synth-ai 0.1.9__py3-none-any.whl → 0.2.1.dev0__py3-none-any.whl

synth_ai/environments/examples/sokoban/engine_helpers/vendored/envs/room_utils.py

@@ -0,0 +1,331 @@
+import random
+import numpy as np
+import marshal
+
+
+def generate_room(
+    dim=(13, 13), p_change_directions=0.35, num_steps=25, num_boxes=3, tries=4, second_player=False
+):
+    """
+    Generates a Sokoban room, represented by an integer matrix. The elements are encoded as follows:
+    wall = 0
+    empty space = 1
+    box target = 2
+    box not on target = 3
+    box on target = 4
+    player = 5
+
+    :param dim:
+    :param p_change_directions:
+    :param num_steps:
+    :return: Numpy 2d Array
+    """
+    room_state = np.zeros(shape=dim)
+    room_structure = np.zeros(shape=dim)
+
+    # Some times rooms with a score == 0 are the only possibility.
+    # In these case, we try another model.
+    for t in range(tries):
+        room = room_topology_generation(dim, p_change_directions, num_steps)
+        room = place_boxes_and_player(room, num_boxes=num_boxes, second_player=second_player)
+
+        # Room fixed represents all not movable parts of the room
+        room_structure = np.copy(room)
+        room_structure[room_structure == 5] = 1
+
+        # Room structure represents the current state of the room including movable parts
+        room_state = room.copy()
+        room_state[room_state == 2] = 4
+
+        room_state, score, box_mapping = reverse_playing(room_state, room_structure)
+        room_state[room_state == 3] = 4
+
+        if score > 0:
+            break
+
+    if score == 0:
+        raise RuntimeWarning("Generated Model with score == 0")
+
+    return room_structure, room_state, box_mapping
+
+
+def room_topology_generation(dim=(10, 10), p_change_directions=0.35, num_steps=15):
+    """
+    Generate a room topology, which consits of empty floors and walls.
+
+    :param dim:
+    :param p_change_directions:
+    :param num_steps:
+    :return:
+    """
+    dim_x, dim_y = dim
+
+    # The ones in the mask represent all fields which will be set to floors
+    # during the random walk. The centered one will be placed over the current
+    # position of the walk.
+    masks = [
+        [[0, 0, 0], [1, 1, 1], [0, 0, 0]],
+        [[0, 1, 0], [0, 1, 0], [0, 1, 0]],
+        [[0, 0, 0], [1, 1, 0], [0, 1, 0]],
+        [[0, 0, 0], [1, 1, 0], [1, 1, 0]],
+        [[0, 0, 0], [0, 1, 1], [0, 1, 0]],
+    ]
+
+    # Possible directions during the walk
+    directions = [(1, 0), (0, 1), (-1, 0), (0, -1)]
+    direction = random.sample(directions, 1)[0]
+
+    # Starting position of random walk
+    position = np.array([random.randint(1, dim_x - 1), random.randint(1, dim_y - 1)])
+
+    level = np.zeros(dim, dtype=int)
+
+    for s in range(num_steps):
+        # Change direction randomly
+        if random.random() < p_change_directions:
+            direction = random.sample(directions, 1)[0]
+
+        # Update position
+        position = position + direction
+        position[0] = max(min(position[0], dim_x - 2), 1)
+        position[1] = max(min(position[1], dim_y - 2), 1)
+
+        # Apply mask
+        mask = random.sample(masks, 1)[0]
+        mask_start = position - 1
+        level[mask_start[0] : mask_start[0] + 3, mask_start[1] : mask_start[1] + 3] += mask
+
+    level[level > 0] = 1
+    level[:, [0, dim_y - 1]] = 0
+    level[[0, dim_x - 1], :] = 0
+
+    return level
+
+
+def place_boxes_and_player(room, num_boxes, second_player):
+    """
+    Places the player and the boxes into the floors in a room.
