Rainforcement Learning

This repository is for studying rainforcement learning. I execute this program in following environment.

4.15.0-36-generic
Python 2.7.12
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=16.04
DISTRIB_CODENAME=xenial
DISTRIB_DESCRIPTION="Ubuntu 16.04.4 LTS"

cartpole_manual.py

This program is playing cartpole in manual.
You can know how difficult this game is.

How to use
$ python manual_cartpole.py  
in terminal,
press -a- the cart move left  
press -d- the cart move right  

cartpole_ql.py

This program is playing cartpole using q-learning.

How to use  
$ python cartpole_ql.py  
you can see learned motion in cartpole.

cartpole_dqn.py

This program is playing cartpole using deep-q-network. You can also select double deep-q-network.

How to use
$ python cartpole_dqn.py

If you need to use ddqn, you have to change following line in cartpole_dqn.py.

DQN_MODE = 0    # 0:DDQN 1:DQN

If program operated correctly, you can see the reward mean as following.

Sometime (about once to 10 times) learning is failed as following.