Merge branch 'develop' into sync-server

.gitignore

@@ -48,6 +48,7 @@ log/*
 # Python
 **/__pycache__/
 __pycache__/
+*.egg-info
 
 newworld/
 ## Non-static Minetest directories or symlinks to these

README.md

@@ -1,6 +1,6 @@
 EleutherAI Alignment Minetest
 =============================
-This is a fork of the Minetest Voxel Engine, designed to support and OAI gym like environment. The library supports 
+This is a fork of the Minetest Voxel Engine, designed to support an OAI gym like environment. The library supports 
 
 Modes of Operation
 ------------------
@@ -11,6 +11,17 @@ The model provides 4 modes of operation for players and agents
 4. Real time recording of player actions
 
 
+Minetester
+==========
+Minetester is the Python package that exposes Minetest environments via the `gym(nasium)` interface.
+After building the minetest executable you can install it with:
+```
+pip install -e .
+```
+To verify the installation run
+```
+python -m minetester.scripts.test_loop
+```
 
 Minetest
 ========

minetester/__init__.py

@@ -0,0 +1 @@
+from minetester.minetest_env import Minetest

hacking_testing/minetest_env.py → minetester/minetest_env.py

@@ -3,168 +3,16 @@
 import os
 import random
 import shutil
-import subprocess
 import uuid
 from typing import Any, Dict, List, Optional, Tuple
 
 import gym
 import matplotlib.pyplot as plt
 import numpy as np
-import proto_python.objects_pb2 as pb_objects
 import psutil
 import zmq
-from proto_python.objects_pb2 import KeyType
-
-# Define default keys / buttons
-KEY_MAP = {
-    "FORWARD": KeyType.FORWARD,
-    "BACKWARD": KeyType.BACKWARD,
-    "LEFT": KeyType.LEFT,
-    "RIGHT": KeyType.RIGHT,
-    "JUMP": KeyType.JUMP,
-    "SNEAK": KeyType.SNEAK,  # shift key in menus
-    "DIG": KeyType.DIG,  # left mouse button
-    "MIDDLE": KeyType.MIDDLE,  # middle mouse button
-    "PLACE": KeyType.PLACE,  # right mouse button
-    "DROP": KeyType.DROP,
-    "HOTBAR_NEXT": KeyType.HOTBAR_NEXT,  # mouse wheel up
-    "HOTBAR_PREV": KeyType.HOTBAR_PREV,  # mouse wheel down
-    "SLOT_1": KeyType.SLOT_1,
-    "SLOT_2": KeyType.SLOT_2,
-    "SLOT_3": KeyType.SLOT_3,
-    "SLOT_4": KeyType.SLOT_4,
-    "SLOT_5": KeyType.SLOT_5,
-    "SLOT_6": KeyType.SLOT_6,
-    "SLOT_7": KeyType.SLOT_7,
-    "SLOT_8": KeyType.SLOT_8,
-    "INVENTORY": KeyType.INVENTORY,
-}
-INV_KEY_MAP = {value: key for key, value in KEY_MAP.items()}
-
-# Define noop action
-NOOP_ACTION = {key: 0 for key in KEY_MAP.keys()}
-NOOP_ACTION.update({"MOUSE": np.zeros(2, dtype=int)})
-
-
-def unpack_pb_obs(received_obs: str):
-    pb_obs = pb_objects.Observation()
-    pb_obs.ParseFromString(received_obs)
-    obs = np.frombuffer(pb_obs.image.data, dtype=np.uint8).reshape(
-        pb_obs.image.height,
-        pb_obs.image.width,
-        3,
-    )
-    last_action = unpack_pb_action(pb_obs.action) if pb_obs.action else None
-    rew = pb_obs.reward
-    done = pb_obs.terminal
-    # TODO receive extra infos
-    info = {}
-    return obs, rew, done, info, last_action
-
-
-def unpack_pb_action(pb_action: pb_objects.Action):
-    action = dict(NOOP_ACTION)
-    action["MOUSE"] = [pb_action.mouseDx, pb_action.mouseDy]
-    for key_event in pb_action.keyEvents:
-        if key_event.key in INV_KEY_MAP and key_event.eventType == pb_objects.PRESS:
-            key_name = INV_KEY_MAP[key_event.key]
-            action[key_name] = 1
-    return action
-
-
-def pack_pb_action(action: Dict[str, Any]):
-    pb_action = pb_objects.Action()
-    pb_action.mouseDx, pb_action.mouseDy = action["MOUSE"]
-    for key, v in action.items():
-        if key == "MOUSE":
-            continue
-        pb_action.keyEvents.append(
-            pb_objects.KeyboardEvent(
-                key=KEY_MAP[key],
-                eventType=pb_objects.PRESS if v else pb_objects.RELEASE,
-            ),
-        )
-    return pb_action
-
-
-def start_minetest_server(
-    minetest_path: str = "bin/minetest",
-    config_path: str = "minetest.conf",
-    log_path: str = "log/{}.log",
-    server_port: int = 30000,
-    world_dir: str = "newworld",
-    sync_port: int = None,
-    sync_dtime: float = 0.001,
-    game_id: str = "minetest",
-):
-    cmd = [
-        minetest_path,
-        "--server",
-        "--world",
-        world_dir,
-        "--gameid",
-        game_id,
-        "--config",
-        config_path,
-        "--port",
-        str(server_port),
-    ]
-    if sync_port:
-        cmd.extend(["--sync-port", str(sync_port)])
-        cmd.extend(["--sync-dtime", str(sync_dtime)])
-    stdout_file = log_path.format("server_stdout")
-    stderr_file = log_path.format("server_stderr")
-    with open(stdout_file, "w") as out, open(stderr_file, "w") as err:
-        server_process = subprocess.Popen(cmd, stdout=out, stderr=err)
-    return server_process
-
-
-def start_minetest_client(
-    minetest_path: str = "bin/minetest",
-    config_path: str = "minetest.conf",
-    log_path: str = "log/{}.log",
-    client_port: int = 5555,
-    server_port: int = 30000,
-    cursor_img: str = "cursors/mouse_cursor_white_16x16.png",
-    client_name: str = "MinetestAgent",
-    xvfb_headless: bool = False,
-    sync_port: int = None,
-):
-    cmd = [
-        minetest_path,
-        "--name",
-        client_name,
-        "--password",
-        "1234",
-        "--address",
-        "0.0.0.0",  # listen to all interfaces
-        "--port",
-        str(server_port),
-        "--go",
-        "--dumb",
-        "--client-address",
-        "tcp://localhost:" + str(client_port),
-        "--record",
-        "--noresizing",
-        "--config",
-        config_path,
-    ]
-    if xvfb_headless:
-        # hide window
-        cmd.insert(0, "-a")  # allow restarts
-        cmd.insert(0, "xvfb-run")
-        # don't render to screen
-        cmd.append("--headless")
-    if cursor_img:
-        cmd.extend(["--cursor-image", cursor_img])
-    if sync_port:
-        cmd.extend(["--sync-port", str(sync_port)])
-
-    stdout_file = log_path.format("client_stdout")
-    stderr_file = log_path.format("client_stderr")
-    with open(stdout_file, "w") as out, open(stderr_file, "w") as err:
-        client_process = subprocess.Popen(cmd, stdout=out, stderr=err)
-    return client_process
+from minetester.utils import (KEY_MAP, pack_pb_action, start_minetest_client,
+                              start_minetest_server, unpack_pb_obs)
 
