On 4-4-2009 8:49, Kevin Wampler wrote:
> Here's a fun, frivolous, and simple little program I wrote recently 
> which seemed just interesting enough to share.  The idea is to write an 
> AI assistant for the game of minesweeper.  The trickiness is that 
> sometimes every available square has a chance of having a mine on it, so 
> running a basic constraint satisfaction algorithm on it won't give you 
> any useful information.
> 
> Instead, you can generate a sequence of random mine assignments which 
> are each consistent with the numbers on the currently visible squares 
> and then calculate a running average of the probability that each square 
> has a mine.  In some cases this probability is 100% or 0%, in which case 
> it (with high likelihood) gives you the same result as a standard 
> logical deduction would, but in other cases the probabilities are very 
> useful for making a move.

The lower one has at least 5 positions where you will end up having a 
50% change and no additional information. The change of surviving this 
minefield are therefore below 3%. Possibly much below that if you have 
had to guess before arriving at the current position. That is not good 
for a game. Perhaps you can employ something like this to find grids 
that can be solved with a reasonable chance.

Post a reply to this message