Alberich wrote:I played Crazystone last night using level one. I beat it easily. But then I tried level two. OMG. Level two must be unbeatable because it literally took my breath away. No matter what I did in level two I was being killed mercilessly. Now the strange thing is...when I skipped to level seven I was able to breathe and play more to my style. Can anybody explain this? And if anyone who can beat level two consistently is clearly a Dan level player regardless of actual rating. That's my opinion.
iPad 1, beat level 2 giving 5 stones with relative ease. A friend recommended me playing bot games giving high handicaps to improve my reading and attacking, so I'm doing that (an iPad works well for it)
Geek of all trades, master of none: the motto for my blog mostlymaths.net
Alberich wrote:I played Crazystone last night using level one. I beat it easily. But then I tried level two. OMG. Level two must be unbeatable because it literally took my breath away. No matter what I did in level two I was being killed mercilessly. Now the strange thing is...when I skipped to level seven I was able to breathe and play more to my style. Can anybody explain this? And if anyone who can beat level two consistently is clearly a Dan level player regardless of actual rating. That's my opinion.
iPad 1, beat level 2 giving 5 stones with relative ease. A friend recommended me playing bot games giving high handicaps to improve my reading and attacking, so I'm doing that (an iPad works well for it)
Crazystone on the iPad (I have the new Ipad 3) is the only go playing software that i'm aware of that allows multiple undo. There are probably others that I'm not aware of at the moment because i haven't had much experience with many of the other popular brands out there. I just tried it out in Crazystone and it works fantastic. The only thing I wish Crazystone HD had was variation trees so I can easily go to any variation and try new lines. I also wish I could see the screen in landscape. It only allows portrait mode. I hope there's an update in the future that allows this.
In the meantime I email myself the games so I can import them into another Go program I have on the iPad and then create variation trees manually.
Last edited by Alberich on Thu Jul 26, 2012 2:58 pm, edited 1 time in total.
Mike Novack wrote:First of all, I suspect that when running on a machine as weak as an ipad Crazy Stone might not be even 1 dan at its highest level. Rémi could probably tell us.
In 9x9 I've seen Crazy Stone beat an EGF 3d a couple of times. In 19x19 I've seen an EGF 2k crush it. (Highest Level setting for both obviously) Small samples obviously but from it's what I've seen from letting friends play it on my iPad 3.
That said, I've seen another beginner struggle a lot with level 2 even after beating level 1 consistently, it seems to be a substantial increase in strength at beginner level.
Boidhre wrote: In 9x9 I've seen Crazy Stone beat an EGF 3d a couple of times. In 19x19 I've seen an EGF 2k crush it. (Highest Level setting for both obviously) Small samples obviously but from it's what I've seen from letting friends play it on my iPad 3.
Sorry if I wasn't clear enough. When I said perhaps not even 1 dan on such (skimpy) hardware I meant 19x19. The MCTS progams are very strong at 9x9 because playouts take less time (games much shorter) and the sample size is a much larger percentage, though still a small percentage, of all possible moves.
Boidhre wrote: In 9x9 I've seen Crazy Stone beat an EGF 3d a couple of times. In 19x19 I've seen an EGF 2k crush it. (Highest Level setting for both obviously) Small samples obviously but from it's what I've seen from letting friends play it on my iPad 3.
Sorry if I wasn't clear enough. When I said perhaps not even 1 dan on such (skimpy) hardware I meant 19x19. The MCTS progams are very strong at 9x9 because playouts take less time (games much shorter) and the sample size is a much larger percentage, though still a small percentage, of all possible moves.
Yeah I figured you meant 19x19 which is why I mentioned the EGF 2k beating it easily.
Boidhre wrote:Yeah I figured you meant 19x19 which is why I mentioned the EGF 2k beating it easily.
When running on an ipad. We have to be careful when talking about how strong these programs are to always specify "when running on what". I don't have Crazy Stones here but if I did and ran it on a moderately powerful laptop would expect perhaps 2 dan (KGS).
Boidhre wrote:Yeah I figured you meant 19x19 which is why I mentioned the EGF 2k beating it easily.
When running on an ipad. We have to be careful when talking about how strong these programs are to always specify "when running on what". I don't have Crazy Stones here but if I did and ran it on a moderately powerful laptop would expect perhaps 2 dan (KGS).
