Kirby's Study Journal

Kirby · Post by **Kirby** » Wed Sep 04, 2019 10:09 am

Thoughts from LZ with barely any playouts...

LZ seems to think tenuki is OK here, but does low:

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . X . . . | $$ | . . B . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . X . . | $$ | . . . . . . . . . . . . . . . O . . . | $$ | . . X . . . . . . . . . . . . O X . . | $$ | . . . . . O . . . . . . . . . O X . . | $$ | . . X O . O X . . , . . . . . , X . . | $$ | . . X . O X O O . . . X . . O . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

LZ didn't like this peep:

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . O . . O . . . . . . X . . . . . | $$ | . . . , . . . . . , . . . . . X . . . | $$ | . . . X . . . . . . . . . X . . X . . | $$ | . . . . . . . . . . . . O O . O O . . | $$ | . . . . . . . . . . . . . . B . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . X . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . X X . . | $$ | . . . . . . . . . . . . . . . O . . . | $$ | . . X . . . . . . . . O . . . O X . . | $$ | . . . . . O . . . . . . . . . O X . . | $$ | . . X O . O X . . , . . . . . , X . . | $$ | . . X . O X O O . . . X . . O . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Wanted to just reduce instead:

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . O . . O . . . . . . X . . . . . | $$ | . . . , . . . . . , . . . . . X . . . | $$ | . . . X . . . . . . . . . X . . X . . | $$ | . . . . . . . . . . . . O O . O O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . X . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . B . X X . . | $$ | . . . . . . . . . . . . . . . O . . . | $$ | . . X . . . . . . . . O . . . O X . . | $$ | . . . . . O . . . . . . . . . O X . . | $$ | . . X O . O X . . , . . . . . , X . . | $$ | . . X . O X O O . . . X . . O . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Right here, LZ seems to think I can get back into the game by just grabbing the 3 stones:

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . O . O . . . . . . . . . . . . . . | $$ | . . X O . . O . . . . O X X . . . . . | $$ | . . X , . . . . . , O . O X . X . . . | $$ | . . . X . . . . . . . . O X . . X . . | $$ | . . . . . . . . . . X . O O O O O . . | $$ | . . . X . . . O . . . . . . X . . . . | $$ | . . . . . X X O . . . X . . . . . . . | $$ | . . X . O O O X X . . O O O . X . . . | $$ | . . . , O X X O X , . O X X X , . . . | $$ | . . . . . . X O O B . O X O . . . . . | $$ | . . O O O O X . . . X X O X . X X . . | $$ | . . X . X X X . . . . . O . . O . . . | $$ | . . X . . . . . . X X O O O . O X . . | $$ | . . . . . O . O . X O X . . . O X . . | $$ | . . X O . O X . . X O . . . . , X . . | $$ | . . X . O X O O . X O X . . O . . . . | $$ | . . . . . . . . . X O O . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

That's all I have time for now. Maybe with more playouts LZ has a different opinion.,

Bill Spight · Post by **Bill Spight** » Wed Sep 04, 2019 12:55 pm

Kirby wrote:Thoughts from LZ with barely any playouts...

{snip}

That's all I have time for now. Maybe with more playouts LZ has a different opinion.,

For analysis, I wouldn't trust LZ with fewer than 10k playouts.

Usually the play in the game, if it is different from LZ's choice will not have many playouts, and, as Uberdude points out, running LZ long enough to bring the number up can be prohibitive. In that case, make the play and base the winrate on LZ's choice, which should have more than 10k playouts. For instance, if the Black peep has fewer than 1k playouts, play it and see what LZ's choice is for White's reply. For instance, it may choose to connect, with 30k playouts. Then take the Black winrate estimate of the connection as the estimate for the peep.

Kirby · Post by **Kirby** » Wed Sep 04, 2019 1:04 pm

Word!

If only I had more than an hour for my lunch break, I could let it run longer

I like the idea, though. Seems like it will save some time.

