Life In 19x19

Not sure if these reports are worth reading or not but here goes...

I've taken a sabbatical from playing humans (except the beginners tourney). I'm trying to force myself to stop playing reactively and think things through and I don't want to keep my opponents waiting while I cogitate.

I've been reading the Learn To Play Go series and trying to incorporate the techniques into my games against Gnu Go using Drago. Gnu Go is ferociously aggressive and I think it's good training. I haven't beaten it yet but I'm getting a little better I think.

I need to get better at invading. My invasions always end with my annihilation.

My typical games right now end with me having a couple groups that are safe but completely surrounded with Gnu Go in control of a large area that I can't seem to penetrate. My invasions always crash and burn. That's pretty much how all my games against humans went as well. Hmmm...maybe there's a pattern there.

I also need to get better at fending off invasions into my prospective territories or at least limiting the damage they cause.

I know I'm better than I was a week ago and that's progress.

The good news: I don't make the same dumb mistakes I used to. I learned not to repeat those.

The bad news: I've discovered a completely new set of dumb mistakes that I need to stop making.

TominNJ wrote:The good news: I don't make the same dumb mistakes I used to. I learned not to repeat those.

The bad news: I've discovered a completely new set of dumb mistakes that I need to stop making.

Don't worry, I've heard this pattern stops around 7 dan (You just keep redefining the magnitude of "dumb"!)

No! Stop playing the computers!

They'll give you weird habits that are hard to break, but no good against human players. KGS is the way, don't worry if they're human, so are you! (I assume)

Consider how you can play such that the opponent doesn't get solid walls around large areas of the board, eg. don't make contact plays against weak stones, don't try to atari unless it has a purpose etc.

I like the report though, good luck finding another new set of mistakes.

Stable wrote:No! Stop playing the computers!

They'll give you weird habits that are hard to break, but no good against human players. KGS is the way, don't worry if they're human, so are you! (I assume):

I'm not sure that's true at this point. I need to pay more attention to what my opponent is doing and try to keep him from isolating my groups. Playing against Gnu Go is perfect for that.

Stable wrote:
Consider how you can play such that the opponent doesn't get solid walls around large areas of the board,

So far, the only way I can do that is to resign earlier in the game.

Stable wrote:
eg. don't make contact plays against weak stones, don't try to atari unless it has a purpose etc.

Yeah, well it's hard to remember my objective is to drain the swamp when I'm up to my eyeballs in alligators.

I've been working on the technique of trying to keep my opponents groups separate so they can't link up and surround me. Sometimes it's difficult to do that and make sure I have enough eyes to avoid disaster. They're learning games and losing doesn't matter if I learn something. It's hard to play that way on KGS. They keep score on there.

Losing to the software doesn't seem to bruise my ego as much as losing on KGS. I guess because I'm trying things and experimenting.

Next technique to practice: I need to keep invading, invading invading until I can do it and live.

Thanks for the encouragement. I could use some sometimes.

TominNJ wrote:

Stable wrote:No! Stop playing the computers!

I'm not sure that's true at this point.

I agree.

TominNJ wrote:They're learning games and losing doesn't matter if I learn something. It's hard to play that way on KGS. They keep score on there. Losing to the software doesn't seem to bruise my ego as much as losing on KGS. I guess because I'm trying things and experimenting.

And this is exactly why.

TominNJ wrote:Yeah, well it's hard to remember my objective is to drain the swamp when I'm up to my eyeballs in alligators.

Nice quote

What handicap do you take against gnu-go?

emeraldemon wrote:What handicap do you take against gnu-go?

none. I play it straight up. I'm not even sure how strong that program is or even if it has any parameters. I'm playing it just the way it came out of the zip file.

I know it's one of the weaker programs out there but it's more than I can handle right now and the price is right.

I started to download Fuego which I *think* is some variation of Gnu Go but I'm not sure.

If I ever get to where I can beat Gnu Go consistently then I'll try to get something stronger and try some human competition.

I wonder if this should be moved to the computer go section at this point...

It's about 8k EGF from memory, but seems to hold 5k or 6k on KGS last I heard. Either way, beating it even will be a good feat of improvement!

TominNJ wrote:

emeraldemon wrote:What handicap do you take against gnu-go?

none. I play it straight up. I'm not even sure how strong that program is or even if it has any parameters. I'm playing it just the way it came out of the zip file.

I know it's one of the weaker programs out there but it's more than I can handle right now and the price is right.

I started to download Fuego which I *think* is some variation of Gnu Go but I'm not sure.

If I ever get to where I can beat Gnu Go consistently then I'll try to get something stronger and try some human competition.

I wonder if this should be moved to the computer go section at this point...

Apparently the GnuGo bots on KGS are 6 kyu. I suspect that's inflated somehow (perhaps people are less serious when playing the bots). My random guess is that if you play it frequently you can beat it at 10 kyu or even earlier.

Fuego doesn't seem to be based on GnuGo. According to its Sensei's Library page it's 2 kyu on KGS.

I played GnuGo more than humans until I could beat it (and then it became rather dull to play, even giving it handicap). Just try to play people occasionally .

I strongly recommend playing it at a handicap. I would play it at two stones, and if you lose, play it at three, and so on. Eventually you will win, even if you have to take 15 stones. Then play it at 14, and so on.