Oh I'm very well aware that we're talking about just an iPad here and it's limited memory especially. I'm deeply curious what the program would be like on this machine, i5 quad core, 8GB Memory etc. Though it'd be nicer with 16GB memory from my understanding with a 3GHz quad core processor? i.e. that the memory would be the bottleneck with MCTS programs above a certain processor speed.
WinPooh wrote:Hi Rémi, thank you for the engine! I play with Adroid version on Sony Ericsson Arc, and it is a tough challenge on Level 10! (I'm 2k KGS). However, once I won at 4 stones as White: http://gokifu.com/s/10up One question: is it possible to export SGF with played game to file? The program only announces "Game saved to file", but I can't find any SGF in the filesystem...
Hi,
Congrats for your win. The level of Crazy Stone may look a bit inconsistent sometimes, because there are some things that are obvious to humans but it does not understand.
If you wish to get the sgf of a game you saved, the app offers the option to email it to you. (Menu/Game Record/Export game via e-mail). So if the game is already saved, then first load it, then email it. The file is saved on disk, but it is a part of the disk that other apps cannot access.
Satorian wrote:To veer slightly off-topic: Does anyone know what strength the Android version achieves on a Tegra 3? Also, does the Android version feature native tablet support? I'm tempted to buy Crazy Stone, but I'm not sure about these two points.
I am running Crazy Stone on my Asus Transformer Prime tablet. It works, and can even use the 4 cores. But the graphics are rather low resolution, and it runs only in portrait mode, which is a bit annoying.
Boidhre wrote:Oh I'm very well aware that we're talking about just an iPad here and it's limited memory especially. I'm deeply curious what the program would be like on this machine, i5 quad core, 8GB Memory etc. Though it'd be nicer with 16GB memory from my understanding with a 3GHz quad core processor? i.e. that the memory would be the bottleneck with MCTS programs above a certain processor speed.
I spent a lot of time optimizing Crazy Stone's memory usage for cell phones, so it is very efficient in this area. The cell phone version uses 20-25 Mb of RAM at most. The PC version probably won't ever use more than 1Gb, even after a long analysis.
Satorian wrote:To veer slightly off-topic: Does anyone know what strength the Android version achieves on a Tegra 3? Also, does the Android version feature native tablet support? I'm tempted to buy Crazy Stone, but I'm not sure about these two points.
I am running Crazy Stone on my Asus Transformer Prime tablet. It works, and can even use the 4 cores. But the graphics are rather low resolution, and it runs only in portrait mode, which is a bit annoying.
Rémi
Thanks for the reply!
Any chance of an Android HD version coming?
And what would you estimate the playing strength to be on the Transformer Prime?
Satorian wrote:To veer slightly off-topic: Does anyone know what strength the Android version achieves on a Tegra 3? Also, does the Android version feature native tablet support? I'm tempted to buy Crazy Stone, but I'm not sure about these two points.
I am running Crazy Stone on my Asus Transformer Prime tablet. It works, and can even use the 4 cores. But the graphics are rather low resolution, and it runs only in portrait mode, which is a bit annoying.
Rémi
Thanks for the reply!
Any chance of an Android HD version coming?
And what would you estimate the playing strength to be on the Transformer Prime?
Unbalance is doing the UI. I don't know much about their plans and don't want to speak for them.
Regarding playing strength, I have no precise data to give a serious estimate. At some point I tried to get rated accounts on KGS to test, but I did not manage to get my accounts rated. I may try again later.
In terms of speed, Crazy stone does about 250 playouts / second / core on the transformer prime, so 1000 playouts / second on 4 cores. This compares to about 2000 playouts / second /core on a PC. Except for speed, the program is otherwise identical.
Rémi wrote:Unbalance is doing the UI. I don't know much about their plans and don't want to speak for them.
Regarding playing strength, I have no precise data to give a serious estimate. At some point I tried to get rated accounts on KGS to test, but I did not manage to get my accounts rated. I may try again later.
In terms of speed, Crazy stone does about 250 playouts / second / core on the transformer prime, so 1000 playouts / second on 4 cores. This compares to about 2000 playouts / second /core on a PC. Except for speed, the program is otherwise identical.
Rémi
Interesting! Thank you very much!
One last question, going a bit beyond Crazy Stone as it is: Do you think there'd be a worthwhile way to use distributed computing similar to SETI@home for Go AI? Of course I'd assume that to be way less efficient than a tightly integrated cluster, but perhaps the sheer scale could offset that. I'm sure there'd be a few thousand Go and AI enthusiasts around to offer their processing power.