Magicwand · Post by **Magicwand** » Wed Sep 04, 2019 2:11 pm

Peep is a bad move. Lizzy is correct.

Bill Spight · Post by **Bill Spight** » Wed Sep 04, 2019 2:12 pm

Kirby wrote:Word!

If only I had more than an hour for my lunch break, I could let it run longer

I like the idea, though. Seems like it will save some time.

Many thanks to yoyoma for basicly giving me the idea.

EdLee · Post by **EdLee** » Wed Sep 04, 2019 2:57 pm

For analysis, I wouldn't trust LZ with fewer than 10k playouts.

Anyone remembers how many playouts in the AG-LeeSedol matches and the AG-KeJie matches ? Was it on the order of 10M playouts ?

Uberdude · Post by **Uberdude** » Wed Sep 04, 2019 3:33 pm

Bill Spight wrote:
Kirby wrote:Thoughts from LZ with barely any playouts...

{snip}

That's all I have time for now. Maybe with more playouts LZ has a different opinion.,
For analysis, I wouldn't trust LZ with fewer than 10k playouts.

Depends how strongly you want to be sure LZ is right. In a game of Kirby's level if LZ says a move is bad with a few hundred playouts and another is better I wouldn't take it as gospel, but it's probably true at least 9 times out of 10. It's like how a pro can often point out your mistakes just based on shape intuition without doing any reading.

Ed, I don't think that was publicly stated, but maybe we can spy in screenshots of AlphaGo movie. AG teaching tool is 10million.

Bill Spight · Post by **Bill Spight** » Wed Sep 04, 2019 3:37 pm

EdLee wrote:
For analysis, I wouldn't trust LZ with fewer than 10k playouts.
Anyone remembers how many playouts in the AG-LeeSedol matches and the AG-KeJie matches ? Was it on the order of 10M playouts ?

Well, I meant playouts for a single option. The total number for all options would be rather bigger, OC.

I used to think 4k playouts was good enough for analysis, but based upon Elf's GoGoD commentary, I now think 10k may be necessary for today's top bots.

EdLee · Post by **EdLee** » Wed Sep 04, 2019 4:49 pm

Hi Bill,

don't think that was publicly stated, but maybe we can spy in screenshots of AlphaGo movie. AG teaching tool is 10million.

Thanks. Ah, part of the secret sauce.

Kirby · Post by **Kirby** » Wed Sep 04, 2019 5:07 pm

Uberdude wrote:In a game of Kirby's level if LZ says a move is bad with a few hundred playouts and another is better I wouldn't take it as gospel, but it's probably true at least 9 times out of 10.

Stated pretty bluntly, but I suppose you're right :-p

Bill Spight · Post by **Bill Spight** » Wed Sep 04, 2019 9:42 pm

Uberdude wrote:
Bill Spight wrote:
Kirby wrote:Thoughts from LZ with barely any playouts...

{snip}

That's all I have time for now. Maybe with more playouts LZ has a different opinion.,
For analysis, I wouldn't trust LZ with fewer than 10k playouts.

Depends how strongly you want to be sure LZ is right. In a game of Kirby's level if LZ says a move is bad with a few hundred playouts and another is better I wouldn't take it as gospel, but it's probably true at least 9 times out of 10. It's like how a pro can often point out your mistakes just based on shape intuition without doing any reading.

Indeed, that's so. However, as the AlphaGo people took pains to point out, humans don't think like bots. They made too strong a case, IMO. It's true that humans don't calculate probablities like bots do, but we do have a general sense, based largely on experience, of our chances of winning a game, and we do base decisions on that assessment. Since the parallel processing of our brains is unconscious, we do not have any direct notion of anything like the number of playouts, and there may be nothing at all like that in our parallel processing. Familiarity may be similar in some ways to a high number of playouts, or maybe not, quien sabe?

