Life In 19x19

Regarding minor updates, yes I think. CS 2013 my version is v1.03 & I remember at least one of those updates installing. CS DL is v1.00.

Drew wrote:Those of you who own CS, could you please share some feedback regarding what features you think it should have but lacks?

EDIT - for example, the Android version which I own does not support a branching move tree. You also cannot move backwards along the move timeline during a game, only once the game is over. This is an archaic limitation compared to the current capabilities of free Android go clients.

Yeah, here are some things I'd like to see.

-Ability to easily export analysed SGF files and easily review them with top variations from CSDL analysis.

-Ability to tell CSDL to use exact time for moves. So if you want it to always minimum use 1 minute to make sure it reads stuff properly... I'd like more flexible time controls and the ability to give CSDL more time, or for it to use more time. Even at 7d level it usually takes longer than me but still misses many crucial things, so I'd like it to take as much time as it needs. Perhaps this is an issue with the search being too limited as remi mentioned before. Top priority to get this fixed!

Really nice program don't get me wrong, but it's barebones and utilitarian. I like that. No need for an excess of features, but I'd like atleast a smooth SGF editor/analyse mode. Current one is clunky and difficult to use.

Drew wrote:Those of you who own CS, could you please share some feedback regarding what features you think it should have but lacks?

The short version is that it should be easy to use Crazy Stone as a tool to analyze games, and it currently isn't. The raw materials are mostly there, but the UI is often actively hostile to the process. For example, here is how I analyze a different move in ChessBase:

• Play a move

and here is how I do it in Crazy Stone:

• Click on the board window to bring it into focus
• Click again on the point I want to play
• Dialog comes up asking if I want to continue, click Yes
• Dialog comes up asking if I want to branch, click Yes
• Now I have left analysis mode, so select Analyze / Analyze Current Game to go back into it.

That's literally five times as many actions.

dfan wrote:

Drew wrote:Those of you who own CS, could you please share some feedback regarding what features you think it should have but lacks?

The short version is that it should be easy to use Crazy Stone as a tool to analyze games, and it currently isn't. The raw materials are mostly there, but the UI is often actively hostile to the process. For example, here is how I analyze a different move in ChessBase:

• Play a move

and here is how I do it in Crazy Stone:

• Click on the board window to bring it into focus
• Click again on the point I want to play
• Dialog comes up asking if I want to continue, click Yes
• Dialog comes up asking if I want to branch, click Yes
• Now I have left analysis mode, so select Analyze / Analyze Current Game to go back into it.

That's literally five times as many actions.

Spock would say: That is highly illogical.

So... you can't just put CrazyStone into "Analysis Mode", where it just keeps running in the background, providing the best found variation? And then, when you click a button, add that variation to the game tree?

Chessbase and it's assorted programs such as Fritz have been able to do something like that one way or another for 20 years now. I understand that for Go, it wasn't really useful, as 10 years ago, the strongest program would play around 6k or so. Anyone with a bit of drive and effort can reach 6k, I think; at least, I feel that this is somewhere around intermediate club level strength, comparable to a 1500-1600 ELO-rating or so in Chess. (

Now, however, we're at the "Deep Blue" moment in Go (a dedicated computer powerful enough to defeat a top human player), and, just like in Chess in 1997, Go engines for normal computers are starting to reach the highest amateur grades. In Chess, that would be somewhere around 2450 ELO, in Go it would be around 6-7d.

I'm going by this assumption between Chess ELO and Go ranking:
1800 ELO is starting point for "serious amateur/advanced club player" <==> 1d amateur
2450 ELO is top amateur level <==> 7d amateur
2500 ELO is entry grandmaster level <==> 1p professional starting out
2850 ELO is top grandmaster level <==> 9p professional in his prime

The algorithms to play go effectively have been found. They are different than the ones for Chess, but it will be a matter of time before Monte Carlo/UCT/Neural Nets get optimized and extended to need less and less power, just like it happened in Chess with MiniMax/AlphaBeta/Quiescence/MTD(f)/Null move etcetera.

Now, 20 years after Deep Blue, a Chess Engine plays at grandmaster level on a somewhat current phone. It may take another 10-20 years for Go to reach that stage, but at some point, it will. For now, the top engines are becoming good enough to run at the strongest amateur level on a powerful off-the shelf computer... just like Chess engines in 1997.

Thus, they will be becoming more and more useful as analysis tools. I wonder if someone is thinking of a professional program like Chessbase, but for Go. At this point programs like Drago and Kombilo are very servicable already (and free even!) but they are only at the level Chessbase was at in the early 90's.

There certainly is a market here.

Babelardus wrote:So... you can't just put CrazyStone into "Analysis Mode", where it just keeps running in the background, providing the best found variation? And then, when you click a button, add that variation to the game tree?

You can run it in analysis mode, and view the current estimated winrate, spread, etc., as well as a list of moves it is considering, but
1) the moment you make a new move (e.g. branch) it turns analysis mode off
2) you don't get to see any variations, just one move ahead

Chessbase and it's assorted programs such as Fritz have been able to do something like that one way or another for 20 years now.

Mogo did it, it was super cool to see what line it expected. I would guess that Crazystone also does it (seems impossible that Remi would not want this feature for his own testing) but the Unbalance software doesn't.

However—here come the apologetics—however, I find that missing this aspect is not particularly crippling and might actually be an unexpected feature in the sense that if CS suggests some alternate move, it's up to you to read and see why it might like it (or, alternatively, reject it as a "bot move"). For instance, in some crosscut fight I defend one side and CS thinks it is easily 1% better that I defend the other, then it's nice to look at the board position one move back and try to read what might happen. In tight tactical battles, though, seeing variations would be very helpful. (Which is the primary use in chess engine reviews.) My problem in go is rarely tactical, so seeing variation trees for me would be neat but not necessarily a feature I'd want to pay another $10 for.

The Go Engine Server for the iPad version was updated on Jun 21st, says the App Store. That's a fabulous way to play when traveling.

Do we know how the levels 1-10 map to kyu/dan rankings? I noticed level 3 was easy win as 6 k KGS. It would be nice to select the strength by a rank in the iPad server version too, especially if it was trained and calibrated by using ranks, and mimics the mistakes that humans make.

In the server, am I playing against against a neural network or Monte Carlo, on each level?

Link to store: https://appsto.re/fi/lI2rC.i

Marcel Grünauer wrote:

Petteri wrote:The Go Engine Server for the iPad version was updated on Jun 21st, says the App Store. That's a fabulous way to play when traveling.

That's very good news; I've been waiting for that. Just the other day I bought a six-month ticket; if it's even mid-dan strength it's a pretty good deal for EUR 12,-.

CrazyStone is pretty strong in local fights, so it's good to discover new moves and blind spots and play through various lines using the Undo function.

It's excellent value! And for a kyu player, it just destroys you tactically. You can see why Alex Dinerchtein said somewhere that strategy is pretty much useless for beginners. No plans survive contact with the enemy!
It'd be nice if Remi posts here to his subscribers the KGS equivalent of the new ipad server version levels, since it uses games from that server to mimic human play.