I am creating a custom gymnasium enviroment. gymnasium is more formal than OpenAI gym.
And I need to be able to import gymnasium but I cannot. And I can do this with Google colab
and Jupyter Lab.
I think Intel provides jupyter notebook enviroment right ?
If import gymnasium works I will solve my problem.
As I posted in my original question I install the required packages;
!pip install gymnasium stable_baselines3 Shimmy
My code:
import gymnasium
def sorting_action_constraint_fn(action):
# Customize the constraint logic based on your requirements
i, j= action
return i != j
class GymnasiumSortingEnv(gymnasium.Env):
def __init__(self, sequence_length):
super(GymnasiumSortingEnv, self).__init__()
self.sequence_length = sequence_length
self.observation_space = gymnasium.spaces.Box(low=0, high=1, shape=(sequence_length,), dtype=np.float32)
self.action_space = gymnasium.spaces.MultiDiscrete([sequence_length, sequence_length])
self.current_sequence = np.zeros(sequence_length)
#self.reset()
def reset(self,seed=None, options=None):
self.current_sequence = np.random.rand(self.sequence_length)
#return self.current_sequence.copy()
return self.current_sequence.copy(), {}
def is_valid_action(self, action):
# Define your constraint function here
return sorting_action_constraint_fn(action)
def step(self, action):
# Validate the action
assert self.action_space.contains(action), f"Invalid action: {action}"
# Swap the elements at the specified indices
i, j = action
#if not sorting_action_constraint_fn(action):
# raise ValueError("Invalid action. Please choose a valid action.")
self.current_sequence[i], self.current_sequence[j] = self.current_sequence[j], self.current_sequence[i]
# Calculate reward
#reward = -np.sum(np.abs(np.diff(self.current_sequence))) # Negative sum of absolute differences
#reward = custom_distance_metric(self.target_sequence, self.current_sequence)
reward = 1 if np.all(np.diff(self.current_sequence) >= 0) else 0
# Check if the sequence is sorted
done = np.all(np.diff(self.current_sequence) >= 0)
# Return the new state, reward, whether the episode is done, and additional info
return self.current_sequence.copy(), reward, done, False, {}
def render(self):
print("Current Sequence:", self.current_sequence)
