Q-learning

Reinforcement Learning dengan Gymnasium di Python

Fouad Trad

Machine Learning Engineer

Pengantar Q-learning

Singkatan dari quality learning
Teknik tanpa model
Mempelajari Q-table optimal melalui interaksi

Diagram menunjukkan langkah-langkah Q-learning termasuk inisialisasi Q-table, memilih aksi untuk dijalankan, menerima reward dari lingkungan, dan memperbarui tabel. Agen mengulang hingga konvergen setelah sejumlah episode.

Q-learning vs. SARSA

SARSA

Gambar menampilkan rumus pembaruan SARSA.

Memperbarui berdasarkan aksi yang diambil
Pembelajar on-policy

Q-learning

Gambar menampilkan rumus pembaruan Q-learning.

Memperbarui independen dari aksi yang diambil
Pembelajar off-policy

Implementasi Q-learning

env = gym.make("FrozenLake", is_slippery=True)


num_episodes = 1000
alpha = 0.1
gamma = 1


num_states, num_actions = env.observation_space.n, env.action_space.n
Q = np.zeros((num_states, num_actions))


reward_per_random_episode = []

Implementasi Q-learning

for episode in range(num_episodes):
    state, info = env.reset()
    terminated = False
    episode_reward = 0

    while not terminated:

        # Pemilihan aksi acak
        action = env.action_space.sample()

        # Ambil aksi dan amati state dan reward baru
        new_state, reward, terminated, truncated, info = env.step(action)

        # Perbarui Q-table
        update_q_table(state, action, new_state)

        episode_reward += reward
        state = new_state

    reward_per_random_episode.append(episode_reward)

Pembaruan Q-learning

Gambar menampilkan rumus pembaruan Q-learning.

def update_q_table(state, action, reward, new_state):

  old_value = Q[state, action]

  next_max = max(Q[new_state])

  Q[state, action] = (1 - alpha) * old_value + alpha * (reward + gamma * next_max)

Menggunakan kebijakan

reward_per_learned_episode = []

policy = get_policy()

for episode in range(num_episodes):
    state, info = env.reset()
    terminated = False
    episode_reward = 0
    while not terminated:
        # Pilih aksi terbaik berdasarkan Q-table yang telah dipelajari
        action = policy[state]
        # Ambil aksi dan amati state baru
        new_state, reward, terminated, truncated, info = env.step(action)
        state = new_state

        episode_reward += reward

    reward_per_learned_episode.append(episode_reward)

Evaluasi Q-learning

import numpy as np
import matplotlib.pyplot as plt


avg_random_reward = np.mean(reward_per_random_episode)
avg_learned_reward = np.mean(reward_per_learned_episode)


plt.bar(['Random Policy', 'Learned Policy'],
        [avg_random_reward, avg_learned_reward],
        color=['blue', 'green'])

plt.title('Rata-rata Reward per Episode')
plt.ylabel('Rata-rata Reward')
plt.show()

Gambar diagram batang menunjukkan kebijakan yang dipelajari menghasilkan return jauh lebih tinggi (sekitar 0,26) dibanding kebijakan acak (sekitar 0,01).

Ayo berlatih!

Reinforcement Learning dengan Gymnasium di Python