Pong-learning-agent

The learning agent actually did not use a markov dicision process because the initial condition is constant. At a resonable alpha gamma and epsilon combination, the learning agent is able to bounce the ball 15 times.