+
+    :param room:
+    :param num_boxes:
+    :return:
+    """
+    # Get all available positions
+    possible_positions = np.where(room == 1)
+    num_possible_positions = possible_positions[0].shape[0]
+    num_players = 2 if second_player else 1
+
+    if num_possible_positions <= num_boxes + num_players:
+        raise RuntimeError(
+            "Not enough free spots (#{}) to place {} player and {} boxes.".format(
+                num_possible_positions, num_players, num_boxes
+            )
+        )
+
+    # Place player(s)
+    ind = np.random.randint(num_possible_positions)
+    player_position = possible_positions[0][ind], possible_positions[1][ind]
+    room[player_position] = 5
+
+    if second_player:
+        ind = np.random.randint(num_possible_positions)
+        player_position = possible_positions[0][ind], possible_positions[1][ind]
+        room[player_position] = 5
+
+    # Place boxes
+    for n in range(num_boxes):
+        possible_positions = np.where(room == 1)
+        num_possible_positions = possible_positions[0].shape[0]
+
+        ind = np.random.randint(num_possible_positions)
+        box_position = possible_positions[0][ind], possible_positions[1][ind]
+        room[box_position] = 2
+
+    return room
+
+
+# Global variables used for reverse playing.
+explored_states = set()
+num_boxes = 0
+best_room_score = -1
+best_room = None
+best_box_mapping = None
+
+
+def reverse_playing(room_state, room_structure, search_depth=100):
+    """
+    This function plays Sokoban reverse in a way, such that the player can
+    move and pull boxes.
+    It ensures a solvable level with all boxes not being placed on a box target.
+    :param room_state:
+    :param room_structure:
+    :param search_depth:
+    :return: 2d array
+    """
+    global explored_states, num_boxes, best_room_score, best_room, best_box_mapping
+
+    # Box_Mapping is used to calculate the box displacement for every box
+    box_mapping = {}
+    box_locations = np.where(room_structure == 2)
+    num_boxes = len(box_locations[0])
+    for l in range(num_boxes):
+        box = (box_locations[0][l], box_locations[1][l])
+        box_mapping[box] = box
+
+    # explored_states globally stores the best room state and score found during search
+    explored_states = set()
+    best_room_score = -1
+    best_box_mapping = box_mapping
+    depth_first_search(
+        room_state, room_structure, box_mapping, box_swaps=0, last_pull=(-1, -1), ttl=300
+    )
+
+    return best_room, best_room_score, best_box_mapping
+
+
+def depth_first_search(
+    room_state, room_structure, box_mapping, box_swaps=0, last_pull=(-1, -1), ttl=300
+):
+    """
+    Searches through all possible states of the room.
+    This is a recursive function, which stops if the tll is reduced to 0 or
+    over 1.000.000 states have been explored.
+    :param room_state:
+    :param room_structure:
+    :param box_mapping:
+    :param box_swaps:
+    :param last_pull:
+    :param ttl:
+    :return:
+    """
+    global explored_states, num_boxes, best_room_score, best_room, best_box_mapping
+
+    ttl -= 1
+    if ttl <= 0 or len(explored_states) >= 300000:
+        return
+
+    state_tohash = marshal.dumps(room_state)
+
+    # Only search this state, if it not yet has been explored
+    if not (state_tohash in explored_states):
+        # Add current state and its score to explored states
+        room_score = box_swaps * box_displacement_score(box_mapping)
+        if np.where(room_state == 2)[0].shape[0] != num_boxes:
+            room_score = 0
+
+        if room_score > best_room_score:
+            best_room = room_state
+            best_room_score = room_score
+            best_box_mapping = box_mapping
+
+        explored_states.add(state_tohash)
+
+        for action in ACTION_LOOKUP.keys():
+            # The state and box mapping  need to be copied to ensure
+            # every action start from a similar state.
+            room_state_next = room_state.copy()
+            box_mapping_next = box_mapping.copy()
+
+            room_state_next, box_mapping_next, last_pull_next = reverse_move(
+                room_state_next, room_structure, box_mapping_next, last_pull, action
+            )
+
+            box_swaps_next = box_swaps
+            if last_pull_next != last_pull:
+                box_swaps_next += 1
+
+            depth_first_search(
+                room_state_next, room_structure, box_mapping_next, box_swaps_next, last_pull, ttl
+            )
+
+
+def reverse_move(room_state, room_structure, box_mapping, last_pull, action):
+    """
+    Perform reverse action. Where all actions in the range [0, 3] correspond to
+    push actions and the ones greater 3 are simmple move actions.
