Fix observations on PPO trainer (Unity-Technologies#340)

* Fix observations on PPO trainer

* tested and fixed the fix

Require tensorflow 1.4.1 (Unity-Technologies#315)

* Require tensorflow 1.4.1

* modified the python/README.md

Initial support for multiple observations (Unity-Technologies#256)

* Initial support for multiple observations

* Fix PPO for continuous control

Bug fix multiple observations

Bug fix multiple observations

Add new flags to ppo best practices

Update installation.md

Merge pull request Unity-Technologies#16 from ASPePeX/keep-checkpoints

--keep-checkpoints=<n> option for ppo.py

formating and added documentation