Could I have jumped 4 stones in strength in a week?

[Babelardus]

Hi

Since the beginning of this week, I noticed that I am suddenly playing stronger than before. While I don't play on KGS (yet), I do play against several bots/engines taking the appropriate handicap. I'm guesstimating my strength using this page at Sensei's Library.

Last week:
Playing against Cosumi, my score was a bit less than even; maybe 40%. Cosumi is estimated to play at +/- 8k. My 40% score suggests I was a bit weaker. Let's say 9k or 10k.

Playing against Gnu Go 3.8 at level 10, I needed to take 4 stones handicap to make it an even game. GNU Go is estimated to play at 5k/6k. Well, Me having to take 4 stones handicap again suggests that I played at 9k/10k level.

Against Fuego and Pachi I didn't stand a chance, not even when taking 5 to 9 handicap stones.

I've been playing like this for some time, until I decided to start watching Nick Sibicky's lectures from the very start; especially the ones in which he reviews kyu games. And I was like: "I do that too. And that too. All bad moves. Crap... Hey, that DOES work? Why doesn't white just... (explains)... Oh. Yeah."

I've been watching a lecture a day, and since the beginning of this week I try to put some of that stuff into practice, with these results:

Now I beat Cosumi most of the time, with score between +0.5 and +12.5. Sometimes I get lazy if I think I've won, and it succeeds at an invasion and I lose.
Against GNU Go on level 10 I'm now obtaining a somewhat better score than 50%, taking no handicap
I have a score of of 35% wins when playing Fuego at 4 stones handicap while before I couldn't even come close. I'll have to try at 5 or 6 stones. (The 2008 version of Fuego is estimated at 2k, but the current version is stronger, and computers are faster; playing against Pachi on my computer, I estimate it at 2d, assuming Pachi's estimate of 4-5d is correct.)

Pachi is estimated at 4d, but that was in 2012. I expect it to be stronger now. Taking 9 stones at a 5d Pachi would be good for a 4k player. If I take 9 stones now, I mostly lose, with margins within komi, and sometimes, rarely, I win.

While I don't know my real rank, I can objectively say that I did jump four stones in strength because of the results against the bots and my changed handicaps. The handicaps between me and the bots as described above are looking consistent, at least to me. If so, and the page on Sensei's library estimates the bots' ranks correctly, I would now be a KGS 5k. I don't really care about the rank; more about the improvement.

I wonder; is it realistic to move up in strength that much, just from watching lectures? In that case, I must keep watching and hoping I encounter things I'm still doing wrong and get an explanation

Edit, PS: I'm not actively trying to find and abuse weak spots the bots may have. That would just be fooling myself.

Edit2: Won against Cosumi again, by 15.5 points, with Black. If you replay it as a stronger player, you'll probably scream at the mistakes I undoubtedly made

[Bill Spight]

Yes, you can jump 4 stones in a week. I did, myself, 11 kyu to 7 kyu. The problem with such a jump against bots is that bots do not learn how you play. It is possible that you learned things that let you take advantage of the weaknesses of bots, even though that was not your aim, and they do not adjust how they play against you. Maybe now would be a good time to set up a 5 kyu account on KGS and see how things go.

Congratulations!

[Babelardus]

Thanks

I've been thinking about a KGS account, but I'm quite reluctant to be honest.

Bots are strong enough now to run alongside a real game and be stronger than the person that is actually playing. In chess, this has been an issue since the beginning. Chess servers such as FICS, where I had accounts in the past, have been ruined because people are dishonest; either not playing themselves, or playing far below their actual rating. I've read/heard that it's called sandbagging in the Go community (in Chess, it's just called 'cheating your opponent').

I don't mind losing at all, if I feel I've played a good game, but losing to a sandbagging/bot-using opponent just boils down to a waste of time. My experiences in the past have really soured me on online play.

Maybe I'll try it and start at 20k and see how it goes.

[jeromie]

With KGS sandbagging isn't much of a problem (in my experience) because you don't get to pick your rank. Other servers may have more of a problem (Tygem is notorious for having strong players in the DDK ranks). But if you start at 20k when you know you're stronger, that's the very definition of sandbagging.

