Android - Crazy Stone Deep Learning

For discussing go computing, software announcements, etc.
Rémi
Lives with ko
Posts: 170
Joined: Sat Jan 14, 2012 4:11 pm
Rank: KGS 4 kyu
GD Posts: 0
Has thanked: 32 times
Been thanked: 119 times
Contact:

Re: Android - Crazy Stone Deep Learning

Post by Rémi »

Satorian wrote:It's been updated today. Couldn't find a change log though.

I'd be interested to know whether the strength ratings differ from the desktop version, especially if they refer to the strength of data set used for that particular playing strength, or whether they are ratings the engine actually achieved on KGS at that setting.


The change is a message that indicates how to open external sgf files.

The weak levels are identical to the desktop version, and are calibrated by playing on KGS. The strongest levels are time-controlled, and strength may vary depending on hardware.
Javerberg
Beginner
Posts: 13
Joined: Mon Jul 28, 2014 3:35 am
GD Posts: 0
Has thanked: 1 time

Re: Android - Crazy Stone Deep Learning

Post by Javerberg »

I just got the ios version, and am really impressed by the weak Levels. They feel much more human than in the old version. A small minus is that you can't import a sgf without using a pc.
Satorian
Dies with sente
Posts: 93
Joined: Sun Jan 22, 2012 7:11 am
Rank: Total beginner
GD Posts: 0
Universal go server handle: Satorian
Has thanked: 26 times
Been thanked: 13 times

Re: Android - Crazy Stone Deep Learning

Post by Satorian »

Rémi wrote:The change is a message that indicates how to open external sgf files.

The weak levels are identical to the desktop version, and are calibrated by playing on KGS. The strongest levels are time-controlled, and strength may vary depending on hardware.


Thank you for the info! Any chance for an offset cursor input method?

Other than desiring the added input method, I'm loving the new Crazy Stone. The higher res graphics, landscape layout for the tablet, the playing style, strength and the analysis mode are all awesome. Money well spent. Definite recommendation from me.
lobotommy
Lives in gote
Posts: 408
Joined: Thu Jul 29, 2010 2:01 am
Rank: EGF 3kyu
GD Posts: 0
Universal go server handle: tommyray (1d/2d)
Location: Poland, Gliwice
Has thanked: 127 times
Been thanked: 94 times

Re: Android - Crazy Stone Deep Learning

Post by lobotommy »

I played 4 games so far against 2d with slow time setting (hardware is iPhone5s). And scored two loses and two wins.
After first two loses I was impressed. Then I changed my approach and played two games very patiently and managed to win twice by killing big groups. Too easily I think. Theoretically I have 3k in EGF rating but it means nothing without playing tournament games, which I don't do recently.

What are your thoughts on strength of this app? Is it stronger on 5d setting than 2/3 kgs dan? EGF dan maybe?
Tsumego/Tesuji apps for iPad, iPhone & Android devices:http://www.lifein19x19.com/forum/viewto ... =18&t=7511
Mike Novack
Lives in sente
Posts: 1045
Joined: Mon Aug 09, 2010 9:36 am
GD Posts: 0
Been thanked: 182 times

Re: Android - Crazy Stone Deep Learning

Post by Mike Novack »

I am not sure how seriously you can take the strength settings when these programs are being run on hardware as weak as the CPU in a phone.

Time and machine power are trade offs. There MIGHT be a program that could play at 5d level when running on a phone, but it would be needing to take a LOT of time per move.

Something to note. For the CPU's of highly portable devices, improvements aren't just in terms of crunch power but crunch power relative to electric power needed.
jorden
Dies in gote
Posts: 27
Joined: Mon Sep 05, 2016 7:21 am
Rank: DDK
GD Posts: 0
Has thanked: 6 times
Been thanked: 3 times

Re: Android - Crazy Stone Deep Learning

Post by jorden »

https://www.gokgs.com/graphPage.jsp?user=crazyphone

@Mike Novack: does this count? Crazystone on a phone on KGS.
lobotommy
Lives in gote
Posts: 408
Joined: Thu Jul 29, 2010 2:01 am
Rank: EGF 3kyu
GD Posts: 0
Universal go server handle: tommyray (1d/2d)
Location: Poland, Gliwice
Has thanked: 127 times
Been thanked: 94 times

Re: Android - Crazy Stone Deep Learning

Post by lobotommy »

jorden wrote:https://www.gokgs.com/graphPage.jsp?user=crazyphone

@Mike Novack: does this count? Crazystone on a phone on KGS.


