Life In 19x19
http://www.lifein19x19.com/

Android - Crazy Stone Deep Learning
http://www.lifein19x19.com/viewtopic.php?f=18&t=13656		 Page 1 of 2
Author:  Majordomo [ Fri Oct 07, 2016 4:03 am ]
Post subject:  Android - Crazy Stone Deep Learning
Didn't see anything about this here, only on reddit - but this:

https://play.google.com/store/apps/details?id=jp.co.unbalance.android.gocsdl

Crazy Stone Deep Learning for Android (and iPhone coming I gather). I just downloaded it and played two games - beating a 6k and losing hard to a 1d. My phone is a Samsung S7 Edge but even so it got really warm when I went to analyse the whole thing (with 2s per move analysis). Anyway, despite a somewhat clunky and overly decorative GUI I like it, it plays faster than GnuGo (and a whole lot better) which has been my only phone AI so far so maybe I'll finally get some Go practice in during my daily commute. The analyse game feature is something I intend to explore more, but I'm very intrigued by it - as a way to more nail down the blunders I make than the fractional % losses. Also, I don't have Crazy Stone for desktop so I can't compare the two.

Thoughts?

One thing, if this really plays at a 5d level, using your phone to cheat, as has been a problem in chess suddenly became a whole lot easier didn't it? I wonder if tournament organisers will care and take measures to prevent it - perhaps they'll be less keen to allow digital tools to record the game while playing (which everyone used for the only tournament I've been to back in April)?
Author:  oren [ Fri Oct 07, 2016 10:22 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
You could have had the 7d version as a subscription to the old champion go for a while, so it's not a new problem. The new version just means you don't need a subscription and internet access.

So far I haven't heard of any issues in real life tournaments. It might require getting up a lot to a quiet spot to ask the computer for help unless it's a life and death problem on the board.
Author:  Majordomo [ Fri Oct 07, 2016 10:58 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
Ah, cool, never knew. Do you know ifthe Champion Go + subscription also has the same analysis features? If so, strange that this one is only 5d if that one was 7d. Hehe I also doubt that Go will have the same issues as Chess still, but it'll be interesting to see how things develop.

Another thing, for analysing the whole game, I'm curious as to which "strength" you're getting at 1s per move compared to the higher.
Author:  oren [ Fri Oct 07, 2016 11:12 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
There is no analyze on the old program. I have used the desktop version for analysis. The 7d to 5d is just the limitation of the phone vs a server most likely.
Author:  Solomon [ Fri Oct 07, 2016 1:01 pm ]
Post subject:  Re: Android - Crazy Stone Deep Learning
I played it at 5d (max) and was very disappointed by its playing level, especially near the end. For instance, my last move in the screenshot below was Q1...Crazy Stone's response speaks for itself:

Image
Author:  Majordomo [ Fri Oct 07, 2016 1:59 pm ]
Post subject:  Re: Android - Crazy Stone Deep Learning
Heh, I didn't count - but I assume you won the game?

Well that kind of thing is disappointing. I did make it play Leela (the recent version, 5d kgs I think?) and while I didn't have time to carry it to the endgame it had an marginal advantage when I ceased (like move 150).

I also read that it has problems with ladders, but maybe only for weaker levels?
Author:  gogeo [ Mon Oct 10, 2016 2:28 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
oren wrote:
There is no analyze on the old program. I have used the desktop version for analysis. The 7d to 5d is just the limitation of the phone vs a server most likely.



Hello,
small question regarding anlaysis - how to use it? I know how to get feedback on next move - but in go sequences I think are more important. So even if move is ok locally - maybe after sequence is very bad. Any idea how to analyze for more movies?

Cheers
gogeo
Author:  dfan [ Mon Oct 10, 2016 4:58 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
gogeo wrote:
oren wrote:
There is no analyze on the old program. I have used the desktop version for analysis. The 7d to 5d is just the limitation of the phone vs a server most likely.

Hello,
small question regarding anlaysis - how to use it? I know how to get feedback on next move - but in go sequences I think are more important. So even if move is ok locally - maybe after sequence is very bad. Any idea how to analyze for more movies?

If you mean the desktop version, I wrote a post Tips for using Crazy Stone to analyze games.
Author:  Solomon [ Mon Oct 10, 2016 8:38 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
Majordomo wrote:
Heh, I didn't count - but I assume you won the game?

Yep, played inattentively and still won handily by over 50 points.
Author:  WinPooh [ Tue Oct 11, 2016 3:22 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
What is the meaning of "Situation" and "Delta" columns in Analysis grid?
Author:  Uberdude [ Tue Oct 11, 2016 4:16 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
I don't have this software, but with CrazyStone "Delta" usually means the difference in winning probability between your move and what CS thinks is the best move. So look for the biggest numbers to find the biggest mistakes in the game. The absolute size (or some average) could perhaps also be used to infer the level of play, for example comparing this Crazy Stone Analysis Remi did of my game a few years ago (before neural nets): http://www.grappa.univ-lille3.fr/~coulo ... /Uberdude/ to a weaker player's http://www.grappa.univ-lille3.fr/~coulo ... /bofinken/ you can see my Deltas are usually smaller, though both games had a "biggest mistake" of 0.11.
Author:  oren [ Tue Oct 11, 2016 9:35 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
Uberdude described delta well. Difference between CS best move and yours. Situation is the score that CS predicts for the end of the game.
Author:  pookpooi [ Tue Oct 11, 2016 9:45 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
Very interesting that this app run on Nexus 5 get 6d on KGS blitz game (crazyphone)
Does that mean blitz magic work too well for computer?
Author:  Majordomo [ Wed Oct 12, 2016 1:48 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
While only a datapoint of 1, I just wanted to try it at a 1d setting and it promptly lost by resign to NeuralZ02 (which is ~3k in the description). While just a few days ago it won against the bot while set to 3k - strange or just expected variance? I've noted in the game list that the bot does seem to beat more 3k's than it loses to but still.

EDIT: And then it trashed it in the second round, so yeah, some variance there.
Author:  Satorian [ Wed Oct 12, 2016 10:15 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
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.
Author:  Rémi [ Wed Oct 12, 2016 2:20 pm ]
Post subject:  Re: Android - Crazy Stone Deep Learning
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.
Author:  Javerberg [ Wed Oct 12, 2016 3:30 pm ]
Post subject:  Re: Android - Crazy Stone Deep Learning
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.
Author:  Satorian [ Thu Oct 13, 2016 6:26 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
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.
Author:  lobotommy [ Thu Nov 03, 2016 3:41 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
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?
Author:  Mike Novack [ Thu Nov 03, 2016 7:05 am ]
Post subject:  Re: Android - Crazy Stone Deep Learning
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.
Page 1 of 2 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/