[dfan]

1) You can definitely make a big jump quickly if parts of your game are severely deficient and holding you back; repairing those areas can make a big difference all at once. Usually this happens at weaker levels but I went from 10k to 6k quite quickly, apparently mostly by simply deciding to play like a 6k instead of a 10k. (Progress since then has been slower...)

I agree with Bill Spight that such an apparent improvement is more possible when you're playing a very small number of opponents and may have learned more how to specifically beat them and less how to play better in general.

2) Computer cheating is the last thing I would worry about on KGS. I'd be surprised if 1% of games have any cheating in them (honestly the thought had not even occurred to me until you brought it up). Then again, I have never personally noticed a cheating problem on the ICC chess server.

As jeromie says, it's a lot harder to sandbag on KGS - you have to lose lots of games intentionally. You won't even be able to set your starting rank at 20k, I believe; you'll start with a ? and your rank will be set automatically once you win and lose a few games.

[Uberdude]

My 40% score suggests I was a bit weaker. Let's say 9k or 10k.
Me having to take 4 stones handicap again suggests that I played at 9k/10k level.

I don't mind losing at all, if I feel I've played a good game, but losing to a sandbagging/bot-using opponent just boils down to a waste of time. My experiences in the past have really soured me on online play.
Maybe I'll try it and start at 20k and see how it goes.

Hmmmm....

[Jhyn]

I don't mind losing at all, if I feel I've played a good game, but losing to a sandbagging/bot-using opponent just boils down to a waste of time. My experiences in the past have really soured me on online play.

Somebody stronger accepts to play with you. A great learning opportunity. At my level, an opportunity people are ready to pay for. How is it a waste of time?

If they play with a bot, how is it any different than playing against a bot yourself?

Anyway this won't happen in a majority of games.

[Uberdude]

My experiences in the past have really soured me on online play.

Is this with Chess? Go players are a nicer bunch of people.

[Babelardus]

With KGS sandbagging isn't much of a problem (in my experience) because you don't get to pick your rank. Other servers may have more of a problem (Tygem is notorious for having strong players in the DDK ranks). But if you start at 20k when you know you're stronger, that's the very definition of sandbagging.

OK, thanks

I understand it may even look like I'm sandbagging myself if I start at 20k, but the point is that I actually don't "know" what my rank should be. The closest thing I can say is that it is (probably) somewhere between 5 and 10k.

I don't mind losing at all, if I feel I've played a good game, but losing to a sandbagging/bot-using opponent just boils down to a waste of time. My experiences in the past have really soured me on online play.

Somebody stronger accepts to play with you. A great learning opportunity. At my level, an opportunity people are ready to pay for. How is it a waste of time?

If they play with a bot, how is it any different than playing against a bot yourself?

Anyway this won't happen in a majority of games.

If someone who is stronger than me decides to play a game, that's fine; I'd love to However, when playing with someone who has '6k'behind their name and he/she pounds you into the ground, it's no fun to find out that their 'other' account is rated at 6d. Good to know that it doesn't happen often on KGS.

My experiences in the past have really soured me on online play.

Is this with Chess? Go players are a nicer bunch of people.

That was with Chess indeed. I haven't played Chess seriously (or against people) for at least 15 years. One of the reasons is that many of the people around here are either casual players, rated 200+ ELO points lower than I am (most of them) or über-serious players who are rated 200+ points higher (only a few). When I quit chess competitively in my late teens, my rating hovered around 1850 ELO. I was still improving, but quit because of university and because I wanted out of the weekly competitions.

When looking through the competition charts of the three (very small) Chess clubs around here, I mostly see very old people (compared to me; in their late 50's up to 70+), playing a game once a week, very late at night. Most are still the same people, still playing at the same ratings, as I played against 15-20 years ago. They don't study Chess anymore; they just play a weekly game. That's fine, but I'm neither one of the weaker nor stronger players, and I don't have the desire to pick up Chess anymore in such a way to hang with the 2000+ ELO guys.

[Mike Novack]

If you were playing against just ONE computer program then the explanation that you had learned how to take advantage of a systemic weakness in the program would be tenable. But since you are reporting about the same amount of improvement against several different programs, that is a far less likely explanation than that you have learned not to make some systemic errors of your own.

