Update repo to work with gymnasium

baseline_test.py

@@ -1,4 +1,4 @@
-import gym
+import gymnasium as gym
 
 from stable_baselines.common.env_checker import check_env
 from stable_baselines.common.policies import MlpPolicy
@@ -27,7 +27,7 @@
 
 model = A2C.load("blokus_weights")
 
-obs = env.reset()
+obs = envreset()
 while True:
     action, _states = model.predict(obs)
     obs, rewards, dones, info = env.step(action)

blokus_gym/__init__.py

@@ -1,4 +1,9 @@
-from gym.envs.registration import register
+from gymnasium.envs.registration import register
+
+register(
+    id='blokus-custom-v0',
+    entry_point='blokus_gym.envs:BlokusCustomEnv',
+)
 
 register(
     id='blokus-hard-v0',

blokus_gym/envs/__init__.py

@@ -1,2 +1,2 @@
 from blokus_gym.envs.blokus_env import BlokusEnv
-from blokus_gym.envs.blokus_envs import BlokusSimpleEnv, BlokusSimpleGreedyEnv, BlokusDuoEnv, BlokusDuoGreedyEnv, BlokusGreedyEnv, BlokusSimpleMinimaxEnv
+from blokus_gym.envs.blokus_envs import BlokusSimpleEnv, BlokusSimpleGreedyEnv, BlokusDuoEnv, BlokusDuoGreedyEnv, BlokusGreedyEnv, BlokusSimpleMinimaxEnv, BlokusCustomEnv

blokus_gym/envs/blokus_env.py

@@ -7,8 +7,10 @@
 from functools import partial
 import os
 import matplotlib.pyplot as plt
-import gym
+import gymnasium as gym
 import cython
+import torch.random
+
 from blokus_gym.envs.game.blokus_game import InvalidMoveByAi
 from blokus_gym.envs.game.blokus_game import BlokusGame
 from blokus_gym.envs.game.board import Board
@@ -25,7 +27,7 @@ def possible_moves_func(dummy, board_size, pieces):
 
 
 class BlokusEnv(gym.Env):
-    metadata = {'render.modes': ['human']}
+    metadata = {'render_modes': ['human'], 'render_fps' : 2}
     rewards = {'won': 1, 'tie-won': 0, 'default': 0, 'invalid': -100, 'lost': -1}
     STATES_FOLDER = "states"
 
@@ -118,7 +120,11 @@ def __get_done_reward(self):
 
         return done, reward
 
-    def reset(self):
+    def reset(self, seed=None, options=None):
+        if seed is not None:
+            np.random.seed(seed)
+            torch.random.seed(seed)
+            random.seed(seed)
         self.init_game()
         return self.blokus_game.board.tensor
 

blokus_gym/envs/blokus_envs.py

@@ -2,7 +2,15 @@
 from blokus_gym.envs.shapes.shapes import get_all_shapes
 from blokus_gym.envs.players.greedy_player import GreedyPlayer
 from blokus_gym.envs.players.minimax_player import MinimaxPlayer
-
+from blokus_gym.envs.players.random_player import RandomPlayer
+
+class BlokusCustomEnv(BlokusEnv):
+    NUMBER_OF_PLAYERS = 3
+    BOARD_SIZE = 10 # This will result in a 10x10 board
+    STATES_FILE = "states.json"  # This needs to be set, if not it will take the base class states
+    all_shapes = [shape for shape in get_all_shapes()
+                  if shape.size == 4]  # This will take only the 4 tiles pieces
+    bot_type = RandomPlayer  # Defaults to RandomPlayer if not passed
 
 class BlokusGreedyEnv(BlokusEnv):
     bot_type = GreedyPlayer

play.py

@@ -1,6 +1,6 @@
 import blokus_gym
 from blokus_gym.envs.blokus_env import BlokusEnv
-import gym
+import gymnasium as gym
 import random
 
 if __name__ == "__main__":
@@ -9,15 +9,16 @@
     print(f"number of possible moves {env.action_space}")
     count = 0
     nb_rounds = 0
-    for _ in range(100):
+    env.reset()
+    for _ in range(10):
         while True:
             # input()
             action = env.action_space.sample()
             count += len(env.ai_possible_indexes())
             nb_rounds += 1
             # action = random.randint(0, 918)
             observation, reward, done, info = env.step(action)
-            env.render("human")
+            env.render()
             # print(env.ai.all_ids_to_move.keys())
             # print(reward)
 

play_custom.py

@@ -0,0 +1,35 @@
+import blokus_gym
+from blokus_gym.envs.blokus_env import BlokusEnv
+import gymnasium as gym
+import random
+
+if __name__ == "__main__":
+    # env = BlokusEnv()
+    env = gym.make("blokus_gym:blokus-custom-v0")  # Make sure to do: pip install -e . in root
+    print(f"number of possible moves {env.action_space}")
+    count = 0
+    nb_rounds = 0
+    for _ in range(100):
+        while True:
+            # input()
+            action = env.action_space.sample()
+            count += len(env.ai_possible_indexes())
+            nb_rounds += 1
+            # action = random.randint(0, 918)
+            observation, reward, done, info = env.step(action)
+            env.render("human")
+            # print(env.ai.all_ids_to_move.keys())
+            # print(reward)
+
+            if done:
+                # print(env.ai.all_ids_to_move.keys())
+                # input()
+                print(f"{'won' if reward == 1 else ('tie-won' if reward == 0 else 'lost')}")
+                # print(env.ai.all_ids_to_move.keys())
+                observation = env.reset()
+                # input()
+                break
+
+    print(f"Average number of moves per turn: {count / nb_rounds:.2f}")
+    print(f"Starter won {env.starter_won / env.games_played * 100:.2f}%")
+    env.close()

rainbow/agent.py

@@ -1,6 +1,6 @@
 import os
 
-import gym
+import gymnasium as gym
 import torch
 import torch.optim as optim
 import torch.nn.functional as F

rainbow/automated_training.py

@@ -1,4 +1,4 @@
-import gym
+import gymnasium as gym
 import pandas as pd
 
 from rainbow.agent import Agent

rainbow/test_model.py

@@ -1,4 +1,4 @@
-import gym
+import gymnasium as gym
 from rainbow.agent import Agent
 
 if __name__ == "__main__":

requirements.txt

@@ -1,5 +1,10 @@
-numpy>=1.17.4
-gym>=0.17.1
-torch>=1.4.0
+cython>=0.29.16
+gym==0.15.7
 matplotlib>=3.1.2
-cython>=0.29.16
+numpy>=1.17.4
+opencv-python==4.5.3.56
+pandas>=1.1.5
+setuptools>=58.0.4
+stable_baselines>=2.10.2
+tensorflow==1.15.2
+torch>=1.4.0