synth-ai 0.2.14__py3-none-any.whl → 0.4.1__py3-none-any.whl

synth_ai/environments/examples/nethack/helpers/observation_utils.py

@@ -1,433 +0,0 @@
-"""Observation processing utilities for NetHack."""
-
-import re
-from typing import Any, Dict, List, Optional, Tuple
-
-
-def format_observation_for_llm(observation: Dict[str, Any]) -> str:
-    """
-    Format NetHack observation for LLM consumption.
-
-    Args:
-        observation: Raw observation dictionary
-
-    Returns:
-        Formatted string suitable for LLM input
-    """
-    lines = []
-
-    # Header with turn count and location
-    lines.append(f"=== NetHack - Turn {observation.get('turn_count', 0)} ===")
-    lines.append(f"Dungeon Level: {observation.get('dungeon_level', 1)}")
-
-    # ASCII map
-    if "ascii_map" in observation:
-        lines.append("\n--- Dungeon Map ---")
-        lines.append(observation["ascii_map"])
-
-    # Game message
-    if "message" in observation and observation["message"]:
-        # Clean null bytes and trailing whitespace
-        message = observation["message"].rstrip("\x00").strip()
-        if message:
-            lines.append(f"\nMessage: {message}")
-
-    # Character stats
-    if "character_stats" in observation:
-        stats = observation["character_stats"]
-        lines.append("\n--- Character Status ---")
-        lines.append(f"HP: {stats.get('hp', 0)}/{stats.get('max_hp', 0)}")
-        lines.append(f"Level: {stats.get('level', 1)} (Exp: {stats.get('experience', 0)})")
-        lines.append(f"AC: {stats.get('ac', 10)}, Gold: {stats.get('gold', 0)}")
-
-        # Attributes
-        attrs = []
-        for attr in [
-            "strength",
-            "dexterity",
-            "constitution",
-            "intelligence",
-            "wisdom",
-            "charisma",
-        ]:
-            if attr in stats:
-                attrs.append(f"{attr[:3].upper()}:{stats[attr]}")
-        if attrs:
-            lines.append(f"Attributes: {' '.join(attrs)}")
-
-    # Inventory summary
-    if "inventory_summary" in observation:
-        lines.append("\n--- Inventory ---")
-        lines.append(observation["inventory_summary"])
-
-    # Menu items if in menu
-    if observation.get("in_menu", False) and "menu_items" in observation:
-        lines.append("\n--- Menu Options ---")
-        for i, item in enumerate(observation["menu_items"]):
-            if i < 26:
-                lines.append(f"{chr(ord('a') + i)}) {item}")
-            else:
-                lines.append(f"{i - 26}) {item}")
-
-    # Score and rewards
-    lines.append("\n--- Progress ---")
-    lines.append(f"Score: {observation.get('score', 0)}")
-    lines.append(f"Total Reward: {observation.get('total_reward', 0.0):.2f}")
-    lines.append(f"Last Reward: {observation.get('reward_last', 0.0):.2f}")
-
-    # Termination status
-    if observation.get("terminated", False):
-        lines.append("\n*** GAME OVER ***")
-
-    return "\n".join(lines)
-
-
-def parse_ascii_map(ascii_map: str) -> Dict[str, Any]:
-    """
-    Parse ASCII map to extract key information.
