There's no point in shuffling out first move bomb hits. If you hit one you lose and move on, in the end it should average out. thats the point of statistical sampling :)
Printable View
There's no point in shuffling out first move bomb hits. If you hit one you lose and move on, in the end it should average out. thats the point of statistical sampling :)
No. You see, the puzzle maker must follow the same rules that the player has to follow. If the maker can shuffle a piece and move it to the far right corner by switching it with the piece currently in that corner, then the player cannot solve the puzzle because the player cannot use the maker's rules. Which makes it mathematically predictable, even if it is only a small percentage that anyone will find a mathematic problem to determine the solution.Quote:
Originally Posted by laserlight
Ah, I see. I had in mind the version where Llyod offered a prize for the solution.Quote:
Only if the program uses a completely random method to place the tiles. If you design the scramble method to shuffle the tiles as you would on a real-life tactile board, this shouldn't be a problem.
I probably should have placed bets that you were going to win... happened to look through your homepage shortly after you replied to my posts :DQuote:
Small confession to make: My current Honours project involves writing an AI to play Minesweeper Flags, the MSN version of the game. If you've never played it, it's almost like winmine.exe, except the object is to open mines, instead of avoiding them.
Well when you start it's the same board and it only changes if you pick a bomb, otherwise it's the same board and even if you do pick a bomb, only the 1 bomb moves, so it's still 99.9% the same board.Quote:
Originally Posted by Decrypt
I agree, and originally that was the plan, However, since that's the rules the windows minesweeper uses, I modified mine to do the same to be consistent with it.Quote:
There's no point in shuffling out first move bomb hits. If you hit one you lose and move on, in the end it should average out. that's the point of statistical sampling
by the way, is the cell 0,0 going to be in the middle, top left, bottom right, top right, or bottom right?
/by the way, i'm a new member so you'll hear more and more from me. so consider this my hello/
0,0 is top left and welcome aboardQuote:
Originally Posted by c89c
Haha!
I implemented my last rule and my score jumped up to 327495 with 103 wins out of 1000 games. I'm not sure how I got a score as close to Dante's with less than half as many wins, but no matter. A few more tweaks and I will be on top again.
H4X!!!!! :eek:
Darn... I'm getting < 40000! Don't think I'll submit this time. Good experience though.
And the program mysteriously crashes while displaying the scores, but only in loops; when I show scores manually it displays correctly :confused:
Hey! You all are using preboxed algorithms! No fair! ;)
If you downloaded the program the 1st day, you might have the 1st version where at the end when it prints the scores the players[i]->getName() was out of scope.Quote:
Originally Posted by jafet
40000? Are you sure that's your score and not the simplebots score?
I got up to 325000-340000 completely with my own algorithms based on how I play minesweeper in real life, so that's not exactly true.Quote:
Originally Posted by jafet
But then, I used something mentioned in Darryl's link to make my "random" guessing better, and now my latest run took 8 minutes and gave this score:Quote:
Originally Posted by Dante Shamest
(emphasis mine) :DCode:RandomBot: -611
SimpleBot: 35780
jlou: 532890
I think small changes in the algorithm can make big differences because of the bonus you get when you solve the puzzle. In a different run I counted 382 total solved (for a score of 539908).
Just something for you all to shoot for. ;)
:eek:Quote:
Originally Posted by jlou
Wow, my morale has just suffered a big blow. Doubt I'll be able to improve my algorithm much better.
Actually I could remove the limiter I mentioned earlier, but it probably won't finish running even after the Iraq war ends. :p
Ackk... don't be discouraged. :)
Everything I did is basically discussed briefly in Darryl's minedefuser link (I haven't read the pdf link yet). There were a few implementation details that I didn't think of right away that helped a lot when I added them. And of course the idea for my last improvement above your score came from that link.
Keep thinking... Hopefully some others are trying as well. The basic AI isn't that hard at all (as indicated by the SimpleBot), so it shouldn't take too much effort to put forth an entry.
That's funny because the mine defuser didn't perform half as well as Dante's, so I guess coupled with your algo, you've super-charged it. Maybe you ought to contact it's author with the suggestionsQuote:
Originally Posted by jlou
I found a small bug in the minefield which is partly why our solve rates are so high. Basically, the "firstmove" check is allowing the first hit bomb to always be avoided even if it isn't actually in the first move. So everybody was getting one mulligan per game. Just moving the mFirstMove = false; code to the end of the uncoverSquare function changed my score to:Slightly more reasonable, but still absurdly high. :pCode:RandomBot: -4206
SimpleBot: 20370
jlou: 332945
Solved: 269
Bad (negative score): 260