Update testing

pettingzoo/classic/rlcard_envs/texas_holdem_no_limit.py

@@ -151,7 +151,7 @@ def __init__(self, num_players=2, render_mode=None):
                                 ),
                                 [100, 100],
                             ),
-                            dtype=np.float64,
+                            dtype=np.float32,
                         ),
                         "action_mask": spaces.Box(
                             low=0, high=1, shape=(self.env.num_actions,), dtype=np.int8

pettingzoo/test/api_test.py

@@ -1,4 +1,3 @@
-import random
 import re
 import warnings
 from collections import defaultdict
@@ -316,8 +315,6 @@ def play_test(env, observation_0, num_cycles):
     """
     env.reset()
 
-    terminated = {agent: False for agent in env.agents}
-    truncated = {agent: False for agent in env.agents}
     live_agents = set(env.agents[:])
     has_finished = set()
     generated_agents = set()
@@ -333,10 +330,13 @@ def play_test(env, observation_0, num_cycles):
         prev_observe, reward, terminated, truncated, info = env.last()
         if terminated or truncated:
             action = None
-        elif isinstance(prev_observe, dict) and "action_mask" in prev_observe:
-            action = random.choice(np.flatnonzero(prev_observe["action_mask"]).tolist())
         else:
-            action = env.action_space(agent).sample()
+            mask = (
+                prev_observe.get("action_mask")
+                if isinstance(prev_observe, dict)
+                else None
+            )
+            action = env.action_space(agent).sample(mask)
 
         if agent not in live_agents:
             live_agents.add(agent)
@@ -382,15 +382,17 @@ def play_test(env, observation_0, num_cycles):
         if not env.agents:
             break
 
-        if isinstance(env.observation_space(agent), gymnasium.spaces.Box):
-            assert env.observation_space(agent).dtype == prev_observe.dtype
         assert env.observation_space(agent).contains(
             prev_observe
         ), "Out of bounds observation: " + str(prev_observe)
 
-        assert env.observation_space(agent).contains(
-            prev_observe
-        ), "Agent's observation is outside of it's observation space"
+        if isinstance(env.observation_space(agent), gymnasium.spaces.Box):
+            assert env.observation_space(agent).dtype == prev_observe.dtype
+        elif isinstance(env.observation_space(agent), gymnasium.spaces.Dict):
+            assert (
+                env.observation_space(agent)["observation"].dtype
+                == prev_observe["observation"].dtype
+            )
         test_observation(prev_observe, observation_0, str(env.unwrapped))
         if not isinstance(env.infos[env.agent_selection], dict):
             warnings.warn(
@@ -407,10 +409,9 @@ def play_test(env, observation_0, num_cycles):
         obs, reward, terminated, truncated, info = env.last()
         if terminated or truncated:
             action = None
-        elif isinstance(obs, dict) and "action_mask" in obs:
-            action = random.choice(np.flatnonzero(obs["action_mask"]).tolist())
         else:
-            action = env.action_space(agent).sample()
+            mask = obs.get("action_mask") if isinstance(obs, dict) else None
+            action = env.action_space(agent).sample(mask)
         assert isinstance(terminated, bool), "terminated from last is not True or False"
         assert isinstance(truncated, bool), "terminated from last is not True or False"
         assert (
@@ -439,7 +440,7 @@ def test_action_flexibility(env):
         if terminated or truncated:
             action = None
         elif isinstance(obs, dict) and "action_mask" in obs:
-            action = random.choice(np.flatnonzero(obs["action_mask"]).tolist())
+            action = env.action_space(agent).sample(obs["action_mask"])
         else:
             action = 0
         env.step(action)
@@ -481,6 +482,9 @@ def progress_report(msg):
 
     env.reset()
     observation_0, *_ = env.last()
+    if isinstance(observation_0, dict) and "observation" in observation_0:
+        observation_0 = observation_0["observation"]
+
     test_observation(observation_0, observation_0, str(env.unwrapped))
 
     non_observe, *_ = env.last(observe=False)

pettingzoo/test/example_envs/generated_agents_env_v0.py

@@ -75,12 +75,25 @@ def reset(self, seed=None, options=None):
         self.truncations = {}
         self.infos = {}
         self.num_steps = 0
+
+        self._obs_spaces = {}
+        self._act_spaces = {}
+        self.types = []
+        self._agent_counters = {}
+        for i in range(3):
+            self.add_type()
         for i in range(5):
             self.add_agent(self.np_random.choice(self.types))
 
         self._agent_selector = agent_selector(self.agents)
         self.agent_selection = self._agent_selector.reset()
 