To say that a pro relies upon intuition without doing any reading is, AFAIK, not exactly accurate. The pro may not be doing any conscious reading, but is surely doing unconscious processing. If we don't call that reading, what do we call it? Maybe we need a new term.

I have related how, after the first Environmental Go Game between Jiang Jujo and Rui Naiwei, almost 20 years ago, Professor Berlekamp and I were going over the game with Jujo. The previous night I had discovered what I was pretty sure was a mistake by Jujo and we were talking about that. Rui Naiwei stopped by and looked at the board for two seconds and rattled off an 8 move combination that she had overlooked during the game. That sequence was not simply the result of intuition, or she would have found it during the play, when she had devoted more time to conscious reading.

Also, bots are telling us that pro intuition is much more fallible than we thought. Moves that were considered go common sense, even knowledge, we now have to admit are probably errors, and not just minor ones. If we are talking about intuition alone, minus unconscious processing, I'm not sure that we can give pros a 90% grade, or bots, either. Quien sabe? Nobody has researched these questions.

Sure, if we simply play a top bot's preferred play, we will play at a superhuman level. But what about other plays? Even plays that are not on the bot's radar? We know of cases where humans have made such plays, which we can prove are better than the bot's choice. There are surely human plays that are better, but we cannot prove that they are.

You don't have to know much in the way of statistics to see that low playouts are a problem with winrate estimates. That does not affect the bot's level of play very much, because its chosen move typically has a large number of playouts. (Besides, bots do not choose moves based upon winrate estimates alone.) All it has to do to play well is to find a good enough move. Its high number of playouts reflects our confidence in its winrate estimate. If all we wanted to do when reviewing games with bots is to find good moves, there is no problem. But humans also want to find mistakes. And there is where the problem with low playouts hits home. If the supposed mistake has a low number of playouts, we cannot be very confident of its winrate estimate, which is what we use for evaluation. From what I have observed, if the number of playouts is less than 1k, we cannot be confident at all.

Not too long ago I thought that 4k playouts was enough to inspire confidence. And recently I lowered my estimate of the margin of error for Elf to 4%. (I think that it's probably 3% for LZ, but I do not have as much experience with its estimates.) And since then I have been shocked to find an example where Elf is demonstrably off by at least 5% with a playout count of around 6.7k.

Now, maybe this is a rare anomaly, but in the past I have found errors with Leela 11 (now no longer a top bot) of at least 20%.

The way bots work is not familiar to humans, and, unfortunately, it seems like we have been encouraged to think that there is such a thing as an objective winrate, which bots figure out, and the number of playouts does not matter, and furthermore, that you can use the difference in winrates to identify errors. That's malarkey, IMHO. We have to learn how to use the numbers that bots tell us. How many people are working on how to do that? I am, in my spare time. Who else? I don't know.

Kirby · Post by **Kirby** » Thu Sep 05, 2019 5:49 pm

If I had to give an evaluation, I'd say that today wasn't a great day. My foot doesn't hurt to walk on it anymore, so I made the resolution to start running again this morning. When my alarm went off at 6:30, I didn't get up - call it lack of willpower, lack of initiative, or just plain laziness... I didn't want to get up, so I slept for longer, and didn't work out in the morning. After getting off to a slow start, I had some breakfast, cleaned up, and got to work later than usual. Getting to work later than usual was OK, though, because today is go club day - I'd be working late, anyway. Since I was off track already, I made no plans to do go problems or stick to my usual routine - I had the single goal of getting a code review out for a project I'm working on. I was able to focus a bit, but progress was slower than I'd hoped. In the early afternoon, I had been making some progress on the project, but my wife texted me and told me that she had smelled something bad in my laptop bag at home. She noticed there was a dish inside of it, so she went to dump out the contents of my laptop bag in the sink... And out fell a dead rat. She was freaked out, and wanted me to home and take care of it. Incidentally, she started getting some allergic reaction to something, and wasn't feeling great. So I went back home to get rid of the rat, then headed back to work. Round trip: about 50 minutes taken away from working on my project at work. So then I continued for awhile. It got to be around 5:00pm, and I was feeling guilty about not helping out at home. So I decided to come home for dinner and forgo the go club today. Coming home, I felt a bit guilty that I didn't put my project up for code review. I suppose I would have felt guilty either way. At home, kids were arguing about stuff, things were somewhat of a mess, and the atmosphere didn't seem that great. I recommended we eat out today, and that cheered the kids up a little bit - that was one of today's highlights - I had a good time seeing them happy. As a side note, I still feel a sense of anxiety, because I'm not completely done with some of the home projects I've been working on - some rooms are only partially completed. I was thinking of working on the house this evening... But overall, I just feel really lazy today.

