Mastering TensorFlow 1.x

Reinforcement learning techniques

Reinforcement learning techniques can be categorized on the basis of the availability of the model as follows:

• Model is available: If the model is available then the agent can plan offline by iterating over policies or the value function to find the optimal policy that gives the maximum reward.
  • Value-iteration learning: In the value-iteration learning approach, the agent starts by initializing the V(s) to a random value and then repeatedly updates the V(s) until a maximum reward is found.
  • Policy-iterative learning: In the policy-iteration learning approach, the agent starts by initializing a random policy p, and then repeatedly updates the policy until a maximum reward is found.

• Model is not available ...