+        # seed observation and action spaces
+        for i, agent in enumerate(self.agents):
+            self.observation_space(agent).seed(seed)
+        for i, agent in enumerate(self.agents):
+            self.action_space(agent).seed(seed)
+
     def seed(self, seed=None):
         self.np_random, _ = gymnasium.utils.seeding.np_random(seed)
 

pettingzoo/test/example_envs/generated_agents_parallel_v0.py

@@ -30,6 +30,7 @@ def __init__(self, max_cycles=100, render_mode=None):
         self.types = []
         self._agent_counters = {}
         self.max_cycles = max_cycles
+        self.rng_seed = None
         self.seed()
         self.render_mode = render_mode
         for i in range(3):
@@ -50,6 +51,8 @@ def add_type(self):
         obs_size = self.np_random.integers(10, 50)
         obs_space = gymnasium.spaces.Box(low=0, high=1, shape=(obs_size,))
         act_space = gymnasium.spaces.Discrete(num_actions)
+        obs_space.seed(self.rng_seed)
+        act_space.seed(self.rng_seed)
         new_type = f"type{type_id}"
         self.types.append(new_type)
         self._obs_spaces[new_type] = obs_space
@@ -65,12 +68,31 @@ def add_agent(self, type):
         return agent_name
 
     def reset(self, seed=None, options=None):
+        self.rng_seed = seed
+
         if seed is not None:
             self.seed(seed=seed)
-        self.agents = []
         self.num_steps = 0
+
+        # Reset spaces and types
+        self._obs_spaces = {}
+        self._act_spaces = {}
+        self.types = []
+        self._agent_counters = {}
+        for i in range(3):
+            self.add_type()
+
+        # Add agents
+        self.agents = []
         for i in range(5):
             self.add_agent(self.np_random.choice(self.types))
+
+        # seed observation and action spaces
+        for i, agent in enumerate(self.agents):
+            self.observation_space(agent).seed(seed)
+        for i, agent in enumerate(self.agents):
+            self.action_space(agent).seed(seed)
+
         return {agent: self.observe(agent) for agent in self.agents}
 
     def seed(self, seed=None):

pettingzoo/test/render_test.py

@@ -1,5 +1,3 @@
-import random
-
 import numpy as np
 
 
@@ -14,7 +12,7 @@ def collect_render_results(env):
                 if terminated or truncated:
                     action = None
                 elif isinstance(obs, dict) and "action_mask" in obs:
-                    action = random.choice(np.flatnonzero(obs["action_mask"]).tolist())
+                    action = env.action_space(agent).sample(obs["action_mask"])
                 else:
                     action = env.action_space(agent).sample()
                 env.step(action)

pettingzoo/test/seed_test.py

@@ -1,109 +1,128 @@
-import hashlib
-import pickle
-import random
-import warnings
-
-import numpy as np
-
-from pettingzoo.utils import parallel_to_aec
-
-
-def hash(val):
-    val = pickle.dumps(val)
-    hasher = hashlib.md5()
-    hasher.update(val)
-    return hasher.hexdigest()
-
-
-def calc_hash(new_env, rand_issue, max_env_iters):
-    cur_hashes = []
-    sampler = random.Random(42)
-    for i in range(3):
-        new_env.reset(seed=i)
-        for j in range(rand_issue + 1):
-            random.randint(0, 1000)
-            np.random.normal(size=100)
-        for agent in new_env.agent_iter(max_env_iters):
-            obs, rew, terminated, truncated, info = new_env.last()
-            if terminated or truncated:
-                action = None
-            elif isinstance(obs, dict) and "action_mask" in obs:
-                action = sampler.choice(np.flatnonzero(obs["action_mask"]).tolist())
-            else:
-                action = new_env.action_space(agent).sample()
-            new_env.step(action)
-            cur_hashes.append(agent)
-            cur_hashes.append(hash_obsevation(obs))
-            cur_hashes.append(float(rew))
-
-    return hash(tuple(cur_hashes))
+from gymnasium.utils.env_checker import data_equivalence
 
 
 def seed_action_spaces(env):
-    if hasattr(env, "possible_agents"):
-        for i, agent in enumerate(env.possible_agents):
+    if hasattr(env, "agents"):
+        for i, agent in enumerate(env.agents):
             env.action_space(agent).seed(42 + i)
 
 
+def seed_observation_spaces(env):
+    if hasattr(env, "agents"):
+        for i, agent in enumerate(env.agents):
+            env.observation_space(agent).seed(42 + i)
+
+
 def check_environment_deterministic(env1, env2, num_cycles):
-    """
-    env1 and env2 should be seeded environments
+    """Check that two AEC environments execute the same way."""
 
-    returns a bool: true if env1 and env2 execute the same way
-    """
+    env1.reset(seed=42)
+    env2.reset(seed=42)
 
-    # seeds action space so that actions are deterministic
+    # seed action spaces to ensure sampled actions are the same
     seed_action_spaces(env1)
     seed_action_spaces(env2)
 
-    num_agents = max(1, len(getattr(env1, "possible_agents", [])))
+    # seed observation spaces to ensure first observation is the same
+    seed_observation_spaces(env1)
+    seed_observation_spaces(env2)
+
+    iter = 0
+    max_env_iters = num_cycles * len(env1.agents)
+
+    for agent1, agent2 in zip(env1.agent_iter(), env2.agent_iter()):
+        assert data_equivalence(agent1, agent2), f"Incorrect agent: {agent1} {agent2}"
+
+        obs1, reward1, termination1, truncation1, info1 = env1.last()
+        obs2, reward2, termination2, truncation2, info2 = env2.last()
+
+        assert data_equivalence(obs1, obs2), "Incorrect observation"
+        assert data_equivalence(reward1, reward2), "Incorrect reward."
+        assert data_equivalence(termination1, termination2), "Incorrect termination."
+        assert data_equivalence(truncation1, truncation2), "Incorrect truncation."
+        assert data_equivalence(info1, info2), "Incorrect info."
 