+    :param room_state:
+    :param room_structure:
+    :param box_mapping:
+    :param last_pull:
+    :param action:
+    :return:
+    """
+    player_position = np.where(room_state == 5)
+    player_position = np.array([player_position[0][0], player_position[1][0]])
+
+    change = CHANGE_COORDINATES[action % 4]
+    next_position = player_position + change
+
+    # Check if next position is an empty floor or an empty box target
+    if room_state[next_position[0], next_position[1]] in [1, 2]:
+        # Move player, independent of pull or move action.
+        room_state[player_position[0], player_position[1]] = room_structure[
+            player_position[0], player_position[1]
+        ]
+        room_state[next_position[0], next_position[1]] = 5
+
+        # In addition try to pull a box if the action is a pull action
+        if action < 4:
+            possible_box_location = change[0] * -1, change[1] * -1
+            possible_box_location += player_position
+
+            if room_state[possible_box_location[0], possible_box_location[1]] in [3, 4]:
+                # Perform pull of the adjacent box
+                room_state[player_position[0], player_position[1]] = 3
+                room_state[possible_box_location[0], possible_box_location[1]] = room_structure[
+                    possible_box_location[0], possible_box_location[1]
+                ]
+
+                # Update the box mapping
+                for k in box_mapping.keys():
+                    if box_mapping[k] == (possible_box_location[0], possible_box_location[1]):
+                        box_mapping[k] = (player_position[0], player_position[1])
+                        last_pull = k
+
+    return room_state, box_mapping, last_pull
+
+
+def box_displacement_score(box_mapping):
+    """
+    Calculates the sum of all Manhattan distances, between the boxes
+    and their origin box targets.
+    :param box_mapping:
+    :return:
+    """
+    score = 0
+
+    for box_target in box_mapping.keys():
+        box_location = np.array(box_mapping[box_target])
+        box_target = np.array(box_target)
+        dist = np.sum(np.abs(box_location - box_target))
+        score += dist
+
+    return score
+
+
+TYPE_LOOKUP = {
+    0: "wall",
+    1: "empty space",
+    2: "box target",
+    3: "box on target",
+    4: "box not on target",
+    5: "player",
+}
+
+ACTION_LOOKUP = {
+    0: "push up",
+    1: "push down",
+    2: "push left",
+    3: "push right",
+    4: "move up",
+    5: "move down",
+    6: "move left",
+    7: "move right",
+}
+
+# Moves are mapped to coordinate changes as follows
+# 0: Move up
+# 1: Move down
+# 2: Move left
+# 3: Move right
+CHANGE_COORDINATES = {0: (-1, 0), 1: (1, 0), 2: (0, -1), 3: (0, 1)}

synth_ai/environments/examples/sokoban/engine_helpers/vendored/envs/sokoban_env.py

@@ -0,0 +1,305 @@
+import gym
+from gym.utils import seeding
+from gym.spaces.discrete import Discrete
+from gym.spaces import Box
+from .room_utils import generate_room
+from .render_utils import room_to_rgb, room_to_tiny_world_rgb
+import numpy as np
+
+
+class SokobanEnv(gym.Env):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array", "raw"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array", "raw"],
+    }
+
+    def __init__(
+        self, dim_room=(10, 10), max_steps=120, num_boxes=4, num_gen_steps=None, reset=True
+    ):
+        # General Configuration
+        self.dim_room = dim_room
+        if num_gen_steps == None:
+            self.num_gen_steps = int(1.7 * (dim_room[0] + dim_room[1]))
+        else:
+            self.num_gen_steps = num_gen_steps
+
+        self.num_boxes = num_boxes
+        self.boxes_on_target = 0
+
+        # Penalties and Rewards
+        self.penalty_for_step = -0.1
+        self.penalty_box_off_target = -1
+        self.reward_box_on_target = 1
+        self.reward_finished = 10
+        self.reward_last = 0
+
+        # Other Settings