Aside from kids being happy for eating out, the other highlight of today was that I received Inseong's review from my game a few days ago.

He had a single overarching theme for my play, which I can relate to: I am not satisfied with a 50-50 result, and I overplay. I played many mistakes, but most of them seem rooted from this idea.

Here are some examples he gave:

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . X . . . . . | $$ | . . . , O . . . . , . . . . . X . . . | $$ | . . X . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . O . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . X X . | $$ | . . . . . . . . . . . . . . . . X O . | $$ | . . . . . . . . . . . . . . O X X O . | $$ | . . . . . . X . . . . . . . . X O . . | $$ | . . O O O . . . . , . . . . . X O . . | $$ | . . X X . . X . . . . . . . . X O . . | $$ | . . . . . X . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

In the board above (I'm white), white has a nice position on the left; black has a nice position on the bottom. But I feel a loss. Why? Because black has a nice position on the bottom. This is what Inseong stated, and it's true - I do feel a loss. I don't like that black can get points there.

The move that I played?

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . X . . . . . | $$ | . . . , O . . . . , . . . . . X . . . | $$ | . . X . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . O . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . X X . | $$ | . . . . . . . . . . . . . . . . X O . | $$ | . . . . . . . . . . . . . . O X X O . | $$ | . . . . . . X . . . . . . . . X O . . | $$ | . . O O O . . . W , . . . . . X O . . | $$ | . . X X . . X . . . . . . . . X O . . | $$ | . . . . . X . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Overplay. Despite black's stones, I don't want to give black points there. Inseong asked in the review if that made sense. Black has a better position on the bottom. So black SHOULD get more points there. That's natural. That's the game. But I don't want to.

Inseong said that his instinct was something like this:

Click Here To Show Diagram Code: [go]$$W $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . X . . . . . | $$ | . . . , O . . . . , . . . . . X . . . | $$ | . . X . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . O . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . X X . | $$ | . . . . . . . . . . . . . . . . X O . | $$ | . . . . . . . . . . 5 . . . O X X O . | $$ | . . . . . . X . . . . . . . . X O . . | $$ | . . O O O . . . . , . . . . . X O . . | $$ | . . X X . . X . . . . . . 4 . X O . . | $$ | . . . . . X . . . . . . . . 2 1 3 . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

This wasn't intuitive to me at all, because the exchanges from

through

make black strong - and then black will get points there. I don't like that.

But Inseong says it's the wrong idea. Black *should* get points there, because black has a position there. He said it'd be equally weird to expect black to get more than 50% on the left where white has a good position. But I don't seem to have a problem with that idea.

Another example is here:

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . X O . . . . . . . . . . . . . . . . | $$ | . X . O . . . . . . . . . X . . . . . | $$ | . W . , O . . . . , . . . . . X . . . | $$ | . . X . . . . . . . . . . . . X X X . | $$ | . . X O . . . . . . . . . . O X O O O | $$ | . X O X . . . . . . . . . . X O X X X | $$ | . O O . . . . . . . . . . X . O X . X | $$ | . . O . . . . . . . . . O X O . O X . | $$ | . . . , . . . . . , . . O X X , O O . | $$ | . . . . . . . . . . . . . O X . . . O | $$ | . . . O . . . . . . . . . O X . X X O | $$ | . . . . . . . . . . . . . . O . X O X | $$ | . . . . . . . . . . O . . . O X X O . | $$ | . . . . . . X X X X O . . . . X O . O | $$ | . . O O O . X O O O X O . . . X O . . | $$ | . . X X . . X . X X X O . . . X O . . | $$ | . . . . . X . . O X O . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

