Final release before split

Releasing before splitting NN module and RL module.

Stack ALE state

Move stacking from agent to env (#294)

* Move stacking from Agent to Env

* Add docstring

* Simplify preprocessor

Fix DQN loss function 2

v0.10.1

Get rid of Cost from DQN (#203)

Fix DQN loss function

Move clipping from cost to agent (#194)

* Move clipping from cost to agent

In the original form

- The form of the loss function was wrong

- It prevents us from computing correct error value

* Add gradient clip test

* Loose cost test

* Fix int test

Add Double DQN

v0.8.0

Add double dqn (#153)

v0.7.0: Update format (#160)

Format Change

v0.6.0: Unify DQN Implementation (#151)

• Cleaned up DQN implementation and equations
• Add bunch of element-wise operations to Tensor

v0.5.0

Fix test log (#132)

Add CartPole

Add CartPole environment

v0.3.0

• Fix Theano Initializers
• Make compatibility between Theano and Tensorflow (almost) 100 %
• Change DQN summary timing unit from episode to update
• Refactor NN components.
• Add layers to run simple MNIST classifier.