Skip to main content

Smartsweeper

a bot that learns to play minesweeper


Nolan Baker
(nbaker)
Using a neural network, the agent learns rules for playing minesweeper over the course of several games.

Visualization: The purple and yellow overlays indicate the agent's guess for where mines are. Purple indicates that the agent thinks it's a mine. Yellow (which really only shows up as light brown or light green) indicates that the agent thinks that there is no mine. Opacity indicates uncertainty. The more opaque a color is, the less certain the agent is about making a move. Conversely, transparency indicates certainty; however, it's unlikely that you won't be able to see any color overlay at all since there is always uncertainty.

Controls: h - switch to human mode and play to your hearts content r - let the robot do the playing left click - clears a space (and space around if the number clicked equals the number of adjacent flags) right click - flag or unflag a square Tap 'g' (for goggles) to cycle between tiles by themselves, tiles with the color overlay, or the color overlay only.

Changes

Links

Home Page
http://code.google.com/p/smartsweeper/

Releases

Smartsweeper beta — 16 May, 2010

Pygame.org account Comments