This is a straightforward adaptation of the baselines DQN implementation to conform to the RLlib API. Files to pay attention to are rllib/dqn/dqn.py and rllib/dqn/example.py; the rest were mostly copied with linter fixes only.

I also fixed up the licensing here by appending the OpenAI MIT license to the top-level LICENSE file.

I have a couple ideas on how to parallelize this with Ray in a followup PR:

• First, we can parallelize rollouts, however to preserve algorithm semantics this requires train_freq to be large enough to allow sufficient parallelism between training steps. Increasing train_freq will probably also require an equivalent increase of batch_size.
• Second, we can parallelize the optimization step. This also requires the batch_size parameter to be increased. We might also consider multiple steps of optimization over replay buffer samples, similar to policy gradient.

There is also literature on parallelizing DQN in other ways but that might be out of scope for now.

On a GPU instance the Pong example spends about equal time in training and rollouts, so both could be potentially valuable.

cc @pcmoritz @royf