Merge pull request #38 from nhamil/conard/moar-sample-bots

Convert most sample bots to support maximizing and minimizing their heuristic

example_tournament.py

@@ -6,26 +6,29 @@
 
 def run_tournament():
     tournament = tilewe.tournament.Tournament([
-        tilewe.engine.MaximizeMoveDifferenceEngine(), 
-        tilewe.engine.LargestPieceEngine(), 
-        tilewe.engine.TileWeightEngine("WallCrawler", 'wall_crawl'),
-        tilewe.engine.TileWeightEngine("Turtle", 'turtle'),
-        tilewe.engine.MostOpenCornersEngine(), 
-        tilewe.engine.RandomEngine(),
+        tilewe.engine.MoveDifferenceEngine(style="max"), 
+        tilewe.engine.MoveDifferenceEngine(style="min"), 
+        tilewe.engine.PieceSizeEngine(style="max"), 
+        tilewe.engine.PieceSizeEngine(style="min"), 
+        tilewe.engine.TileWeightEngine(style='wall_crawl'),
+        tilewe.engine.TileWeightEngine(style='turtle'),
+        tilewe.engine.OpenCornersEngine(style="max"), 
+        tilewe.engine.OpenCornersEngine(style="min"), 
+        tilewe.engine.RandomEngine(name="Random"),
     ])
-
+    
     results = tournament.play(100, n_threads=multiprocessing.cpu_count(), move_seconds=1, elo_mode="estimated")
 
     # print the result of game 1
-    print(results.match_data[0].board)
+    # print(results.match_data[0].board)
 
     # print the total real time the tournament took and the average duration of each match, in seconds
     print(f"Tournament ran for {round(results.real_time, 4)}s with avg " + 
           f"match duration {round(results.average_match_duration, 4)}s\n")
 
-    # print the engine rankings sorted by win_counts desc and then by avg_scores asc
-    print(results.get_engine_rankings_display('win_counts', 'desc'))
-    print(results.get_engine_rankings_display('avg_scores', 'asc'))
+    # print the engine rankings sorted by win_counts desc and then by elo_error_margin asc
+    # print(results.get_engine_rankings_display('win_counts', 'desc'))
+    # print(results.get_engine_rankings_display('elo_error_margin', 'asc'))
 
 if __name__ == '__main__':
     run_tournament()

pyproject.toml

@@ -7,7 +7,7 @@ authors = [
 ]
 description = "A tile-placing game for AI development fun"
 readme = "README.md"
-requires-python = "==3.10.*"
+requires-python = ">=3.10"
 classifiers = [
     "Programming Language :: Python :: 3",
     "License :: OSI Approved :: MIT License",

tilewe/__init__.py

@@ -255,3 +255,7 @@ def is_legal(self, move: Move, for_player: Color=None) -> bool:
         ...
 
 from ctilewe import *  # noqa: E402, F401, F403
+
+N_PIECE_TILES:    list[int] = [n_piece_tiles(piece)    for piece in range(PIECE_COUNT)]  # noqa: E241, E272
+N_PIECE_CORNERS:  list[int] = [n_piece_corners(piece)  for piece in range(PIECE_COUNT)]  # noqa: E241, E272
+N_PIECE_CONTACTS: list[int] = [n_piece_contacts(piece) for piece in range(PIECE_COUNT)]  # noqa: E241, E272

tilewe/engine.py

@@ -78,16 +78,23 @@ def __init__(self, name: str="Random", estimated_elo: float=None):
     def on_search(self, board: tilewe.Board, _seconds: float) -> tilewe.Move: 
         return random.choice(board.generate_legal_moves()) 
 
-class MostOpenCornersEngine(Engine): 
+class OpenCornersEngine(Engine): 
     """
     Plays the move that results in the player having the most
     playable corners possible afterwards, i.e. maximizing the
     possible moves on the next turn.
     Fairly weak but does result in decent board coverage behavior.
     """
 
-    def __init__(self, name: str="MostOpenCorners", estimated_elo: float=None):
-        super().__init__(name, 15.0 if estimated_elo is None else estimated_elo)
+    def __init__(self, name: str=None, style: str="max", estimated_elo: float=None):
+        if style not in ["max", "min"]:
+            raise ValueError("Invalid style, must be 'max' or 'min'")
+
+        name = name or ("MostOpenCorners" if style == "max" else "LeastOpenCorners")
+        estimated_elo = (15.0 if style == "max" else -250.0) if estimated_elo is None else estimated_elo
+        self.func = min if style == "min" else max
+
+        super().__init__(name, estimated_elo)
 
     def on_search(self, board: tilewe.Board, _seconds: float) -> tilewe.Move:
         moves = board.generate_legal_moves() 
@@ -100,9 +107,9 @@ def corners_after_move(m: tilewe.Move) -> int:
                 corners = board.n_player_corners(player) 
                 return corners
 
-        return max(moves, key=corners_after_move)
+        return self.func(moves, key=corners_after_move)
 