-
-    Args:
-        ascii_map: Raw ASCII map string
-
-    Returns:
-        Dictionary with extracted map information
-    """
-    lines = ascii_map.strip().split("\n")
-
-    map_info = {
-        "width": 0,
-        "height": len(lines),
-        "player_position": None,
-        "stairs_positions": [],
-        "door_positions": [],
-        "item_positions": [],
-        "monster_positions": [],
-        "wall_positions": [],
-        "floor_positions": [],
-    }
-
-    if lines:
-        map_info["width"] = max(len(line) for line in lines)
-
-    # Common NetHack ASCII symbols
-    symbols = {
-        "@": "player",
-        "<": "stairs_up",
-        ">": "stairs_down",
-        "+": "closed_door",
-        "-": "open_door_horizontal",
-        "|": "open_door_vertical",
-        ".": "floor",
-        "#": "corridor",
-        " ": "wall",
-        "$": "gold",
-        "*": "gem",
-        "!": "potion",
-        "?": "scroll",
-        "/": "wand",
-        "=": "ring",
-        '"': "amulet",
-        "[": "armor",
-        ")": "weapon",
-        "(": "tool",
-        "%": "food",
-        "^": "trap",
-    }
-
-    # Monster symbols (letters)
-    for y, line in enumerate(lines):
-        for x, char in enumerate(line):
-            pos = (x, y)
-
-            if char == "@":
-                map_info["player_position"] = pos
-            elif char == "<":
-                map_info["stairs_positions"].append(("up", pos))
-            elif char == ">":
-                map_info["stairs_positions"].append(("down", pos))
-            elif char in ["+", "-", "|"]:
-                map_info["door_positions"].append((char, pos))
-            elif char in ["$", "*", "!", "?", "/", "=", '"', "[", ")", "(", "%"]:
-                map_info["item_positions"].append((char, pos))
-            elif char.isalpha() and char != "@":
-                # Store both the character and position for monster identification
-                map_info["monster_positions"].append((char, pos))
-            elif char == ".":
-                map_info["floor_positions"].append(pos)
-            elif char in ["#", " "]:
-                map_info["wall_positions"].append(pos)
-
-    return map_info
-
-
-def extract_game_context(observation: Dict[str, Any]) -> Dict[str, Any]:
-    """
-    Extract high-level game context from observation.
-
-    Args:
-        observation: Raw observation dictionary
-
-    Returns:
-        Dictionary with game context information
-    """
-    context = {
-        "in_combat": False,
-        "in_shop": False,
-        "at_stairs": False,
-        "items_nearby": False,
-        "doors_nearby": False,
-        "low_health": False,
-        "hungry": False,
-        "encumbered": False,
-        "in_menu": observation.get("in_menu", False),
-        "game_over": observation.get("terminated", False),
-    }
-
-    # Parse map for context
-    if "ascii_map" in observation:
-        map_info = parse_ascii_map(observation["ascii_map"])
-
-        # Check if player is near important features
-        if map_info["player_position"]:
-            px, py = map_info["player_position"]
-
-            # Check for nearby monsters (within 2 squares)
-            # Exclude pets from combat detection
-            pet_symbols = ["f", "d"]  # f = kitten, d = dog
-            for monster_char, (mx, my) in map_info["monster_positions"]:
-                if abs(mx - px) <= 2 and abs(my - py) <= 2:
-                    # Only trigger combat for non-pet monsters
-                    if monster_char not in pet_symbols:
-                        context["in_combat"] = True
-                        break
-
-            # Check for stairs
-            for stair_type, (sx, sy) in map_info["stairs_positions"]:
-                if sx == px and sy == py:
-                    context["at_stairs"] = True
-                    context["stairs_type"] = stair_type
-                    break
-
-            # Check for nearby items
-            for _, (ix, iy) in map_info["item_positions"]:
-                if abs(ix - px) <= 1 and abs(iy - py) <= 1:
-                    context["items_nearby"] = True
-                    break
-
-            # Check for nearby doors
-            for _, (dx, dy) in map_info["door_positions"]:
-                if abs(dx - px) <= 1 and abs(dy - py) <= 1:
-                    context["doors_nearby"] = True
-                    break
-
-    # Check health status
-    if "character_stats" in observation:
-        stats = observation["character_stats"]
-        hp = stats.get("hp", 0)
-        max_hp = stats.get("max_hp", 1)
-        if hp < max_hp * 0.3:
-            context["low_health"] = True
-
-    # Check for specific messages
-    message = observation.get("message", "").lower()
-    if "hungry" in message or "weak" in message:
-        context["hungry"] = True
-    if "burdened" in message or "stressed" in message:
-        context["encumbered"] = True
-    if "shop" in message or "shopkeeper" in message:
-        context["in_shop"] = True
-
-    return context
-
-
-def simplify_observation(observation: Dict[str, Any]) -> Dict[str, Any]:
-    """
-    Create a simplified observation for agents that need less detail.
