Reinforcement-taxi-v3

Q-learning, SARSA, Value-iteration for Taxi-v3 problem of python Gym library.

Taxi-v3

The taxi-v3 problem is a classic reinforcement learning problem in the Python library Gym. The goal is to learn an agent how to navigate a grid-world environment as a taxi driver, picking up passengers and dropping them off at their desired locations. The environment is represented by a 5×5 grid, with walls blocking certain paths and passengers and destinations represented by colored squares. 500 states. The agent receives a reward for successfully picking up and dropping off passengers. However, the agent also incurs a penalty for each time step is taken and for illegal actions, such as attempting to pick up a passenger who is already in the taxi. Actions: move south, move north, move east, move west, pick up a passenger, drop off a passenger.
More info here: [Gym Documentation]: https://www.gymlibrary.dev

Project files:

• 'agents.py' contains algorithm classes:

  • Qlearning
  • Sarsa
  • Value-iteration

• 'main.py' contains:

  • running the train and tests methods for each algorithm

• 'requirements.txt' contains:

  • Libraries Installation

Overview

For each algorithm, you can run a training and test methods.
agent1 = QlearningAgent() choose the algorithm.
agent1.train() run a training method.
After that, policy file will be generated and saved to main path.
Similarly, you can run test method agent1.test()

If you want to change hyperparameters such as: learning rate, discount factor, epochs, use:
QlearningAgent(0.9, 0.6, 2000)