-    # checks deterministic behavior if seed is set
-    hashes = []
-    num_seeds = 2
-    max_env_iters = num_cycles * num_agents
-    envs = [env1, env2]
-    for x in range(num_seeds):
-        hashes.append(calc_hash(envs[x], x, max_env_iters))
+        if termination1 or truncation1:
+            break
 
-    return all(hashes[0] == h for h in hashes)
+        mask1 = obs1.get("action_mask") if isinstance(obs1, dict) else None
+        mask2 = obs2.get("action_mask") if isinstance(obs2, dict) else None
+        assert data_equivalence(mask1, mask2), f"Incorrect action mask: {mask1} {mask2}"
 
+        action1 = env1.action_space(agent1).sample(mask1)
+        action2 = env2.action_space(agent2).sample(mask2)
 
-def hash_obsevation(obs):
-    try:
-        val = hash(obs.tobytes())
-        return val
-    except AttributeError:
-        try:
-            return hash(obs)
-        except TypeError:
-            warnings.warn("Observation not an int or an Numpy array")
-            return 0
+        assert data_equivalence(
+            action1, action2
+        ), f"Incorrect actions: {action1} {action2}"
 
+        env1.step(action1)
+        env2.step(action2)
 
-def test_environment_reset_deterministic(env1, num_cycles):
+        iter += 1
+
+        if iter >= max_env_iters:
+            break
+
+    env1.close()
+    env2.close()
+
+
+def check_environment_deterministic_parallel(env1, env2, num_cycles):
+    """Check that two parallel environments execute the same way."""
+    env1.reset(seed=42)
+    env2.reset(seed=42)
+
+    # seed action spaces to ensure sampled actions are the same
     seed_action_spaces(env1)
-    hash1 = calc_hash(env1, 1, num_cycles)
+    seed_action_spaces(env2)
+
+    # seed observation spaces to ensure first observation is the same
+    seed_observation_spaces(env1)
+    seed_observation_spaces(env2)
+
+    iter = 0
+    max_env_iters = num_cycles * len(env1.agents)
+
+    env1.reset(seed=42)
+    env2.reset(seed=42)
+
     seed_action_spaces(env1)
-    hash2 = calc_hash(env1, 2, num_cycles)
-    assert hash1 == hash2, "environments kept state after and reset(seed)"
+    seed_action_spaces(env2)
+
+    while env1.agents:
+        actions1 = {agent: env1.action_space(agent).sample() for agent in env1.agents}
+        actions2 = {agent: env2.action_space(agent).sample() for agent in env2.agents}
+
+        assert data_equivalence(actions1, actions2), "Incorrect action seeding"
+
+        obs1, rewards1, terminations1, truncations1, infos1 = env1.step(actions1)
+        obs2, rewards2, terminations2, truncations2, infos2 = env2.step(actions2)
+
+        iter += 1
+
+        assert data_equivalence(obs1, obs2), "Incorrect observations"
+        assert data_equivalence(rewards1, rewards2), "Incorrect values for rewards"
+        assert data_equivalence(terminations1, terminations2), "Incorrect terminations."
+        assert data_equivalence(truncations1, truncations2), "Incorrect truncations"
+        assert data_equivalence(infos1, infos2), "Incorrect infos"
+
+        if iter >= max_env_iters or any(terminations1) or any(truncations1):
+            break
+
+    env1.close()
+    env2.close()
 
 
-def seed_test(env_constructor, num_cycles=10, test_kept_state=True):
+def seed_test(env_constructor, num_cycles=10):
     env1 = env_constructor()
-    if test_kept_state:
-        test_environment_reset_deterministic(env1, num_cycles)
     env2 = env_constructor()
 
-    assert check_environment_deterministic(
-        env1, env2, num_cycles
-    ), "The environment gives different results on multiple runs when initialized with the same seed. This is usually a sign that you are using np.random or random modules directly, which uses a global random state."
+    check_environment_deterministic(env1, env2, 500)
 
 
-def parallel_seed_test(parallel_env_fn, num_cycles=10, test_kept_state=True):
-    def aec_env_fn():
-        parallel_env = parallel_env_fn()
-        env = parallel_to_aec(parallel_env)
-        return env
+def parallel_seed_test(parallel_env_fn):
+    env1 = parallel_env_fn()
+    env2 = parallel_env_fn()
 
-    seed_test(aec_env_fn, num_cycles, test_kept_state)
+    check_environment_deterministic_parallel(env1, env2, 500)