+        self.viewer = None
+        self.max_steps = max_steps
+        self.action_space = Discrete(len(ACTION_LOOKUP))
+        screen_height, screen_width = (dim_room[0] * 16, dim_room[1] * 16)
+        self.observation_space = Box(
+            low=0, high=255, shape=(screen_height, screen_width, 3), dtype=np.uint8
+        )
+
+        if reset:
+            # Initialize Room
+            _ = self.reset()
+
+    def seed(self, seed=None):
+        self.np_random, seed = seeding.np_random(seed)
+        return [seed]
+
+    def step(self, action, observation_mode="rgb_array"):
+        assert action in ACTION_LOOKUP
+        assert observation_mode in ["rgb_array", "tiny_rgb_array", "raw"]
+
+        self.num_env_steps += 1
+
+        self.new_box_position = None
+        self.old_box_position = None
+
+        moved_box = False
+
+        if action == 0:
+            moved_player = False
+
+        # All push actions are in the range of [0, 3]
+        elif action < 5:
+            moved_player, moved_box = self._push(action)
+
+        else:
+            moved_player = self._move(action)
+
+        self._calc_reward()
+
+        done = self._check_if_done()
+
+        # Convert the observation to RGB frame
+        observation = self.render(mode=observation_mode)
+
+        info = {
+            "action.name": ACTION_LOOKUP[action],
+            "action.moved_player": moved_player,
+            "action.moved_box": moved_box,
+        }
+        if done:
+            info["maxsteps_used"] = self._check_if_maxsteps()
+            info["all_boxes_on_target"] = self._check_if_all_boxes_on_target()
+
+        return observation, self.reward_last, done, info
+
+    def _push(self, action):
+        """
+        Perform a push, if a box is adjacent in the right direction.
+        If no box, can be pushed, try to move.
+        :param action:
+        :return: Boolean, indicating a change of the room's state
+        """
+        change = CHANGE_COORDINATES[(action - 1) % 4]
+        new_position = self.player_position + change
+        current_position = self.player_position.copy()
+
+        # No push, if the push would get the box out of the room's grid
+        new_box_position = new_position + change
+        if (
+            new_box_position[0] >= self.room_state.shape[0]
+            or new_box_position[1] >= self.room_state.shape[1]
+        ):
+            return False, False
+
+        can_push_box = self.room_state[new_position[0], new_position[1]] in [3, 4]
+        can_push_box &= self.room_state[new_box_position[0], new_box_position[1]] in [1, 2]
+        if can_push_box:
+            self.new_box_position = tuple(new_box_position)
+            self.old_box_position = tuple(new_position)
+
+            # Move Player
+            self.player_position = new_position
+            self.room_state[(new_position[0], new_position[1])] = 5
+            self.room_state[current_position[0], current_position[1]] = self.room_fixed[
+                current_position[0], current_position[1]
+            ]
+
+            # Move Box
+            box_type = 4
+            if self.room_fixed[new_box_position[0], new_box_position[1]] == 2:
+                box_type = 3
+            self.room_state[new_box_position[0], new_box_position[1]] = box_type
+            return True, True
+
+        # Try to move if no box to push, available
+        else:
+            return self._move(action), False
+
+    def _move(self, action):
+        """
+        Moves the player to the next field, if it is not occupied.
+        :param action:
+        :return: Boolean, indicating a change of the room's state
+        """
+        change = CHANGE_COORDINATES[(action - 1) % 4]
+        new_position = self.player_position + change
+        current_position = self.player_position.copy()
+
+        # Move player if the field in the moving direction is either
+        # an empty field or an empty box target.