 
 class Minetest(gym.Env):
@@ -514,13 +362,6 @@ def step(self, action: Dict[str, Any]):
         return next_obs, rew, done, info
 
     def render(self, render_mode: str = "human"):
-        if render_mode is None:
-            gym.logger.warn(
-                "You are calling render method without specifying any render mode. "
-                "You can specify the render_mode at initialization, "
-                f'e.g. gym("{self.spec.id}", render_mode="rgb_array")',
-            )
-            return
         if render_mode == "human":
             if self.render_img is None:
                 # Setup figure
@@ -540,6 +381,12 @@ def render(self, render_mode: str = "human"):
             plt.draw(), plt.pause(1e-3)
         elif render_mode == "rgb_array":
             return self.last_obs
+        else:
+            raise NotImplementedError(
+                "You are calling 'render()' with an unsupported"
+                f" render mode: '{render_mode}'. "
+                f"Supported modes: {self.metadata['render.modes']}"
+            )
 
     def close(self):
         if self.render_fig is not None:

hacking_testing/test_loop.py → minetester/scripts/test_loop.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-from minetest_env import Minetest
+from minetester import Minetest
 
 env = Minetest(
     seed=42,
@@ -9,7 +9,7 @@
     sync_dtime=0.05,
 )
 
-render = False
+render = True
 obs = env.reset()
 done = False
 while not done:

hacking_testing/test_loop_parallel.py → minetester/scripts/test_loop_parallel.py

@@ -1,7 +1,7 @@
 from typing import Any, Dict, Optional
 
 from gym.wrappers import TimeLimit
-from minetest_env import Minetest
+from minetester import Minetest
 from stable_baselines3.common.vec_env import DummyVecEnv, SubprocVecEnv
 
 if __name__ == "__main__":