-class LargestPieceEngine(Engine): 
+class PieceSizeEngine(Engine): 
     """
     Plays the best legal move prioritizing the following, in order:
         Piece with the most squares (i.e. most points)
@@ -113,24 +120,31 @@ class LargestPieceEngine(Engine):
     ties, it's effectively a greedy form of RandomEngine.
     """
 
-    def __init__(self, name: str="LargestPiece", estimated_elo: float=None):
-        super().__init__(name, 30.0 if estimated_elo is None else estimated_elo)
+    def __init__(self, name: str=None, style: str="max", estimated_elo: float=None):
+        if style not in ["max", "min"]:
+            raise ValueError("Invalid style, must be 'max' or 'min'")
+
+        name = name or ("LargestPiece" if style == "max" else "SmallestPiece")
+        estimated_elo = (30.0 if style == "max" else -150.0) if estimated_elo is None else estimated_elo
+        self.func = min if style == "min" else max
+
+        super().__init__(name, estimated_elo)
 
     def on_search(self, board: tilewe.Board, _seconds: float) -> tilewe.Move:
         moves = board.generate_legal_moves() 
         random.shuffle(moves) 
 
         def score(m: tilewe.Move): 
-            pc = m.piece 
-            return tilewe.n_piece_tiles(pc) * 100 + \
-                tilewe.n_piece_corners(pc) * 10 + \
-                tilewe.n_piece_contacts(pc) 
+            pc = m.piece
+            return (tilewe.N_PIECE_TILES[pc] * 100 + 
+                tilewe.N_PIECE_CORNERS[pc] * 10 + 
+                tilewe.N_PIECE_CONTACTS[pc])
 
-        best = max(moves, key=score)
+        best = self.func(moves, key=score)
 
         return best
 
-class MaximizeMoveDifferenceEngine(Engine): 
+class MoveDifferenceEngine(Engine): 
     """
     Plays the move that results in the player having the best difference 
     in subsequent legal move counts compared to all opponents. That is,
@@ -141,8 +155,15 @@ class MaximizeMoveDifferenceEngine(Engine):
     getting access to an open area on the board, etc.
     """
 
-    def __init__(self, name: str="MaximizeMoveDifference", estimated_elo: float=None):
-        super().__init__(name, 50.0 if estimated_elo is None else estimated_elo)
+    def __init__(self, name: str=None, style: str="max", estimated_elo: float=None):
+        if style not in ["max", "min"]:
+            raise ValueError("Invalid style, must be 'max' or 'min'")
+
+        name = name or ("MaxMoveDiff" if style == "max" else "MinMoveDiff")
+        estimated_elo = (50.0 if style == "max" else -200.0) if estimated_elo is None else estimated_elo
+        self.func = min if style == "min" else max
+
+        super().__init__(name, estimated_elo)
 
     def on_search(self, board: tilewe.Board, _seconds: float) -> tilewe.Move:
         moves = board.generate_legal_moves() 
@@ -159,7 +180,7 @@ def eval_after_move(m: tilewe.Move) -> int:
                     total += n_moves * (1 if color == player else -1)
                 return total
 
-        return max(moves, key=eval_after_move)
+        return self.func(moves, key=eval_after_move)
 
 class TileWeightEngine(Engine):
     """
@@ -229,29 +250,34 @@ class TileWeightEngine(Engine):
         'turtle': -40.0
     }
 
+    names = {
+        'wall_crawl': 'WallCrawler',
+        'turtle': 'Turtle'
+    }
+
     def __init__(self, 
-                 name: str="TileWeight", 
-                 weight_map: str='wall_crawl', 
+                 name: str=None, 
+                 style: str='wall_crawl', 
                  custom_weights: list[int | float]=None, 
                  estimated_elo: float=None): 
         """
-        Current `weight_map` built-in options are 'wall_crawl' and 'turtle'
+        Current `style` built-in options are 'wall_crawl' and 'turtle'
         Can optionally provide a custom set of weights instead
         """
 
-        est_elo: float = 0.0 if estimated_elo is None else estimated_elo
-
         if custom_weights is not None:
             if len(custom_weights) != 20 * 20:
                 raise Exception("TileWeightEngine custom_weights must be a list of exactly 400 values")
             self.weights = custom_weights
+            name = name or "CustomTileWeights"
+            est_elo: float = 0.0 if estimated_elo is None else estimated_elo
 
         else:
-            if weight_map not in self.weight_maps:
-                raise Exception("TileWeightEngine given invalid weight_map choice")
-            self.weights = self.weight_maps[weight_map]
-            if estimated_elo is None: 
-                est_elo = self.weight_elos[weight_map] 
+            if style not in self.weight_maps:
+                raise Exception("TileWeightEngine given invalid style choice")
+            self.weights = self.weight_maps[style]
+            name = name or self.names[style]
+            est_elo: float = self.weight_elos[style] if estimated_elo is None else estimated_elo
 
         super().__init__(name, estimated_elo=est_elo)