Nice, I will check higher settings in my next play :]

[Update after few more games]
It's really strong. 30 second byo for it and it's crashing me every time by two digits score now.
Tsumego/Tesuji apps for iPad, iPhone & Android devices:http://www.lifein19x19.com/forum/viewto ... =18&t=7511
WinPooh
Dies in gote
Posts: 30
Joined: Fri Dec 30, 2011 6:22 am
Rank: KGS 2 kyu
GD Posts: 0
KGS: WinPooh
IGS: WinPooh
DGS: WinPooh
Been thanked: 3 times

Re: Android - Crazy Stone Deep Learning

Post by WinPooh »

Is it possible in Android version to save automated game analysis results for further review?
Ophitoxaemia
Dies in gote
Posts: 25
Joined: Sat May 07, 2016 4:39 pm
GD Posts: 0
Has thanked: 4 times
Been thanked: 4 times

Re: Android - Crazy Stone Deep Learning

Post by Ophitoxaemia »

I just played CSDL on iOS at 1D setting and it made a huge dragon which died. It didn't try to make shape even when it had a chance to.
lobotommy
Lives in gote
Posts: 408
Joined: Thu Jul 29, 2010 2:01 am
Rank: EGF 3kyu
GD Posts: 0
Universal go server handle: tommyray (1d/2d)
Location: Poland, Gliwice
Has thanked: 127 times
Been thanked: 94 times

Re: Android - Crazy Stone Deep Learning

Post by lobotommy »

Ophitoxaemia wrote:I just played CSDL on iOS at 1D setting and it made a huge dragon which died. It didn't try to make shape even when it had a chance to.


Don't use preset strenght settings - they have no relation to anything, I don't know what 1d is in CrazyStone. Japanese 1d (weak, close to maybe Polish 3kyu)? KGS (even weaker, like 5 kyu)? Or maybe it's just for making certain people feel good that they can play on dan level. Use time settings instead. Set for example a 20 second for computer player as a byoyomi time and then try to win. I guarantee you will lose badly.
Tsumego/Tesuji apps for iPad, iPhone & Android devices:http://www.lifein19x19.com/forum/viewto ... =18&t=7511
Rémi
Lives with ko
Posts: 170
Joined: Sat Jan 14, 2012 4:11 pm
Rank: KGS 4 kyu
GD Posts: 0
Has thanked: 32 times
Been thanked: 119 times
Contact:

Re: Android - Crazy Stone Deep Learning

Post by Rémi »

lobotommy wrote:
Ophitoxaemia wrote:I just played CSDL on iOS at 1D setting and it made a huge dragon which died. It didn't try to make shape even when it had a chance to.


Don't use preset strenght settings - they have no relation to anything, I don't know what 1d is in CrazyStone. Japanese 1d (weak, close to maybe Polish 3kyu)? KGS (even weaker, like 5 kyu)? Or maybe it's just for making certain people feel good that they can play on dan level. Use time settings instead. Set for example a 20 second for computer player as a byoyomi time and then try to win. I guarantee you will lose badly.
The strength levels of Crazy Stone were calibrated by playing on KGS. The 1d level is pure neural network, and it has a tendency to believe that any very large group is alive. IIRC, 2d starts to use playouts, which help to cure this weakness.
lobotommy
Lives in gote
Posts: 408
Joined: Thu Jul 29, 2010 2:01 am
Rank: EGF 3kyu
GD Posts: 0
Universal go server handle: tommyray (1d/2d)
Location: Poland, Gliwice
Has thanked: 127 times
Been thanked: 94 times

Re: Android - Crazy Stone Deep Learning

Post by lobotommy »

Thanks Rémi, mystery solved.
Tsumego/Tesuji apps for iPad, iPhone & Android devices:http://www.lifein19x19.com/forum/viewto ... =18&t=7511
Ophitoxaemia
Dies in gote
Posts: 25
Joined: Sat May 07, 2016 4:39 pm
GD Posts: 0
Has thanked: 4 times
Been thanked: 4 times

Re: Android - Crazy Stone Deep Learning

Post by Ophitoxaemia »

Great explanation, thanks.

Played again on 2d setting. It reacted differently and didn't make a dragon.
And
Gosei
Posts: 1464
Joined: Tue Sep 25, 2018 10:28 am
GD Posts: 0
Has thanked: 212 times
Been thanked: 215 times

Re: Android - Crazy Stone Deep Learning

Post by And »

I use the latest version of CrazyStone DeepLearning Pro 1.2.0 purchased a year and a half ago, the regular version has been updated many times during this time. has the neural network changed since then? I'm wondering if there is a difference in the 1-2 dan game?
Post Reply