moves.sort(key = lambda e : e[1], reverse = True)
  98                 #sys.stderr.write(sys.argv[0] + ": Moves for " + str(piece) + " are " + str(moves) + "\n")
  99
 100                 piece.last_moves = moves
 101                 piece.selected_moves = None
 102
 103
 104
 105
 106                 return moves
 107
 108         def select_best(self, colour):
 109
 110                 self.depth += 1
 111                 all_moves = {}
 112                 for p in self.board.pieces[colour]:
 113                         self.choice = p # Temporarily pick that piece
 114                         m = self.prioritise_moves(p)
 115                         if len(m) > 0:
 116                                 all_moves.update({p : m[0]})
 117
 118                 if len(all_moves.items()) <= 0:
 119                         return None
 120
 121
 122                 opts = all_moves.items()
 123                 opts.sort(key = lambda e : e[1][1], reverse = True)
 124
 125                 if self.depth >= self.max_depth:
 126                         self.depth -= 1
 127                         return list(opts[0])
 128
 129                 if self.recurse_for >= 0:
 130                         opts = opts[0:self.recurse_for]
 131                 #sys.stderr.write(sys.argv[0] + " : Before recurse, options are " + str(opts) + "\n")
 132
 133                 # Take the best few moves, and recurse
 134                 for choice in opts[0:self.recurse_for]:
 135                         [xx,yy] = [choice[0].x, choice[0].y] # Remember position
 136                         [nx,ny] = choice[1][0] # Target
 137                         [choice[0].x, choice[0].y] = [nx, ny] # Set position
 138                         target = self.board.grid[nx][ny] # Remember piece in spot
 139                         self.board.grid[xx][yy] = None # Remove piece
 140                         self.board.grid[nx][ny] = choice[0] # Replace with moving piece
 141
 142                         # Recurse
 143                         best_enemy_move = self.select_best(opponent(choice[0].colour))
 144                         choice[1][1] -= best_enemy_move[1][1] / float(self.depth + 1.0)
 145
 146                         [choice[0].x, choice[0].y] = [xx, yy] # Restore position
 147                         self.board.grid[nx][ny] = target # Restore taken piece
 148                         self.board.grid[xx][yy] = choice[0] # Restore moved piece
 149
 150
 151
 152                 opts.sort(key = lambda e : e[1][1], reverse = True)
 153                 #sys.stderr.write(sys.argv[0] + " : After recurse, options are " + str(opts) + "\n")
 154
 155                 self.depth -= 1
 156                 return list(opts[0])
 157
 158
 159
 160         # Returns [x,y] of selected piece
 161         def select(self):
 162                 #sys.stderr.write("Getting choice...")
 163                 self.choice = self.select_best(self.colour)[0]
 164
 165                 #sys.stderr.write(" Done " + str(self.choice)+"\n")
 166                 return [self.choice.x, self.choice.y]
 167
 168         # Returns [x,y] of square to move selected piece into
 169         def get_move(self):
 170                 #sys.stderr.write("Choice is " + str(self.choice) + "\n")
 171                 self.choice.selected_moves = self.choice.last_moves
 172                 moves = self.prioritise_moves(self.choice)
 173                 if len(moves) > 0:
 174                         return moves[0][0]
 175                 else:
 176                         return AgentRandom.get_move(self)
 177