代码阅读笔记

# DQN
discount_factor = 0.99
learning_rate = 0.001
epsilon = 1.0
epsilon_decay = 0.999
epsilon_min = 0.01
batch_size = 64
train_start = 1000
memory = deque(maxlen=2000)
mini_batch = random.sample(self.memory, batch_size)

# PG
discount_factor = 0.99
learning_rate = 0.001
discounted_rewards -= np.mean(discounted_rewards)
discounted_rewards /= np.std(discounted_rewards)

# A2C
discount_factor = 0.99
actor_lr = 0.001
critic_lr = 0.005

# A3C
actor_lr = 0.001
critic_lr = 0.001
discount_factor = .99
hidden1, self.hidden2 = 24, 24
threads = 8

原来的 README

---

Minimal and clean examples of reinforcement learning algorithms presented by RLCode team.

Maintainers - Woongwon, Youngmoo, Hyeokreal, Uiryeong, Keon

From the basics to deep reinforcement learning, this repo provides easy-to-read code examples. One file for each algorithm. Please feel free to create a Pull Request, or open an issue!

Dependencies

1. Python 3.5
2. Tensorflow 1.0.0
3. Keras
4. numpy
5. pandas
6. matplot
7. pillow
8. Skimage
9. h5py

Install Requirements

pip install -r requirements.txt

Table of Contents

Grid World - Mastering the basics of reinforcement learning in the simplified world called "Grid World"

• Policy Iteration
• Value Iteration
• Monte Carlo
• SARSA
• Q-Learning
• Deep SARSA
• REINFORCE

CartPole - Applying deep reinforcement learning on basic Cartpole game.

• Deep Q Network
• Double Deep Q Network
• Policy Gradient
• Actor Critic (A2C)
• Asynchronous Advantage Actor Critic (A3C)

Atari - Mastering Atari games with Deep Reinforcement Learning

• Breakout - DQN, DDQN Dueling DDQN A3C
• Pong - Policy Gradient

OpenAI GYM - [WIP]

• Mountain Car - DQN