Simple library of reinforcement learning (Q learning and Deep Q Network).

Q learning is a classical algorithm for reinforcement learning and deep Q Network (DQN) is originated from Human-level control through deep reinforcement learning by Volodymyr Mnih et al.

Also, now Double-DQN is appearing soon and Dueling Network in prep.

With this library, you can solve some sample tasks with either algorithm. Other tasks like some games are now in prep.

You can also define your own task with little effort and solve it with q-learning or dqn algorithm easily.

• tqdm

• matplotlib v1.5.1

• numpy v1.10.1

• seaborn

• chainer v1.5.1 +

• Other packages for chainer.

Learning the shortest way to the destination of given field, using q-learning algorithm with epsilon-greedy strategy.

$ python qlearn.py --task "Searchway"

Learning how to swimg up the pendulum, using deep-q network.

$ python qlearn.py --task "Pendulum"

Learning the tasks using Deep Q Network.

$ python dqn.py --task "Searchway"

$ python dqn.py --task "Pendulum"

Now in prep...

shiba24, March, 2016