add Training for AI and AI Model and allow to collect rl data from BestBattleSnake

snakes/TrainedBattleSnake.py

@@ -0,0 +1,91 @@
+from pathlib import Path
+from typing import Any
+import random, json, os
+
+from server.TrainBattleSnakeAI import MOVES, extract_feature_values
+from snakes.TemplateSnake import TemplateSnake
+
+class TrainedBattleSnake(TemplateSnake):
+  VERSION = "0.1.0"
+
+  def __init__(self):
+    super().__init__()
+    self.name = "TrainedBattleSnake"
+    self.version = self.VERSION
+    self._model_path:Path|None=None
+    self._model_data:dict[str, Any]|None=None
+
+  def choose_move(self, game_data) -> str:
+    self.game_board = game_data
+    self.calculations = []
+
+    safe_positions = self.find_safe_positions(add_to_calculations=True)
+    if not safe_positions:
+      self.add_to_history({"turn": game_data.get_turn(), "reason": "no_safe_moves"})
+      return "up"
+
+    model = self._load_model()
+    if not model:
+      move = random.choice(list(safe_positions.keys()))
+      self.add_to_history({
+        "turn": game_data.get_turn(),
+        "move": move,
+        "reason": "model_missing",
+        "safe_moves": list(safe_positions.keys()),
+      })
+      return move
+
+    row = {
+      "turn": game_data.get_turn(),
+      "game_board": game_data.get_game_board_as_dict(),
+    }
+    scores = self._predict_scores(model, row)
+
+    best_safe_move = max(safe_positions.keys(), key=lambda move: scores.get(move, float("-inf")))
+    self.add_to_history({
+      "turn": game_data.get_turn(),
+      "move": best_safe_move,
+      "safe_moves": list(safe_positions.keys()),
+      "scores": {move: round(scores.get(move, 0.0), 5) for move in MOVES},
+    })
+    return best_safe_move
+
+  def _load_model(self) -> dict[str, Any] | None:
+    env_path = os.getenv("TRAINED_SNAKE_MODEL", "models/battlesnake_softmax_v2.json")
+    path = Path(env_path)
+
+    if self._model_path == path and self._model_data is not None:
+      return self._model_data
+
+    if not path.exists() or not path.is_file():
+      self._model_path = path
+      self._model_data = None
+      return None
+
+    payload = json.loads(path.read_text(encoding="utf-8"))
+    model = payload.get("model")
+    if not isinstance(model, dict):
+      self._model_path = path
+      self._model_data = None
+      return None
+
+    self._model_path = path
+    self._model_data = model
+    return model
+
+  def _predict_scores(self, model:dict[str, Any], row:dict[str, Any]) -> dict[str, float]:
+    return self._predict_scores_softmax_v2(model, row)
+
+  def _predict_scores_softmax_v2(self, model:dict[str, Any], row:dict[str, Any]) -> dict[str, float]:
+    features = extract_feature_values(row)
+    weights = model.get("weights", {})
+    bias = model.get("bias", {})
+    scores:dict[str, float] = {}
+
+    for move in MOVES:
+      move_weights = weights.get(move, {})
+      score = float(bias.get(move, 0.0))
+      for name, value in features.items():
+        score += float(move_weights.get(name, 0.0)) * float(value)
+      scores[move] = score
+    return scores