Deep Learning Essentials

Simple reinforcement learning example

In this section, we will implement a simple reinforcement learning example for a game of DuskDrive:

1. We will use gym game environment for this purpose and universe Docker environment for running it:

import gymimport universeenv = gym.make('flashgames.DuskDrive-v0')env.configure(remotes=1)

2. Once the environment is configured to run it, we reset it to a random start state as follows:

observations = env.reset()

Once the game is initialized to a random start state, we need to continuously provide an action. Since DuskDrive is a racing game, one simple strategy is to simply move forward using the up arrow key.

3. We run this strategy in an infinite loop, as follows:

while True: action = [[('KeyEvent', 'ArrowUp', ...