It would have been really helpful if you displayed those changing bit patterns in a monospace font.
@JoBot__
2 жыл бұрын
I agree.
@thehobbyshow5049
3 жыл бұрын
Oh, I thought you could solve this with a neat-python implementation :/ Doesn't help me with my problem but nice video nonetheless :D
@tylermett4333
2 жыл бұрын
I am so glad I found this channel!
@neatai6702
2 жыл бұрын
Welcome aboard!
@JamesCetrangelo
2 жыл бұрын
A strange game. The only winning move is not to play.
@LMitchellAnimation
2 жыл бұрын
Found you from a recommend and am enthralled. Hopefully this bump means big things, this stuff is entertaining and detailed!
@neatai6702
2 жыл бұрын
Welcome aboard!
@Kraus-
2 жыл бұрын
Lol that trash talk in the beginning.
@smileyp4535
2 жыл бұрын
What is the optimal draw? What does it do against itself?
@catkook543
2 жыл бұрын
6:36 it missed a winning move!
@thegozer100
2 жыл бұрын
It would be interesting to see this algorithm applied to connect 4 with a neural net
@neatai6702
2 жыл бұрын
Good idea.. I'll add it to the list
@oberguga
2 жыл бұрын
Tic-tac-toe 3x3 can be simply and efficiently implementet as full List of possible situation with 4 rotation and two mirroring... It no need an AI. On infinite board with 5 in row it actually became a task for AI
@oberguga
2 жыл бұрын
@@Wariowa345 Infinite board means that new X or O can be placed everywhere, but it doesnt mean that it should be or had any significance. Algoritm for infinity board analyse only relevant edge of marked area(4 symbols deep) and consider new step in a similar area around. So if player place X or O in a middle of nowhere AI look around in square 9x9 and should decide that it's irreleant and do something in area where more interesting happening. For not algorithmic AI it means that it should scan area from edge of marked area to 4-5 cells from it with aperture 9x9 and calculate importance of every cell, then place X or O in most important place.
@oberguga
2 жыл бұрын
@@Wariowa345 it need analyse only one neighborhood 9x9 every step (near place where player make his) . Other area already mapped and analysed. So real discovery area less then 9⁴=6.5k cells. It is manageble for AI. Choosing one best cell if it already evaluated is trivial and can be done by simple for loop over analysed cells. Structure of net can mirror the fact that play field has simmetries and real field of possible situations 8 times smaller than it looks. Also probably parts of 9x9 window irrelevant for evaluation and all that should be analysed is just intersection of X and + patterns (approx 33 cells for input of an AI part), on same area(same AI analyser applied to every of 33 cells in neighborhood of last player step). It sounds quite manageble.))
@derp2375
Жыл бұрын
Can you apply this in a game like checkers?
@neatai6702
Жыл бұрын
its on my (long) list of projects... but, yes, I don't see why not..
@JoBot__
2 жыл бұрын
Wow, this vid has less views than I thought it'd have. Nice job on it anyways.
@neatai6702
2 жыл бұрын
thanks for the feedback !
@catkook543
2 жыл бұрын
6:23 ok, that x placement is sub-optimal. if you place your first move in a corner, and your opponent places theirs in the middle you want to play directly across from them
@mysteriouslyseeing
2 жыл бұрын
At this point, the AI understands that every game, when played optimally, will end in a draw, and it doesn't differentiate between different draws.
@catkook543
2 жыл бұрын
@@mysteriouslyseeing ok, but if your opponent plays in the middle in response to you playing in the corner, then playing the corner opposite from them is the only move that can force a win depending on how your opponent plays next
@mysteriouslyseeing
2 жыл бұрын
@@catkook543 That move is definitely optimal for humans, but is as optimal as all the other moves that result in a draw to the computer.
@catkook543
2 жыл бұрын
@@mysteriouslyseeing well, there's only 4 possible moves their opponent can make to make it not a guaranteed win, out of 6 so assuming your opponent picks randomly that's a 33% chance you get a guaranteed win. though i don't think the alternative move options boost those odds by very much.
@mysteriouslyseeing
2 жыл бұрын
@@catkook543 That's the thing though, the AI doesn't assume that it's opponent will play randomly. It assumes that it's opponent will play optimally.
Пікірлер: 31