I have weaknesses on the outside, and there's no way to kill black. But I want to kill black - or threaten to. But it could have been a disaster:

Click Here To Show Diagram Code: [go]$$B $$ --------------------------------------- $$ | . O . . . . . . . . . . . . . . . . . | $$ | . X O . . . . . . . . . . . . . . . . | $$ | . X . O 5 . . . . . . . . X . . . . . | $$ | . O X 3 O . . . . , . . . . . X . . . | $$ | . O X 2 4 . . . . . . . . . . X X X . | $$ | . X X O 1 . . . . . . . . . O X O O O | $$ | . X O X . . . . . . . . . . X O X X X | $$ | . O O . . . . . . . . . . X . O X . X | $$ | . . O . . . . . . . . . O X O . O X . | $$ | . . . , . . . . . , . . O X X , O O . | $$ | . . . . . . . . . . . . . O X . . . O | $$ | . . . O . . . . . . . . . O X . X X O | $$ | . . . . . . . . . . . . . . O . X O X | $$ | . . . . . . . . . . O . . . O X X O . | $$ | . . . . . . X X X X O . . . . X O . O | $$ | . . O O O . X O O O X O . . . X O . . | $$ | . . X X . . X . X X X O . . . X O . . | $$ | . . . . . X . . O X O . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

But I wanted a result better than what was possible for that board position.

I am reminded of the game from the US Go Congress I posted about in this post: viewtopic.php?p=247321#p247321

Here's what I said in that post:

Kirby wrote: But I couldn't find a way to invade properly on the top.

Despite this... I played there anyway!

I was looking for a chance for my opponent to screw up - a way to break into the top with a bad move on his part. I was putting the fate of the game into his hands, hoping that he'd make a mistake.

...

The question that remained in my mind for quite some time was, "WHY did I play in this way?"
It wasn't a mistake in reading - I didn't read out a variation that worked, and then be surprised by a move that I didn't expect. The fact is, I didn't read out a variation that worked. But I wanted to play there anyway.

Why? The question is exceedingly interesting to me. Why did I want to play a sequence that I hadn't thought to work?

In the days that followed the US Go Congress, I came to an answer: I am not confident when the game becomes close - and in these cases, I succumb to overplay in hopes that my opponent will make a mistake.
Here I had a reasonably close game that I could very well have won if I had just played in a normal way. I didn't require special moves. But my lack of confidence when the game is close led me to want to overplay - I wanted to break into the top, whether it worked or not! I had an expectation that my opponent could likely make a mistake, and then I could set the score such that the game wasn't close anymore.

More than the moves themselves in this sequence, the psychological state that I was in to decide on making these moves was, and remains, very fascinating to me.

To overcome this, I need to be more confident when the game is close. I need to be willing to play normal moves when I'm just a couple of points ahead - and not something weird to give my opponent a chance.

I feel that this is the same problem that Inseong is describing to me. When there's a 50-50 result on the board, I am nervous. I don't want a 50-50 result. I want 80-20; or 90-10. But 50-50 - that is too scary. And when black has an advantage in a certain local area - I want to get a better result than black. Because I'm not confident of the result when the game is close.

I don't really know how to fix this psychological problem. Maybe it's a matter of playing some games, consciously not trying to do this. But games are 200 or 300 moves, and often, in concentration, it's hard to deviate from my nature... I have a feeling that studying endgame could be one way to help. It's boring to me... But maybe it'll give me confidence when the game is close.

I wonder if there are other ways that could help me overcome this psychological problem...