Seems like these lectures have really helped you.

BTW, when thinking about the strength of computer programs, especially the computationally intensive MCTS ones, remember that hardware matters. If you were running Pachi on your own machine probably more like 1-2 dan unless you are giving it a LOT of time per move. When you see "was 4 dan back in 20xx" take a look at "was running on" (how powerful a machine). To compare with your machine, don't just count cores; look up benchmark data.

[Babelardus]

If you were playing against just ONE computer program then the explanation that you had learned how to take advantage of a systemic weakness in the program would be tenable. But since you are reporting about the same amount of improvement against several different programs, that is a far less likely explanation than that you have learned not to make some systemic errors of your own.

It is the reason why I am using more than one engine.

It is also the reason why I'm compiling engines myself, to be able to have the most powerful version available under Windows; the Linux guys often don't compile for Windows, or the version is only 32-bit, or behind in updates. (Stupid Oakfoam 0.20 won't compile. It says I'm missing a needed library. It's installed, because Fuego uses the same library and it DOES compile. Grrr.)

Seems like these lectures have really helped you.

They have, because in the early kyu game reviews, I saw Nick comment on (or die laughing about/swearing over) many moves that I also played. I also saw him point out several things. "Black can attach here..." Yeah, everybody says that, but WHY? When I tried it, it did never work, until I saw Nick's explanation. Same with his shape video's. "This/That shape is bad." Yeah, I know already. WHY? Nick explained.

Just knowing that something is good or bad is only useful if you know why that is.

Now I'm looking for more lectures, because now, I'm having a different problem. I 'just' play a game, kill some stones, lose a few stones myself, and have the general feeling that I didn't do anything really wrong, but then I lose by 25 or 30 points. Why? I see a territory forming, and if I invade, the group gets killed. If I don't invade, I lose by about the size of that territory.

I seem to be unable to prevent my opponent (the engines) to create a larger territory than I can, without me being able to get an invasion to live, ever. And I don't know why. I did see a video about invasions in Nick's playlist... maybe that will add another two stones

BTW, when thinking about the strength of computer programs, especially the computationally intensive MCTS ones, remember that hardware matters. If you were running Pachi on your own machine probably more like 1-2 dan unless you are giving it a LOT of time per move. When you see "was 4 dan back in 20xx" take a look at "was running on" (how powerful a machine). To compare with your machine, don't just count cores; look up benchmark data.

Do you know if it is listed anywhere, if I can see somewhere what computer the bots where/are running on?

This self-built computer will probably outrun most 'normal' off the shelf computers for the next 7 or 8 years. I've just replaced my old computer, which was 8 years old as well, but which is *still* in the midrange today, except for graphics card power. (Fortunately, I don't play many new games.)

This new rig is about as high end as you can possibly purchase for desktop use (two CPU's, twelve cores, and upgraded to 32 GB RAM today). I'm just hoping Pachi's 4d designation doesn't come from running on a 128 core cluster sporting 256GB of RAM, because then my rig would indeed be underpowered.

[skydyr]

Now I'm looking for more lectures, because now, I'm having a different problem. I 'just' play a game, kill some stones, lose a few stones myself, and have the general feeling that I didn't do anything really wrong, but then I lose by 25 or 30 points. Why? I see a territory forming, and if I invade, the group gets killed. If I don't invade, I lose by about the size of that territory.

I seem to be unable to prevent my opponent (the engines) to create a larger territory than I can, without me being able to get an invasion to live, ever. And I don't know why. I did see a video about invasions in Nick's playlist... maybe that will add another two stones

The easiest way to get help with this sort of problem is to post a few games here and ask for comments or reviews, so you can get feedback as to where you are going wrong. It's not always where you think it is.

Regarding invading vs losing, one thing to do is to try to anticipate the territory formation earlier, and to play to prevent it, even if it means giving up a bit somewhere else instead. The other thing that you don't seems to have mentioned is that you have a few options to deal with large moyos (frameworks that could turn into territory). You COULD invade, but it's not always the best option. The other choices are to reduce (playing a cap or shoulder hit or something with the aim of making your opponent's potential territory smaller) and/or to set up a competing territory to counter it elsewhere on the board.