Why? I taught someone to play once who refused to take handicaps. We played 7x7 games (I printed out a board), she took black without komi, and I always won. It seemed like maybe she was improving, but how could I know? Finally I forced her to take 5(!) stones on the 7x7. She lost that game, but quickly learned to beat me at 5, and eventually pushed all the way to winning with black and no komi on the 7x7. I'm 100% convinced that if she hadn't taken those handicaps and worked her way up, she never would have gotten any better.

If you say "i will play gnugo even until I win", you are trying to jump straight to 10k (or 6k) without ever winning a game! If you can beat it at 9 stones, you can feel good about being a respectable 15k or whatever.

TominNJ wrote:Yeah, well it's hard to remember my objective is to drain the swamp when I'm up to my eyeballs in alligators.

Truth that.

I just remember that I learned all sorts of tricks playing igowin that only work against that bot and probably held me back around 18k for a while. Still, ego is ego. I found losing to a human easier than playing "honestly" against a bot.

The strength of GnuGo depends on the time settings. On KGS where the settings for bots are quite fast, it is around 6-8k. On OGS where you (as a human) have lot of time (and the bot still plays fast) it is around 12-15k. To avoid playing the same opponent all the time, you could try the bot Aya. When I was 12-15k, it was my favorite program. It is only a little bit weaker than GnuGo but has a different style. Fuego is not recommendable, since it often makes nonsense move, if it detects that it is comfortably ahead (or hopelessly behind).

@Tom: If you like exercises, you could try out 321go.org. It contains a go course that comes along with tons of exercises. You have to sign up there, but the site keeps track of your progress. Once you have gone through the program, you should be around 15k.

emeraldemon wrote:I strongly recommend playing it at a handicap. I would play it at two stones, and if you lose, play it at three, and so on. Eventually you will win, even if you have to take 15 stones. Then play it at 14, and so on.

Why? I taught someone to play once who refused to take handicaps. We played 7x7 games (I printed out a board), she took black without komi, and I always won. It seemed like maybe she was improving, but how could I know? Finally I forced her to take 5(!) stones on the 7x7. She lost that game, but quickly learned to beat me at 5, and eventually pushed all the way to winning with black and no komi on the 7x7. I'm 100% convinced that if she hadn't taken those handicaps and worked her way up, she never would have gotten any better.

If you say "i will play gnugo even until I win", you are trying to jump straight to 10k (or 6k) without ever winning a game! If you can beat it at 9 stones, you can feel good about being a respectable 15k or whatever.

Good advice.

Take handicap and focus on killing gnugo's groups and live with your own. With handicap you'll sooner get to the fighting and can skip that lofty mapping out moyo and territory stage called the opening. Fighting skill is the single most important skill to master when you start out (I'm not sure if it ever stops being most important).

Play on smaller boards. 19x19 is very vast and it's difficult to know what mistake(s) lead to your demise. On smaller boards the feedback comes more quickly. I.e. it gets more apparent what moves lost/won the game.

prokofiev wrote:Apparently the GnuGo bots on KGS are 6 kyu. I suspect that's inflated somehow (perhaps people are less serious when playing the bots). My random guess is that if you play it frequently you can beat it at 10 kyu or even earlier.

GnuGo (the most recent one) played without hamete is about 6k, to within a stone either way. I say that because I remember my KGS rank and the handicap stones I took from GnuGo were very well correlated. I haven't played it recently. It's tactical fighting is quite strong compared to my skills when I was DDK, so it's definitely a good learning aid. If your corner is killable GnuGo will let you know

Also, I say without hamete because if you really know GnuGo you can game it to win using sequences and such it's terrible at (sort of meta game hamete). But that's really missing the point, isn't it? If you play it the same way you would play a random opponent on KGS it's 6kyu.

Numsgil wrote:

prokofiev wrote:Apparently the GnuGo bots on KGS are 6 kyu. I suspect that's inflated somehow (perhaps people are less serious when playing the bots). My random guess is that if you play it frequently you can beat it at 10 kyu or even earlier.

GnuGo (the most recent one) played without hamete is about 6k, to within a stone either way. I say that because I remember my KGS rank and the handicap stones I took from GnuGo were very well correlated. I haven't played it recently. It's tactical fighting is quite strong compared to my skills when I was DDK, so it's definitely a good learning aid. If your corner is killable GnuGo will let you know

Also, I say without hamete because if you really know GnuGo you can game it to win using sequences and such it's terrible at (sort of meta game hamete). But that's really missing the point, isn't it? If you play it the same way you would play a random opponent on KGS it's 6kyu.

Probably the difference in our experiences comes down to what karaklis said about time constraints. I tended to take my time when playing GnuGo.

As for hamete, I don't know any specific sequences that work as you describe, but basically as soon as GnuGo stopped trouncing me tactically, it was simply the case that every game I'd play some normal sente move against one of its groups that every human would defend against and GnuGo would play away, and I'd kill its group and coast (even if at a tactical disadvantage) to an easy win. Thus the 'it became dull' sentiment from earlier. It was great for working out the tactical kinks up to that point, though, and I still occasionally play computers on 9x9 (where there aren't enough important groups to play away as above).