-
-    Args:
-        observation: Full observation dictionary
-
-    Returns:
-        Simplified observation dictionary
-    """
-    simplified = {
-        "turn": observation.get("turn_count", 0),
-        "level": observation.get("dungeon_level", 1),
-        "hp": 0,
-        "max_hp": 0,
-        "message": observation.get("message", ""),
-        "terminated": observation.get("terminated", False),
-        "reward": observation.get("reward_last", 0.0),
-    }
-
-    # Extract HP
-    if "character_stats" in observation:
-        stats = observation["character_stats"]
-        simplified["hp"] = stats.get("hp", 0)
-        simplified["max_hp"] = stats.get("max_hp", 0)
-
-    # Extract key map features
-    if "ascii_map" in observation:
-        map_info = parse_ascii_map(observation["ascii_map"])
-        simplified["player_pos"] = map_info["player_position"]
-        simplified["monsters_nearby"] = len(map_info["monster_positions"])
-        simplified["items_nearby"] = len(map_info["item_positions"])
-        simplified["at_stairs"] = any(
-            pos == map_info["player_position"] for _, pos in map_info["stairs_positions"]
-        )
-
-    return simplified
-
-
-def extract_inventory_from_message(message: str) -> List[Dict[str, Any]]:
-    """
-    Extract inventory information from NetHack inventory messages.
-
-    Args:
-        message: Inventory message string
-
-    Returns:
-        List of inventory items
-    """
-    items = []
-
-    # Common inventory line patterns
-    # Example: "a - a blessed +1 long sword (weapon in hand)"
-    # Example: "b - an uncursed food ration"
-    pattern = r"^([a-zA-Z])\s*-\s*(.+?)(?:\s*\(([^)]+)\))?$"
-
-    for line in message.split("\n"):
-        match = re.match(pattern, line.strip())
-        if match:
-            letter, description, status = match.groups()
-
-            item = {
-                "letter": letter,
-                "description": description.strip(),
-                "status": status.strip() if status else None,
-            }
-
-            # Parse quantity
-            qty_match = re.match(r"^(\d+)\s+(.+)", description)
-            if qty_match:
-                item["quantity"] = int(qty_match.group(1))
-                item["name"] = qty_match.group(2)
-            else:
-                item["quantity"] = 1
-                item["name"] = description
-
-            # Identify item type
-            item["type"] = identify_item_type(description)
-
-            items.append(item)
-
-    return items
-
-
-def identify_item_type(description: str) -> str:
-    """
-    Identify the type of an item from its description.
-
-    Args:
-        description: Item description string
-
-    Returns:
-        Item type string
-    """
-    desc_lower = description.lower()
-
-    # Weapons
-    if any(
-        word in desc_lower
-        for word in [
-            "sword",
-            "dagger",
-            "spear",
-            "axe",
-            "mace",
-            "bow",
-            "arrow",
-            "dart",
-            "knife",
-        ]
-    ):
-        return "weapon"
-
-    # Armor
-    if any(
-        word in desc_lower
-        for word in [
-            "armor",
-            "mail",
-            "helmet",
-            "boots",
-            "gloves",
-            "shield",
-            "cloak",
-            "robe",
-        ]
-    ):
-        return "armor"
-
-    # Food
-    if any(
-        word in desc_lower
-        for word in [
-            "food",
-            "ration",
-            "corpse",
-            "egg",
-            "fruit",
-            "meat",
-            "candy",
-            "cookie",
-        ]
-    ):
-        return "food"
-
-    # Potions
-    if "potion" in desc_lower:
-        return "potion"
-
-    # Scrolls
-    if "scroll" in desc_lower:
-        return "scroll"
-
-    # Wands
-    if "wand" in desc_lower:
-        return "wand"
-
-    # Rings
-    if "ring" in desc_lower:
-        return "ring"
-
-    # Amulets
-    if "amulet" in desc_lower:
-        return "amulet"
-
-    # Tools
-    if any(
-        word in desc_lower
-        for word in [
-            "pick",
-            "key",
-            "lamp",
-            "candle",
-            "bag",
-            "sack",
-            "horn",
-            "whistle",
-            "mirror",
-        ]
-    ):
-        return "tool"
-
-    # Gems/stones
-    if any(word in desc_lower for word in ["gem", "stone", "rock", "crystal"]):
-        return "gem"
-
-    # Gold
-    if "gold" in desc_lower:
-        return "gold"
-
-    return "unknown"

synth_ai/environments/examples/nethack/helpers/recording_wrapper.py

@@ -1,200 +0,0 @@
-"""Environment wrapper that adds trajectory recording capabilities."""