Even if you're in a situation where you must invade or lose, there are still a couple possibilities as to what is going wrong. First is that your opponent is too strong locally, and there's no real chance. The other is that you need to make your life and death skills better because you should live but don't.

Finally, an invasion doesn't necessarily have to live to be successful. If you invade an area, you might be able to give up the group in such a way that you can get lots of reducing moves on the outside that your opponent has to respond to suboptimally in order to keep the invasion dead.

As I said, though, post some game examples where things go pear-shaped for you and we'll be able to tell you which of these apply, or if it's something else entirely.

[Mike Novack]

Do you know if it is listed anywhere, if I can see somewhere what computer the bots where/are running on?

This self-built computer will probably outrun most 'normal' off the shelf computers for the next 7 or 8 years. I've just replaced my old computer, which was 8 years old as well, but which is *still* in the midrange today, except for graphics card power. (Fortunately, I don't play many new games.)

This new rig is about as high end as you can possibly purchase for desktop use (two CPU's, twelve cores, and upgraded to 32 GB RAM today). I'm just hoping Pachi's 4d designation doesn't come from running on a 128 core cluster sporting 256GB of RAM, because then my rig would indeed be underpowered.

Listing per se, no. But often you see described the hardware (what CPU's) and those you can look up at any of the benchmark sites for a crunch comparison. A powerful video chip used as a video chip wouldn't help, but some of the programs might be able to make use of vector processing. Memory might or might not help a great deal. That could depend on whether the implementation needs a lot of data kept in memory.

[Babelardus]

As I said, though, post some game examples where things go pear-shaped for you and we'll be able to tell you which of these apply, or if it's something else entirely.

If I encounter such a situation again, I'll post the game. One thing I have to say is that I dislike moyo games with a passion. i really do. They make me feel like playing a positional chess game: nothing seems to happen. In a positional Chess game, the players move their pieces endlessly, often 50-60 moves, in the hope to get on tiny small advantage somewhere that can be used to win the game. In a moyo game, I feel as if players are just fencing off big chunks of territory, and that's that.

In Chess, I look for extremely tactical combinations, as deep as I can get them. Then I move a pawn, or sacrifice a knight against a pawn, and the board explodes. In Go, I like 'snake games'; fights all over the board, where players end up with a nearly full board, and the territory is something like 45 points per player. A game that ends with 100 points per player is just... meh.

That's another thing that goes wrong often. I'm having one little fight, and then my mom got scared and I had to move to my uncle in Bel Air.... Eh, wait. Wrong Youtube video. But seriously, often I'm doing well in a fight, just a few moves away from killing a humongous group... at which point I make a mistake, and I end up with one liberty short somewhere among my cutting stones, 50 moves later. It has to be a mistake by me, because you're not going to tell me an engine can already read 50 moves ahead right now and already knew it would win the fight in that way.

I sometimes have the same in Chess, starting a combination, breaking open the board, beginning a checkmate attack sacrificing half of my pieces, and then being unable to actually play the checkmate move because of the bishop that landed at square a2 15 moves ago now protects the the g8 square, because the pawn of the same color that was in the way on d5 got trashed and the diagonal is now open.

If this happens, I sometimes have the urge to... eh... trash something.

Listing per se, no. But often you see described the hardware (what CPU's) and those you can look up at any of the benchmark sites for a crunch comparison. A powerful video chip used as a video chip wouldn't help, but some of the programs might be able to make use of vector processing. Memory might or might not help a great deal. That could depend on whether the implementation needs a lot of data kept in memory.

I'll see if I can find something. I'm quite confident though, that if a bot is not running on something like a 4+ CPU server, cluster, or the cloud (i.e., it's running on a normal computer), my current rig will be more powerful than anything available in 2008 or even 2014. As I said, if I'd want to upgrade this computer now, I'd need to move to a 4-CPU server or something.

[Bill Spight]

Moyo games are boring?

You might take a look at these games: viewtopic.php?p=60253#p60253

Let me recommend the fourth one, with Chisaku's great sacrifice.