Refactoring interface code

[hill-a]

• bug fixes for some RL algorithms
• changed the structure of the rl_baselines folder
• added a register to rl_baselines, so only one place needs to be updated to add algorithms
• changed all the RL algorithms so they inherit a common class
• added a README.md in rl_baselines folder
• added a register to environments, so only one place needs to be updated to add environments
• added a README.md in environments folder
• added a register to state_representation, so only one place needs to be updated to add SRL models
• added a README.md in state_representation folder
• added checks for incompatible environment and SRL model (joints with mobile robot for example)
• added end-to-end testing (dataset generation -> SRL training -> RL training)

• made the dataset generator deterministic

  :exclamation: This code breaks the loading of old saved RL algorithms in enjoy_baselines.py.

TODO

• [x] update test script for code coverage
• [x] run exhaustive RL/env tests to validate refactoring
• [x] Fix the reset for ddpg and deepq in enjoy_baselines.py
• [x] run SRL tests to validate refactoring

[hill-a]

Test: 9 passed in 3296.36 seconds for 4cac14f

[hill-a]

Running tests, ready for review.

[hill-a]

enjoy_baselines with plotting, ppo2 mlp, relative, shaped-reward on this branch causes this on my personal computer after one episode of the enjoy_baselines script. Updating pybullet changes nothing.

Not quite sure what to make of this, or why it is happening.

[araffin]

@kalifou or @hill-a could you integrate the forward-model branch of srl too ?

[hill-a]

tests: 13 passed in 3555.42 seconds for 2695991, with 71% coverage

[hill-a]

tests: 13 passed in 3475.96 seconds for 824c4e6, 75% coverage

[hill-a]

• [x] remove globals for SAVE_INTERVAL and LOG_INTERVAL