+        if self.room_state[new_position[0], new_position[1]] in [1, 2]:
+            self.player_position = new_position
+            self.room_state[(new_position[0], new_position[1])] = 5
+            self.room_state[current_position[0], current_position[1]] = self.room_fixed[
+                current_position[0], current_position[1]
+            ]
+
+            return True
+
+        return False
+
+    def _calc_reward(self):
+        """
+        Calculate Reward Based on
+        :return:
+        """
+        # Every step a small penalty is given, This ensures
+        # that short solutions have a higher reward.
+        self.reward_last = self.penalty_for_step
+
+        # count boxes off or on the target
+        empty_targets = self.room_state == 2
+        player_on_target = (self.room_fixed == 2) & (self.room_state == 5)
+        total_targets = empty_targets | player_on_target
+
+        current_boxes_on_target = self.num_boxes - np.where(total_targets)[0].shape[0]
+
+        # Add the reward if a box is pushed on the target and give a
+        # penalty if a box is pushed off the target.
+        if current_boxes_on_target > self.boxes_on_target:
+            self.reward_last += self.reward_box_on_target
+        elif current_boxes_on_target < self.boxes_on_target:
+            self.reward_last += self.penalty_box_off_target
+
+        game_won = self._check_if_all_boxes_on_target()
+        if game_won:
+            self.reward_last += self.reward_finished
+
+        self.boxes_on_target = current_boxes_on_target
+
+    def _check_if_done(self):
+        # Check if the game is over either through reaching the maximum number
+        # of available steps or by pushing all boxes on the targets.
+        return self._check_if_all_boxes_on_target() or self._check_if_maxsteps()
+
+    def _check_if_all_boxes_on_target(self):
+        empty_targets = self.room_state == 2
+        player_hiding_target = (self.room_fixed == 2) & (self.room_state == 5)
+        are_all_boxes_on_targets = np.where(empty_targets | player_hiding_target)[0].shape[0] == 0
+        return are_all_boxes_on_targets
+
+    def _check_if_maxsteps(self):
+        return self.max_steps == self.num_env_steps
+
+    def reset(
+        self,
+        second_player: bool = False,
+        render_mode: str = "rgb_array",
+        seed=None,
+        *,
+        max_attempts: int = 5,
+    ):
+        if seed is not None:
+            self.seed(seed)
+
+        for attempt in range(max_attempts):
+            try:
+                self.room_fixed, self.room_state, self.box_mapping = generate_room(
+                    dim=self.dim_room,
+                    num_steps=self.num_gen_steps,
+                    num_boxes=self.num_boxes,
+                    second_player=second_player,
+                )
+                break  # success
+            except (RuntimeError, RuntimeWarning) as e:
+                print(f"[SOKOBAN] {e}  – retry {attempt + 1}/{max_attempts}")
+                if attempt == max_attempts - 1:
+                    # fallback: trivial 3×3 room with player only
+                    self.room_fixed = np.pad(np.ones((1, 1), dtype=int), 1, constant_values=0)
+                    self.room_state = self.room_fixed.copy()
+                    self.room_state[1, 1] = 5
+                    self.box_mapping = {}
+
+        self.player_position = np.argwhere(self.room_state == 5)[0]
+        self.num_env_steps = 0
+        self.reward_last = 0
+        self.boxes_on_target = 0
+
+        return self.render(render_mode)
+
+    def render(self, mode="human", close=None, scale=1):
+        assert mode in RENDERING_MODES
+
+        img = self.get_image(mode, scale)
+
+        if "rgb_array" in mode:
+            return img
+
+        elif "human" in mode:
+            from gym.envs.classic_control import rendering
+
+            if self.viewer is None:
+                self.viewer = rendering.SimpleImageViewer()
+            self.viewer.imshow(img)
+            return self.viewer.isopen
+
+        elif "raw" in mode:
+            arr_walls = (self.room_fixed == 0).view(np.int8)
