Minesweeper rules code#

The environment's current state, action, reward and new state are collectively called a transition. Each action is assigned a reward based on our reward structure. Its state is updated every time the agent takes an action. The Agent: This is essentially the computer, which takes actions on the environment based on what it thinks will result the highest reward / lowest penalty.The Reward Structure: Rather than explicit rules, we indicate to the computer what is beneficial or detrimental to performing a task by assigning rewards and/or penalties on specific conditions.The following are the key components of RL: Reinforcement Learning (RL) is an area of machine learning that aims to train a computer to accomplish a task. This is where Reinforcement Learning comes in! What is Reinforcement Learning? But can a computer learn to play Minesweeper without these clear-cut commands? In other words, can a computer learn the logic of Minesweeper without it being aware of the rules firsthand? This means that a computer solver can be obtained by explicitly programming it to take specific actions from specific states. Given the logical rules of the game, Minesweeper can actually be solved using brute force algorithms such as a combination of if-else statements. You progress through the game by revealing numbers and deducing where it is safe to click next.Ĭan a computer learn to beat Minesweeper? By clicking on tiles you reveal numbers which indicate how many mines are in the tiles around them. The objective - click on all tiles except the ones containing mines. Minesweeper is a classic game of logic, dating back to 1989. Using Reinforcement Learning to Beat Minesweeper.Sentdex's Deep-Q Learning Tutorial Series.

Minesweeper rules code#

Some code in this repo were adapted from: Below is a comparison of a model playing Minesweeper before training and after training on ~half a million games! After all, Reinforcement Learning is all about having a machine learning model improve through trial-and-error. Of course, since this is a Reinforcement Learning project, the above statement should be a given.