🏋️ Gymnasium Integration¶
This guide explains how to use Gymnasium environments with PAMIQ Core framework.
Installation¶
To use the Gymnasium integration, install PAMIQ Core with the gym extra:
Overview¶
The pamiq_core.gym module provides seamless integration between Gymnasium (the fork of OpenAI Gym) and PAMIQ Core's interaction system. This allows you to use any Gymnasium environment as a PAMIQ Core environment.
Key Components¶
GymEnvironment¶
GymEnvironment wraps any Gymnasium environment to make it compatible with PAMIQ Core's Environment interface:
from pamiq_core.gym import GymEnvironment
# Create from environment ID
env = GymEnvironment("CartPole-v1")
# Or use existing Gymnasium environment
import gymnasium as gym
gym_env = gym.make("CartPole-v1", render_mode="human")
env = GymEnvironment(gym_env)
GymAgent¶
GymAgent is an abstract base class for implementing agents that interact with Gymnasium environments:
from pamiq_core.gym import GymAgent
import numpy as np
class MyCartPoleAgent(GymAgent[np.ndarray, int]):
def on_reset(self, obs, info):
"""Called when environment resets."""
return 0 # Initial action
def on_step(self, obs, reward, terminated, truncated, info):
"""Called at each environment step."""
# Simple policy: move right if pole is tilting right
return 1 if obs[2] > 0 else 0
Advanced Features¶
Manual Environment Reset¶
Agents can request environment reset by setting the need_reset flag:
class EarlyStoppingAgent(GymAgent[np.ndarray, int]):
def on_step(self, obs, reward, terminated, truncated, info):
# Request reset if pole angle is too large
if abs(obs[2]) > 0.5:
self.need_reset = True
return 1 if obs[2] > 0 else 0
API Reference¶
More details, Checkout to the API Reference