v0.2.0: PyTorch RL Modules, more Algorithms, and DMC examples

v0.2.0

Added

• Introduce cherry.nn.Policy, cherry.nn.ActionValue, and cherry.nn.StateValue.
• Algorithm class utilities for: A2C, PPO, TRPO, DDPG, TD3, SAC, and DrQ/DrQv2.
• DMC examples for SAC, DrQ, and DrQv2.
• N-steps returns sampling in ExperienceReplay.

Changed

• Discontinue most of cherry.wrappers.

Fixed

• Fixes return value of StateNormalizer and RewardNormalizer wrappers.
• Requirements to generate docs.

v0.1.3

======

Added
-----

* A CHANGELOG.md file.

Changed
-------

* Travis testing with different versions of Python (3.6, 3.7), torch (1.1, 1.2, 1.3, 1.4), and torchvision (0.3, 0.4, 0.5).

Fixed
-----

* Bugfix when using `td.discount` with replays coming from vectorized environments (@galatolofederico)
* env.action_size and env.state_size when the number of vectorized environments is 1. (thanks @galatolofederico)
* Actor-critic integration test being to finicky.
* `cherry.onehot` support for numpy's float and integer types. (thanks @ngoby)

First open-source release.

This is a first open-source release in order to get feedback on the high-level API choices. Most likely, things will change by the time we reach v0.2.0.