tunesmith's study journal

[tunesmith]

Creating a study journal sounds like fun.

I've posted once so far, over in Amateurs: http://www.lifein19x19.com/forum/viewto ... 12&t=11490

I am a complete beginner, and seem to be hanging out in the 20kyu range lately.

Since then, I attended the Seattle Go Center once - the lecture was a bit too advanced for me, but afterwards I played a 15kyu player who gave me five stones. I lost, but not by a ton - it was fun, and my first person-to-person game on a full-size goban. It's too bad I can't play back the human games in an SGF file.

I haven't tried playing a human in KGS yet, I feel like I am still making too many blunders and I want to solidify a bit. So I'm sticking with playing AI for now.

My saga lately has been in playing against GnuGo using either sente.ch's Goban, or GoGui. I basically start with a 9-stone handicap and work my way up. Playing 9x9, I can sometimes win with a 2-stone handicap.

Then I tried 19x19. That's been the big drama lately. The first time I tried with a 9-stone handicap, I still lost by 88. Then, I lost by 44. Then, I lost by 15. This was all over the last week or so. And just now, finally, just a couple minutes ago, I finally won my first 19x19 against GnuGo, by making white resign. I think the main thing I focused on this time was extending on hane instead of crosscutting, which was reinforced by me finishing Janice Kim's Volume II last night. Anyway, it seemed to help me create territory.

I might post those game records here for my own posterity. I'm sort of nervous to play GnuGo again since I hate the idea of taking one step back after two steps forward, so I think I'll take a couple of days before I try eight stones.

The other thing I've been doing is geeking out on making GoGui play against itself. I found that Fuego 1.1 beats GnuGo pretty readily, while Pachi beats Fuego (all just using default settings). I was just able to play Fuego for the first time yesterday ('brew install fuego' now works when it didn't before). I played 9x9 giving myself a 3-stone handicap, and I lost by 0.5, argh! But the game record ended up being really interesting because I was able to find the one latest move where if I let fuego take over my own moves, it was still able to eke out a win, and another variation of my own where black was able to win faster after I gave fuego control again. Both variations involved invasions - I feel like I have a lot to learn in knowing when I can invade a corner and expect to be able to make life.

[tunesmith]

To celebrate beating GnuGo at 19x19 (with 9 stones handicap), I thought I should try to get a ranking on KGS. After first playing a human (my first go server human!) and winning by resignation when I thought I was on the way to losing, I realized it was a "Free" game and didn't count for my rankings. So then I tried to find a ranked bot, and found the 16kyu LibertyBot.

It was kind of a strange game, because I was intimidated by its ranking and I honestly spent most of the game feeling like I was being demolished, but then it turned out I only lost by 9.5 - and if I had managed to defeat that invasion in the lower left, I think I would have been able to eke out the win. Maybe C2 at would have done it? So I must have stumbled into some good moves somewhere in there (unless black was holding back near the end after counting it was ahead).



KGS ranked me at "17k?" after this game. Since that time I played the 23k "Weak Bot" at a six-stone handicap, and beat it, so now KGS sees me as "17k". I'm sure it'll vary wildly for a while.

[S2W]

Hello tunesmith - welcome to life in 19x19

Here are are few variations - take them all with a pinch of salt

[tunesmith]

Thanks S2W! I appreciate the comments.

Yeah, it looks like a lot of my loss was due to moves like and and W-158 and W-172. These are the sorts of moves I currently expect myself to recognize and I just didn't. I think I was feeling a bit jackrabbity because I was playing a 16kyu at zero handicap, intimidated.

The variations around W-228 and W-230 are especially interesting for me because I don't quite feel able to recognize those situations in games yet. I think I can solve them as problems since a problem usually communicates that something clever is able to be done, but they still kind of sneak up on me in games. I should still probably try making some problems out of these scenarios... I guess any time there is a cut point somewhere on a wall I should try to read it a bit more to see if there's a way to take advantage of it. Anyway, thanks for those. W-240 is a good next step up for me, too - I feel like I still haven't quite had that first experience of committing a successful invasion - I tend to either miss opportunities, or take opportunities and regret it later.

[Elom]

Poorly done by a weak player, with too much text, and may not be extremely useful, but I just wanted to add a few thoughts (take them with caution )



Maybe if you improve your feel for shape and haengma, you could play with few worries!

Edit: Realised that a third of the commentary went missing, and fixed a few errors.

[tunesmith]

Wow Elom, those are great comments, thanks! Some of it is beyond me and I'll be busy looking up some definitions now. Definitely saving this to review a few times over the next little while.

[EdLee]

Hi tunesmith,

My $0.02

getting out is big. P15, at 5-5, is a simple way.
The last empty corner is big, but you don't want to be boxed in by B here.

B can box you in with P15. P15 is a big shared vital point.

P15 still very big.

P15.

P15. B takes the UR corner; you take the LR corner.

a pincer feels quite natural.

P4 and F4 are two big, shared vital points.

Have you already finished about 100 games yet ?
If yes, good! If no, do it.
Either way, continue to play, play, and play.
Enjoy.

[tunesmith]

A few days ago I was able to beat GnuGo on 19x19, at a 9-stone handicap, for the first time. So tonight I tried to play with an 8-stone handicap and lost. This is the game - I ran the record back through GnuGo afterward to see where GnuGo would have moved as black, and those moves are in the commentary, along with some of my own comments.

[tunesmith]

I challenged LibertyBot again after my last match - all the comments must have helped! I won my "revenge match" convincingly, and KGS now has me ranked at 15 kyu. My rank seems really inflated to me, of course - I'm still regularly getting stumped on 20-kyu level go problems. But this game was encouraging in a couple of ways because I was able to fight off a couple of invasions that probably would have screwed me up before, and I won a couple of capturing races, too.

It's also seeming a bit more obvious that LibertyBot is probably making some really weird moves that a human of that level wouldn't make - I'm imagining it's strong in some ways and weak in others. Some of those capturing races were weird, like I couldn't believe I got . So I have to get back to playing humans soon. Still expecting my rank to bounce all over the place for a while.

[tentano]

Don't rank yourself by go problem ratings. They're not a reliable indicator.

The only true rank is when you enter into a competitive system, and you win and lose some games.

If you want to be more certain it's a deserved rank, just play more games. The results of those games will prove what you're worth, over and over. If after a few dozen games, it still says you're 15k, then that's what you are.

The bot's rating is valid, even though it makes different mistakes than a human player. It has played against a lot of ranked humans, so you can be sure it's accurately rated.

[skydyr]

Some comments:


In general, I'd recommend looking at capturing races and counting the liberties, so that you know if and when you need to capture and don't waste moves needlessly. Making sure that you don't get broken shape is the other. When white pushes straight through two of your stones or groups, without getting cut in return, it can't end well.