+            arr_goals = (self.room_fixed == 2).view(np.int8)
+            arr_boxes = ((self.room_state == 4) + (self.room_state == 3)).view(np.int8)
+            arr_player = (self.room_state == 5).view(np.int8)
+
+            return arr_walls, arr_goals, arr_boxes, arr_player
+
+        else:
+            super(SokobanEnv, self).render(mode=mode)  # just raise an exception
+
+    def get_image(self, mode, scale=1):
+        if mode.startswith("tiny_"):
+            img = room_to_tiny_world_rgb(self.room_state, self.room_fixed, scale=scale)
+        else:
+            img = room_to_rgb(self.room_state, self.room_fixed)
+
+        return img
+
+    def close(self):
+        if self.viewer is not None:
+            self.viewer.close()
+
+    def set_maxsteps(self, num_steps):
+        self.max_steps = num_steps
+
+    def get_action_lookup(self):
+        return ACTION_LOOKUP
+
+    def get_action_meanings(self):
+        return ACTION_LOOKUP
+
+
+ACTION_LOOKUP = {
+    0: "no operation",
+    1: "push up",
+    2: "push down",
+    3: "push left",
+    4: "push right",
+    5: "move up",
+    6: "move down",
+    7: "move left",
+    8: "move right",
+}
+
+# Moves are mapped to coordinate changes as follows
+# 0: Move up
+# 1: Move down
+# 2: Move left
+# 3: Move right
+CHANGE_COORDINATES = {0: (-1, 0), 1: (1, 0), 2: (0, -1), 3: (0, 1)}
+
+RENDERING_MODES = ["rgb_array", "human", "tiny_rgb_array", "tiny_human", "raw"]

synth_ai/environments/examples/sokoban/engine_helpers/vendored/envs/sokoban_env_fixed_targets.py

@@ -0,0 +1,66 @@
+from .sokoban_env import SokobanEnv
+from .render_utils import room_to_rgb_FT, room_to_tiny_world_rgb_FT
+from gym.spaces import Box
+
+
+class FixedTargetsSokobanEnv(SokobanEnv):
+    def __init__(self, dim_room=(10, 10), max_steps=120, num_boxes=3, num_gen_steps=None):
+        super(FixedTargetsSokobanEnv, self).__init__(dim_room, max_steps, num_boxes, num_gen_steps)
+        screen_height, screen_width = (dim_room[0] * 16, dim_room[1] * 16)
+        self.observation_space = Box(low=0, high=255, shape=(screen_height, screen_width, 3))
+        self.boxes_are_on_target = [False] * num_boxes
+
+        _ = self.reset()
+
+    def get_image(self, mode, scale=1):
+        if mode.startswith("tiny_"):
+            img = room_to_tiny_world_rgb_FT(
+                self.room_state, self.box_mapping, self.room_fixed, scale=scale
+            )
+        else:
+            img = room_to_rgb_FT(self.room_state, self.box_mapping, self.room_fixed)
+
+        return img
+
+    def step(self, action, observation_mode="rgb_array"):
+        observation, self.reward_last, done, info = super(FixedTargetsSokobanEnv, self).step(
+            action, observation_mode
+        )
+
+        return observation, self.reward_last, done, info
+
+    def _calc_reward(self):
+        self._update_box_mapping()
+
+        # Every step a small penalty is given, This ensures
+        # that short solutions have a higher reward.
+        self.reward_last = self.penalty_for_step
+
+        for b in range(len(self.boxes_are_on_target)):
+            previous_state = self.boxes_are_on_target[b]
+
+            # Calculate new state
+            box_id = list(self.box_mapping.keys())[b]
+            new_state = self.box_mapping[box_id] == box_id
+
+            if previous_state and not new_state:
+                # Box was pushed of its target
+                self.reward_last += self.penalty_box_off_target
+            elif not previous_state and new_state:
+                # box was pushed on its target
+                self.reward_last += self.reward_box_on_target
+
+            self.boxes_are_on_target[b] = new_state
+
+    def _update_box_mapping(self):
+        if self.new_box_position is not None:
+            box_index = list(self.box_mapping.values()).index(self.old_box_position)
+            box_id = list(self.box_mapping.keys())[box_index]
+            self.box_mapping[box_id] = self.new_box_position
+
+    def _check_if_all_boxes_on_target(self):
+        for key in self.box_mapping.keys():
+            if not key == self.box_mapping[key]:
+                return False
+
+        return True