-
-import logging
-from pathlib import Path
-from typing import Any, Dict, Optional, Tuple
-
-from src.synth_env.environment.tools import EnvToolCall
-from src.synth_env.examples.nethack.environment import NetHackEnvironment
-from src.synth_env.examples.nethack.helpers.trajectory_recorder import (
-    TrajectoryRecorder,
-)
-
-logger = logging.getLogger(__name__)
-
-
-class RecordingNetHackEnvironment(NetHackEnvironment):
-    """NetHack environment with automatic trajectory recording."""
-
-    def __init__(
-        self,
-        save_dir: str = "temp/nethack_trajectories",
-        auto_record: bool = True,
-        **kwargs,
-    ):
-        """Initialize recording wrapper.
-
-        Args:
-            save_dir: Directory to save trajectories
-            auto_record: Whether to automatically record all episodes
-            **kwargs: Arguments passed to NetHackEnvironment
-        """
-        super().__init__(**kwargs)
-
-        self.save_dir = Path(save_dir)
-        self.save_dir.mkdir(parents=True, exist_ok=True)
-
-        self.auto_record = auto_record
-        self.recorder = TrajectoryRecorder(save_dir)
-        self.is_recording = False
-        self.trajectory_id = None
-
-    async def start(self, **kwargs) -> Tuple[Dict[str, Any], Dict[str, Any]]:
-        """Start environment and optionally begin recording."""
-        public_state, private_state = await super().start(**kwargs)
-
-        if self.auto_record:
-            # Extract character role from task metadata
-            character_role = "adventurer"  # default
-            if hasattr(self.task_instance, "metadata") and hasattr(
-                self.task_instance.metadata, "character_role"
-            ):
-                character_role = self.task_instance.metadata.character_role
-
-            task_id = self.task_instance.task_id if self.task_instance else None
-            self.trajectory_id = self.recorder.start_recording(character_role, task_id)
-            self.is_recording = True
-
-            logger.info(f"Started recording trajectory: {self.trajectory_id}")
-
-            # Record initial state
-            obs = self._extract_observation(public_state)
-            self.recorder.record_step("reset", obs, 0.0, False, {})
-
-        return public_state, private_state
-
-    async def process_action(
-        self, tool_calls: list[EnvToolCall]
-    ) -> Tuple[Any, float, bool, Dict[str, Any]]:
-        """Process action and record if enabled."""
-        # Execute action
-        observation, reward, terminated, info = await super().process_action(tool_calls)
-
-        # Record step if recording
-        if self.is_recording and self.recorder.is_recording:
-            # Extract action from tool calls
-            action = "unknown"
-            if tool_calls and len(tool_calls) > 0:
-                if hasattr(tool_calls[0], "args") and "action" in tool_calls[0].args:
-                    action = tool_calls[0].args["action"]
-
-            # Extract observation data
-            obs_data = self._extract_observation(observation)
-
-            # Record step
-            self.recorder.record_step(action, obs_data, reward, terminated, info)
-
-        # If episode ended, stop recording
-        if terminated and self.is_recording:
-            await self._finalize_recording(observation)
-
-        return observation, reward, terminated, info
-
-    def _extract_observation(self, state: Any) -> Dict[str, Any]:
-        """Extract observation data from state object."""
