Train a Smartcab How to Drive

Udacity Machine Learning - Reinforcement Learning Project The goal of this project was to implement a Q-Learning algorithm so that a smartcab agent can learn to navigate to a destination within a predefined time in a grid-world.

The methodology for the implementation of this code is found in the report found in the main directory of this repository.

Install

This project requires Python 2.7 with the pygame library installed:

https://www.pygame.org/wiki/GettingStarted

Code

Open smartcab/agent.py and implement LearningAgent. Follow TODOs for further instructions.

Run

Make sure you are in the top-level project directory smartcab/ (that contains this README). Then run:

python smartcab/agent.py

OR:

python -m smartcab.agent

Automated Data Collection

To run the automated data collection uncomment the script according to single scenario or iterative.

For iterative collection

• Comment out the Single Scenario Function call
• set the number of times for each iteration
• choose ranges and steps for Gamma, Epsilon and Epsilon Decay

    gamma_values = [(x / 100.0) for x in  range(30, 91, 10)]
    epsilon_values = [(x / 100.0) for x in range(50, 55, 10)]
    epsilon_decay_values = [(x / 100.0) for x in range(90, 100, 3)]

    number_of_times = 5
        for i in range(0, number_of_times):
        iterative_data_collection(gamma_values, epsilon_values, epsilon_decay_values, str(i))

• Run python automated_data_collection.py

For Single Scenario

• Comment out the Iterative Data Collection Function Call
• Set the variables for Gamma, Epsilon, Epsilon Decay and the number of times

    single_scenario_repeat_data_collection(0.50, 0.50, 0.99, 10)

• Run python automated_data_collection.py