What Reinforcement Learning is

In Reinforcement Learning (RL), a software agent makes observations and takes actions inside an environment, and in return it receives rewards. The agent’s whole job is to learn a way of acting — a policy — that maximizes its expected reward over time. Nobody hands it labeled examples; it has to figure out what works through trial and error.

That loop shows up almost everywhere:

• a robot getting closer to (or further from) a target destination
• a game-playing program scoring points
• a thermostat balancing comfort against energy use
• a trading bot buying and selling based on market moves

Phase 7 flow

  flowchart TD
  A["Agent ↔ Environment Loop"] --> B["Policy Search & OpenAI Gym"]
  B --> C["Markov Decision Processes"]
  C --> D["Q-Learning"]
  D --> E["Deep Q-Networks (DQN)"]
  E --> F["Policy Gradients (REINFORCE)"]

Suggested practice environment

The classic first environment is CartPole from OpenAI Gym: balance a pole on a moving cart by pushing it left or right. It is small enough to reason about by hand, yet rich enough to show off both value-based methods (Q-Learning, DQN) and policy-based methods (policy gradients).