Bill Spight · Post by **Bill Spight** » Thu Sep 05, 2019 10:00 pm

One thing that strikes me about your overplays, at least, these ones, is that your opponents let you get away with them. (Quick thought: Maybe you should play against stronger players who will punish your overplays.

)

As for the one in the top left corner, it could have ended badly for you, but it didn't. You saw the kill, you just missed the weakness in your own position, but you saw that before he did.

All's well that ends well.

The one on the bottom side really surprised me. Not that I don't sympathize with wanting to rein in Black's moyo. It is not only fairly large, it is easy to develop it. OTOH, White's sphere of influence is loose, and it is not all that clear how to develop it. That's one reason why on White's previous turn I would have played the keima on the edge of Black's moyo and White sphere of influence, instead of enclosing the Black stone in the top left.

Boy was I surprised! You attacked a live group instead of protecting your three stone wall that had neither an eye nor an extension!

And Black fell for it. He attacked your lone stone instead of your weak group, and then captured it in such a way that you were able to build a successful invasion. Wow!

Yeah, making overplays is a bad habit, but you seem to be getting rewarded for doing so. Technically, one thing that may be lacking is viewing the board through your opponent's eyes. For instance, when he played the second line turn on the right side, did you think he was only preventing you from connecting? Maybe he was, but I expect that he was aware of weakening your corner.

Also, perhaps you don't do much planning. As Znosko-Borovsky says, at first you don't worry about what your opponent will do, you get clear what you want to do. But go is a game of trade-offs, and any plan needs to consider what your opponent can get in exchange for what you get. You remarked before switching to the top left corner that it wasn't clear which was bigger, the left side or the bottom. Indeed, that was the case, at least for players like us. But there is your potential trade-off right there: If I stake out the left side, my opponent can stake out the bottom, and if I invade or reduce the bottom, he can go into the left side. Then you can consider that exchange before deciding where to play.

Kirby · Post by **Kirby** » Fri Sep 06, 2019 4:42 am

Indeed, I gave only part of the review. Inseong’s comment for my opponent was that he passively accepts my overplays. Inseong’s point to me was that I want an 80-20 result. If my opponent lets me get it, I win the game; if he punishes my overplay, I lose the game. He said I should be more conscious of this tendency I have, or else, I will be unlikely to advance my level.

Maybe poor planning is part of it. Sometimes I’ve figured that maybe Inseong sees this as my tendency to overplay, but maybe if I just read more, the overplay won’t come. In other words, one theory I have is that some players may have tendency to overplay when the situation isn't clear to them; some may have tendency to underplay. Some may play in a different weird way when they haven't read everything out. I happen to overplay when the situation is unclear. But maybe the solution, then, isn't to focus on overplay, but rather to make the situation more clear (i.e. by reading or planning).

That being said, I’m not sure if this is consistent with the Go Congress game I referenced... there, I read the invasion didn’t work, and played it anyway...

EdLee · Post by **EdLee** » Fri Sep 06, 2019 5:46 am

Hi Kirby,

Inseong’s point to me was that I want an 80-20 result.

Seems you ( and Bill and Mr. Inseong ) have already ID'd the problem exactly. So, to beat a dead horse:

One way for you to consistently get a 80-20 edge for a major portion of your game is to play someone who's 2 stones below you, in an even game ( 6.5 or 7.5 komi; nigiri ). In other words, it's an under handicapped game and it's unfair to your opponent. This way, you'll feel 80-20 most of the time.

If you play an identical twin of yourself, how do you get 80-20 ? This is exactly what happens with opponents your level, by definition.

If you play an even game with someone 2 stones above you, then you get the 20-80 feeling.

( BTW, before 2015, many pros felt 50-50, except when facing a top title holder, in which case maybe they felt under 50% odds. Today, I'd say any pro who keeps up-to-date understands their odd is ~0% in an even game vs. a superhuman bot.

)

Life In 19x19

Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re:

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal

Re: Kirby's Study Journal