-        if hasattr(state, "__dict__"):
-            # Convert state object to dict
-            obs = {}
-
-            # Extract relevant fields
-            if hasattr(state, "ascii_map"):
-                obs["ascii_map"] = state.ascii_map
-            if hasattr(state, "message"):
-                obs["message"] = state.message
-            if hasattr(state, "character_stats"):
-                obs["player_stats"] = state.character_stats
-            if hasattr(state, "inventory"):
-                obs["inventory"] = state.inventory
-            if hasattr(state, "position"):
-                obs["player_stats"] = obs.get("player_stats", {})
-                obs["player_stats"]["x"] = state.position[0]
-                obs["player_stats"]["y"] = state.position[1]
-            if hasattr(state, "dungeon_level"):
-                obs["player_stats"] = obs.get("player_stats", {})
-                obs["player_stats"]["depth"] = state.dungeon_level
-            if hasattr(state, "in_menu"):
-                obs["in_menu"] = state.in_menu
-            if hasattr(state, "menu_items"):
-                obs["menu_items"] = state.menu_items
-
-            return obs
-
-        # If it's already a dict, return as is
-        return state if isinstance(state, dict) else {}
-
-    async def _finalize_recording(self, final_state: Any):
-        """Finalize and save recording."""
-        if not self.is_recording:
-            return
-
-        # Determine final status
-        final_status = "completed"
-        if hasattr(final_state, "message"):
-            msg = final_state.message.lower()
-            if "die" in msg or "killed" in msg:
-                final_status = "died"
-            elif "quit" in msg:
-                final_status = "quit"
-            elif "time limit" in msg or "truncated" in msg:
-                final_status = "truncated"
-
-        # Extract achievements if available
-        achievements = {}
-        # TODO: Extract achievements from game state
-
-        # Stop recording
-        self.recorder.stop_recording(final_status, achievements)
-
-        # Save trajectory
-        filepath = self.recorder.save_trajectory()
-        logger.info(f"Saved trajectory to: {filepath}")
-
-        # Get and log summary
-        summary = self.recorder.get_summary()
-        logger.info(f"Trajectory summary: {summary}")
-
-        self.is_recording = False
-
-    def start_recording(
-        self, character_role: Optional[str] = None, task_id: Optional[str] = None
-    ) -> str:
-        """Manually start recording."""
-        if self.is_recording:
-            logger.warning("Recording already in progress")
-            return self.trajectory_id
-
-        if character_role is None:
-            character_role = "adventurer"
-            if hasattr(self, "engine") and hasattr(self.engine, "character_role"):
-                character_role = self.engine.character_role
-
-        self.trajectory_id = self.recorder.start_recording(character_role, task_id)
-        self.is_recording = True
-        logger.info(f"Started manual recording: {self.trajectory_id}")
-
-        return self.trajectory_id
-
-    def stop_recording(self, save: bool = True) -> Optional[str]:
-        """Manually stop recording."""
-        if not self.is_recording:
-            logger.warning("No recording in progress")
-            return None
-
-        self.recorder.stop_recording()
-
-        filepath = None
-        if save:
-            filepath = self.recorder.save_trajectory()
-            logger.info(f"Saved trajectory to: {filepath}")
-
-        self.is_recording = False
-        return filepath
-
-    def get_recording_status(self) -> Dict[str, Any]:
-        """Get current recording status."""
-        return {
-            "is_recording": self.is_recording,
-            "trajectory_id": self.trajectory_id,
-            "current_step": self.recorder.current_step if self.is_recording else 0,
-            "total_reward": self.recorder.total_reward if self.is_recording else 0.0,
-        }