ComeçarComece de graça

Solving 8x8 Frozen Lake with Q-learning

In this exercise, you'll apply the Q-learning algorithm to learn an optimal policy for navigating through the 8x8 Frozen Lake environment, this time with the "slippery" condition enabled. The challenge introduces stochastic transitions, making the agent's movement unpredictable and thus more closely simulating real-world scenarios.

A Q-table Q has been initialized and pre-loaded for you, along with the update_q_table() function from the previous exercise and an empty list rewards_per_episode that will contain the total reward accumulated through each episode.

Este exercício faz parte do curso

Reinforcement Learning with Gymnasium in Python

Ver curso

Instruções do exercício

  • For each episode, execute the selected action and observe the reward and next state.
  • Update the Q-table.
  • Append the total_reward to the rewards_per_episode list.

Exercício interativo prático

Experimente este exercício completando este código de exemplo.

for episode in range(10000):
    state, info = env.reset()
    total_reward = 0
    terminated = False
    while not terminated:
        action = env.action_space.sample()
        # Execute the action
        next_state, reward, terminated, truncated, info = ____
        # Update the Q-table
        ____
        state = next_state
        total_reward += reward
    # Append the total reward to the rewards list    
    rewards_per_episode.____(____)
print("Average reward per random episode: ", np.mean(rewards_per_episode))
Editar e executar o código