[OSPP-Week2] Carry on building the framework and implement CRR algorithm

[pilgrimygy]

@findmyway
Last week plan

Last week, I did the following work:

1. Build a simple offline framework and CartPole datasets (including different random behavior policies).
2. Test the offline DQN algorithm.
3. Write a simple CQL loss component. After testing, it can greatly improve the performance of offline DQN.

Week 2:
This week, I will try to complete the framework and implement the Critic Regularized Regression (CRR) algorithm. I wanna implement discrete and continuous versions.

[pilgrimygy]

If implementing continuous CRR, maybe I need to write an offline AC algorithm as the benchmark and make a Pendulum Dataset.

[pilgrimygy]

I think that maybe we need to directly use the written online algorithm as an offline benchmark instead of rewriting it. @findmyway

[findmyway]

Agree

[pilgrimygy]

I want to know where you think we should put the dataset.

[findmyway]

Just create a new package under src/ folder named ReinforcementLearningDatasets