It's so hard to beat Crazy Stone

[Mike Novack]

Move 10: it is in the joseki book that black can not hanne. CS didnt know joseki.
Move 14: i know it is written that you capture the cut stone but not on this board.
Move 28: really? i guess CS is satisfied with what appears to be influence...
Move 50: this is a perfect example of computer playing influence without any plan.
Move 54: CS is trying to capture cut stone but can not read what he will giveup catch one stone.
Move 64: kyuish attack.
Move 70, 72: appears to be an attack that is not really an attack...
Move 78: When board gets complicated.. CS plays Crazy moves.
Move 100 and 101 trade: helped white and lost sente
Move 110, 112: whY????
Move 120, 124, 146, 150, 152, 154............: crazy computer moves

I understand that computer is trying hard to make the game even after losing big in the beginning.
but if you play a move...it should have meaning. It least ddk moves are full of purpose although it is wrong.
This computer lacks spirit and reading ability.

I will be honest and say that i can beat CS in 4 stone game.

Magicwand, you are mistaking the concept (of how a MCTS evaluator works). It isn't reading. It isn't planning. It is simply making the move that analysis of a large number of playouts between equal opponents indicated had the highest chance of winning. That means "insanely complicated, dependent on a large amount of scattered small bits of aji" is viable even though unintelligible to human thinking. So it can make rather weird moves, leave lots of positions unfinished, etc. In human terms, think of this as a player with extremely "loose" style. Its strength is unbalanced, more tactical than strategic ability.

However, you are convinced you could beat the program even if it had several handicap stones. Well that is an empirical matter, easily settled one way or the other. Play against it. OK, you don't have access to playing against it in a slow game (you lack the hardware and the software) but you can always play against it in a fast game. To compensate, allow it one less stone.

In other words, lets see you take on the Crazy Stone bot on KGS allowing it three stones. It should be back up there shortly since the tournament is over. If not, you could contact Remi to see what could be set up.

[Mike Novack]

I'm not seeing a lot of the strongest MCTS bots available at the moment. But since you are willing to let the bot have a few stones handicap, that shouldn't matter. For example, the bot ManyFaces is active at the moment, a weaker program and running on weaker hardware than CrazyStone.

So Magicwand, let us see what happens if you play some games against ManyFaces at four stones. If you can't beat ManyFaces at four stones it's sure you wouldn't be able to take CrazyStone at four.

NOTE: ManyFaces will make fewer "weird" moves than some of the others, but this isn't because its MCTS evaluator is so different but rather because it is applying a prefilter to what moves it is sending to the MCTS evaluator.

[Pippen]

I'd love a match up in an even game between Magicwand and CS, with the time setting FJ had in his match against the program (something like 60min, 5/30). The necessary steps would be:

1. Contact Remi Coulom, https://plus.google.com/u/0/+R%C3%A9miC ... razy/posts, see if he's interested
2. Set up a date where the "real deal" CS will be ready on KGS
3. Set up a time setting, e.g. 60min., 5/30s

Go!

I'd bet my house on CS, but I'd root for Magciwand^^.

[Magicwand]

I'd love a match up in an even game between Magicwand and CS, with the time setting FJ had in his match against the program (something like 60min, 5/30). The necessary steps would be:

1. Contact Remi Coulom, https://plus.google.com/u/0/+R%C3%A9miC ... razy/posts, see if he's interested
2. Set up a date where the "real deal" CS will be ready on KGS
3. Set up a time setting, e.g. 60min., 5/30s

Go!

I'd bet my house on CS, but I'd root for Magciwand^^.

Not unless you have multiple house...
dont put yourself on homeless situation over me .
I have tried some evengame against strongest bot i can find and was really disappointed on their strength.
i wont bet my house over it but my pride worth more than my house..

[Magicwand]

question:
can Crazy Stone play turnbased game against me???

[mitsun]

Someone with a copy of CrazyStone could set up a Malkovich game and relay the moves. We could also get (hidden) commentary from CrazyStone giving its estimated winning chances along the way, and perhaps also a listing of some of the other moves it considered playing at interesting junctures.

[Mike Novack]

Someone with a copy of CrazyStone could set up a Malkovich game and relay the moves.

Somebody with a copy and some serious hardware.

For example, MFOG12 playing on my machine at home is about 1 dan, but the bot on KGS about 3 dan on Fotland's machine. And that machine is much less powerful than what CrazyStone was running on in this tournament.

Now if somebody has a home machine with at least an i7-4970 in it, then maybe just a stone or two weaker than at the tournament. Maybe a serious gamer among us with a machine powerful enough?

In the meantime, Magicwand, since you are so confident taking CrazyStone at four stones, why not give us an example game or two against the much weaker ManyFaces bot at four stones.

[Boidhre]

Someone with a copy of CrazyStone could set up a Malkovich game and relay the moves.

Somebody with a copy and some serious hardware.

For example, MFOG12 playing on my machine at home is about 1 dan, but the bot on KGS about 3 dan on Fotland's machine. And that machine is much less powerful than what CrazyStone was running on in this tournament.

Now if somebody has a home machine with at least an i7-4970 in it, then maybe just a stone or two weaker than at the tournament. Maybe a serious gamer among us with a machine powerful enough?

In the meantime, Magicwand, since you are so confident taking CrazyStone at four stones, why not give us an example game or two against the much weaker ManyFaces bot at four stones.

I have the 2012 version of Crazy Stone and an i5-4690. Weaker than what you want but still reasonably strong.

[Pippen]

Someone with a copy of CrazyStone

Remember: The commercial version of CS is a different animal. It is about 1d only. Magicwand would have to play against the premium version using extraordinary computer power like in the best of 5 game between FJ(6d) and CS which FJ won 3-1 (and games were all pretty close except for the last one when CS had hardware problems). IMO the matchup should be: Magicwand against the CS that lost to Chikun...even game, maybe someone could ask Remi if he could do it the Malkovich way.... (A HC game would be boring because CS will be too strong for Magicwand, even if he'd bet his house against it now , assuming he's 4d KGS like his profile says)

[Uberdude]

I find it a little perverse if arrogance is what allows someone who would not normally have the opportunity to play a strong version of CrazyStone to do so.

[wineandgolover]

I find it a little perverse if arrogance is what allows someone who would not normally have the opportunity to play a strong version of CrazyStone to do so.

Perverse, but fun.

[Mike Novack]

Magicwand would have to play against the premium version using extraordinary computer power like in the best of 5 game between FJ(6d) and CS which FJ won 3-1 (and games were all pretty close except for the last one when CS had hardware problems).

No, not "extraordinary power" as computers go, just something at the top end of the "workstation" class. Perhaps an order of magnitude more than most folks have at home but only 2-3 times what a serious "gamer" might have.

Nor am I so sure that the commercial program is an entirely different program. Might just be a matter of parameters. This is the sort of thing we would need to ask Remi. Is it just a matter of parameters and if so, how to override the defaults (that the commercial program is using) and what would be the values to use if running on a .........(those optimizing for a given amount of crunch power)

At least I know that were I designing/coding I'd have the variables that controlled performance and that might have to be experimentally adjusted supplied as run time parameters instead of having to recompile the program each time to change them.