Reinforcement Learning Algorithms Tutorial

Enviroment

macOS 10.13.6
python 3.5.6

Requirements.txt

numpy==1.14.5
tensorflow==1.12.1
gym==0.12.1

Command for install

pip3 install -r requirements.txt

Q-Learning

Medium Tutorial

The complete tutorial is released.

Problem Definition

We use FrozenLake-v0 without slippery and define it in class Environment

Basic Q-Learning (with table method)

python3 frozenlake_unslippery.py --max_episodes=2000 --max_actions=99 --discount=0.95 --exploration_rate=1.0

Deep Q-Learning (Deep Q Network)

max_episodes ≥ 15000.

python3 deep_frozenlake_unslippery.py --max_episodes=20000 --max_actions=99 --discount=0.95 --exploration_rate=1.0 --hidden_units=10

Double Q-Learning (Double Q Network)

The complete tutorial is released.

Problem Definition

Take CartPole-v1 as example

run

python3 main.py --max_episodes=20000 --max_actions=10000 --exploration_rate=1.0 --exploration_decay=0.0001 --batch_size=64