Updates for capture assignment.

Author: eriq-augustine

pacai/agents/greedy.py

@@ -63,7 +63,7 @@ def evaluate_state(self,
 
         score = 0.0
         for key, value in features.items():
-            score += (self.weights.get(key, 1.0) * value)
+            score += (self.weights.get(key, 0.0) * value)
 
         return score
 

pacai/capture/agents.py

@@ -7,24 +7,35 @@
 import pacai.core.features
 import pacai.search.distance
 
+GHOST_IGNORE_RANGE: float = 2.5
+
 class DefensiveAgent(pacai.agents.greedy.GreedyFeatureAgent):
     """
     A capture agent that prioritizes defending its own territory.
     """
 
-    def __init__(self, **kwargs) -> None:
+    def __init__(self,
+            override_weights: dict[str, float] | None = None,
+            **kwargs) -> None:
         kwargs['feature_extractor_func'] = _extract_baseline_defensive_features
         super().__init__(**kwargs)
 
         self._distances: pacai.search.distance.DistancePreComputer = pacai.search.distance.DistancePreComputer()
         """ Precompute distances. """
 
+        # Set base weights.
         self.weights['on_home_side'] = 100.0
         self.weights['stopped'] = -100.0
-        self.weights['reverse'] = -2
+        self.weights['reverse'] = -2.0
         self.weights['num_invaders'] = -1000.0
         self.weights['distance_to_invader'] = -10.0
 
+        if (override_weights is None):
+            override_weights = {}
+
+        for (key, weight) in override_weights.items():
+            self.weights[key] = weight
+
     def game_start(self, initial_state: pacai.core.gamestate.GameState) -> None:
         self._distances.compute(initial_state.board)
 
@@ -33,16 +44,25 @@ class OffensiveAgent(pacai.agents.greedy.GreedyFeatureAgent):
     A capture agent that prioritizes defending its own territory.
     """
 
-    def __init__(self, **kwargs) -> None:
+    def __init__(self,
+            override_weights: dict[str, float] | None = None,
+            **kwargs) -> None:
         kwargs['feature_extractor_func'] = _extract_baseline_offensive_features
         super().__init__(**kwargs)
 
         self._distances: pacai.search.distance.DistancePreComputer = pacai.search.distance.DistancePreComputer()
         """ Precompute distances. """
 
+        # Set base weights.
         self.weights['score'] = 100.0
         self.weights['distance_to_food'] = -1.0
 
+        if (override_weights is None):
+            override_weights = {}
+
+        for (key, weight) in override_weights.items():
+            self.weights[key] = weight
+
     def game_start(self, initial_state: pacai.core.gamestate.GameState) -> None:
         self._distances.compute(initial_state.board)
 
@@ -95,18 +115,40 @@ def _extract_baseline_offensive_features(
     features: pacai.core.features.FeatureDict = pacai.core.features.FeatureDict()
     features['score'] = state.get_normalized_score(agent.agent_index)
 
+    # Note the side of the board we are on.
+    features['on_home_side'] = int(state.is_ghost(agent_index = agent.agent_index))
+
+    # Prefer moving over stopping.
+    features['stopped'] = int(action == pacai.core.action.STOP)
+
+    # Prefer not turning around.
+    # Remember that the state we get is already a successor, so we have to look two actions back.
+    agent_actions = state.get_agent_actions(agent.agent_index)
+    if (len(agent_actions) > 1):
+        features['reverse'] = int(action == state.get_reverse_action(agent_actions[-2]))
+
     current_position = state.get_agent_position(agent.agent_index)
     if (current_position is None):
         # We are dead and waiting to respawn.
         return features
 
     food_positions = state.get_food(agent_index = agent.agent_index)
-
     if (len(food_positions) > 0):
         food_distances = [agent._distances.get_distance(current_position, food_position) for food_position in food_positions]
         features['distance_to_food'] = min(distance for distance in food_distances if (distance is not None))
     else:
         # There is no food left, give a large score.
         features['distance_to_food'] = -100000
 
+    ghost_positions = state.get_nonscared_opponent_positions(agent_index = agent.agent_index)
+    if (len(ghost_positions) > 0):
+        ghost_distances = [agent._distances.get_distance(current_position, ghost_position) for ghost_position in ghost_positions.values()]
+        features['distance_to_ghost'] = min(distance for distance in ghost_distances if (distance is not None))
+        if (features['distance_to_ghost'] > GHOST_IGNORE_RANGE):
+            features['distance_to_ghost'] = 1000
+
+        features['distance_to_ghost_squared'] = features['distance_to_ghost'] ** 2
+    else:
+        features['distance_to_ghost'] = 0
+
     return features

pacai/student/capture-team.txt

@@ -0,0 +1 @@
+pacai