Introduction to Ray AIR - RLlib & Coding Exercises

[emmyscode]

New Copy and Code for section on "Reinforcement Learning on Ray AIR" which includes the following major changes:

• Introduction and Brief Primer on Reinforcement Learning
• Better Introduction to RLlib with key features and benefits
• gif of CartPole environment and description of problem statement
• 1(a) Importing Relevant Packages: description, cleaned up and organized imports
• 1(b) Define a Training Function: removed Tuner, simplified code to essentials, description for what an RLTrainer does and how it connects to the other components of Ray AIR
• 1(c) Define an Evaluation Function: walkthrough of evaluation in RL, described termination states, discussed online evaluation, simplified code for evaluation
• 1(d) Online Reinforcement Learning: minimal changes for the final training, evaluation, and rewards
• Added Coding Exercise: asks learner to incorporate Ray Tune into the initial training loop to extend knowledge of Ray AIR to RLlib to give it a more e2e feel.
• Summary and Key Concepts: general copy here

[emmyscode]

Updates to be Made

• Add subsection heading for CartPole example to signify a transition
• Add in the robot diagram that describes agents, states, actions, and environment
• Add highlighted images of the Map of Ray asset for every library (Data, Train, Tune, Serve, RLlib)
• Swap order of code and explanation (code, then explanation)
• Change the style of code explanation to be more bulleted for easier mapping, less sentence-y
• Make note of and investigate purely RLlib examples vs. incorporation with Ray AIR

[emmyscode]

Revise Coding Exercises

For each library, include two coding exercises (one simple, one involved) with solution and explanation for every exercise.