# Error with custom environments

**URL:** <https://discuss.ray.io/t/error-with-custom-environments/16000>\
**Category:** RLlib\
**Created:** [October 14, 2024, 9:02pm UTC](https://discuss.ray.io/t/error-with-custom-environments/16000 "2024-10-14T21:02:14Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Kush\_Patel](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/kush_patel/32/6634_2.png) [@Kush\_Patel](https://discuss.ray.io/u/Kush_Patel)\
**Post date:** [October 14, 2024, 9:02pm UTC](https://discuss.ray.io/t/error-with-custom-environments/16000/1 "2024-10-14T21:02:15Z")

</div>

from ray.rllib.algorithms.ppo import PPOConfig

def initialize\_ppo\_trainer():  
global ppo\_trainer

```
register_env("custom_env", lambda config2: CustomEnv(config2))

# Configure PPO
ppo_config = (
    PPOConfig()
    .environment("custom_env") # Name of the registered environment
    .framework("torch") # Use torch or tf depending on your setup
    .rollouts(num_rollout_workers=0, create_env_on_local_worker=True) # Adjust number of workers
    .training(
        train_batch_size=BATCH_SIZE,
        sgd_minibatch_size=min(32, BATCH_SIZE), # Minibatch size <= train batch size
    )
    .resources(num_gpus=0) # Adjust GPU resources as needed
)

# Build the PPO trainer
ppo_trainer = ppo_config.build()
logger.info("PPO trainer initialized.")

```

CustomEnv:

class CustomEnv(gym.Env):  
“”"  
Custom Environment for Incident Response using OpenAI Gym interface.  
“”"

```
metadata = {'render.modes': ['human']}

def __init__ (self, config: Dict[str, Any] = None):
    super(CustomEnv, self). __init__ ()
    
    # Define action and observation space
    # Actions: Discrete actions corresponding to incident response strategies

    self.action_space = spaces.Discrete(len(ACTIONS)) # len(ACTIONS) = 9
    
    # Observation space: 77 continuous features, normalized between 0 and 1
    # Adjust the low and high values based on actual data ranges for better normalization
    self.observation_space = spaces.Box(
        low=0.0, 
        high=1.0, 
        shape=(len(STATE_FEATURES),), 
        dtype=np.float32
    )

    print(f"CustomEnv Initialized: Observation space - {self.observation_space}, Action space - {self.action_space}")
    logger.info(f"CustomEnv Initialized: Observation space - {self.observation_space}, Action space - {self.action_space}")

    
    # Initialize state
    self.state = self.reset()
    
    # Optional: Define additional parameters from config if needed
    if config is not None:
        self.config = config
    else:
        self.config = {}

```

ValueError: `observation_space` not provided in PolicySpec for default\_policy and env does not have an observation space OR no spaces received from other workers’ env(s) OR no `observation_space` specified in config!

---

<div class="post-metadata">

**Author:** ![Kush\_Patel](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/kush_patel/32/6634_2.png) [@Kush\_Patel](https://discuss.ray.io/u/Kush_Patel)\
**Post date:** [October 14, 2024, 9:21pm UTC](https://discuss.ray.io/t/error-with-custom-environments/16000/2 "2024-10-14T21:21:06Z")

</div>

Even on this simple code, gives me it:

import ray  
from ray.rllib.algorithms.ppo import PPOConfig  
from ray.tune.registry import register\_env  
import gym

# Initialize Ray

ray.init(ignore\_reinit\_error=True)

# Define the default gym environment creator

def env\_creator(env\_config):  
return gym.make(“CartPole-v1”)

# Register the environment

register\_env(“cartpole\_env”, env\_creator)

# Create a PPO configuration and train the agent

config = PPOConfig().environment(env=“cartpole\_env”).framework(“torch”).rollouts(num\_rollout\_workers=1)

# Build the PPO trainer using the config

trainer = config.build()

# Train for a few iterations

for i in range(3):  
result = trainer.train()  
print(f"Iteration {i + 1}: reward = {result[‘episode\_reward\_mean’]}")

# Cleanup Ray

ray.shutdown()