Apparently, I still don't understand Leela

Fedya · #1

I've been trying to play more aggressively, and the results so far have been less than successful. I seem to get a bunch of weak groups, and my attempts to stop my opponents from getting good moyos don't really work.

In this game, when I went over it, I noticed Leela seemed to think my chances weren't that bad. As far as I can tell, I'm either supposed to be able to kill something (first I tried the left, then the top; neither worked), or get a bunch of territory in the center (that didn't work either).

I had Black.

Tryss · #2

at B6 should work :study:

jlt · #3

You were supposed to kill the Q18 group?

Bill Spight · #4

Looking at your games over the years, I say Bravo! :clap:

Yeah, you lost this one, and you took your eye off the ball a couple of times, but you made a lot of good plays, and your game made sense.

I'm off to bed, but I'll make a few more comments mannana.

Bill Spight · #5

A brave effort!

A few comments. No guarantee of correctness, OC.

BTW, I had to insert the escape character, \, in this file, as well.

Fedya · #6

Tryss: for :b67:

Leela spends 99.5% of its effort on C11. So when I played that, Leela looks at either B5 or D11 for :w68:

which both lead to complicated fighting, but with White saving the group.

jlt: At the end of the game, Leela suggests R18 and more complicated fighting, instead of resigning. I didn't see that the Q18 group was in any danger.

Fedya · #7

Fast-forward to 67 to see how Leela saved the group on the left for White:

Bill Spight · #8

Fedya wrote:

Fast-forward to 67 to see how Leela saved the group on the left for White:

Leela gets :b69:

wrong unless it does not want to capture the White stones.

Joaz Banbeck · #9

Bill Spight wrote:

...***tesuji...

Fedya wrote:

Fast-forward to 67 to see how Leela saved the group on the left for White...

Silly computer.

Black 73 at B10 still kills.

( 4 fills at C11 and 8 fills at B6 )

Click Here To Show Diagram Code: [go]$$Bc ( 4 fills at C11 and 8 fills at B6 ) $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . O . . O . . . | $$ | . . . . . O . . . . . X . O . . . . . | $$ | . . . O . . . X . , . X O . . X X . . | $$ | . . . . . . . . . . . X . O X . . . . | $$ | . . . . . . . . . . . . . O X . O . . | $$ | . . . . . . . . . . . . . . X . . . . | $$ | . . 2 . . . . . . . . . . . . . O . . | $$ | . O X O . . . . . . . . . . . X . . . | $$ | . 1 O X X . . . . , . . . . . X O . . | $$ | . 3 O X O . . . . . . . . . . X O . . | $$ | . X X O X X . . . . . . . . X O O . . | $$ | . O 5 O X O . . . . . . . . . . . . . | $$ | 6 X O O O X O . . . . . . . X . X . . | $$ | . O 7 X X X . . . . . . . . . . . . . | $$ | 9 X X X O X . . . , X . . O . O . . . | $$ | . . O X O O . O . O . X . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

You should know by now that if Bill and Leela disagree, you should believe Bill.

Bill Spight · **#10**

Joaz Banbeck wrote:

You should know by now that if Bill and Leela disagree, you should believe Bill.

Better make that Uberdude, not me.

mitsun · **#11**

Bill already gave a good review, but let me add my thoughts. I will try to not be too redundant, except where warranted

I think your first slightly wrong move was :b19:

. Taking the corner is purely territorial and not really bad, but at this stage of the game, efficiency of stones is more important than territory. So locally the best move is Q14, keeping everything connected and making your thickness truly impressive. Try to appreciate that this move is at least as good as taking the corner.

You did well in the fighting in the lower right and lower left, up to :b47:

. This was your worst failure of the game. Please play over this position and convince yourself that E5 is the only move. Even without reading, this is clearly the required shape move, and here it works well tactically. Fortunately for you, W returned the favor with :w48:

, blocking the wrong way. W should block at E5, paying attention to the important side, giving up one stone on a small scale. My version of Leela says your mistake cost 20% win rate, while your opponent's mistake cost 25% win rate. The game is still pretty even at this point.

:b59:

was a similar position to :b47:

, but you got it right this time, bravo! Now W is in serious trouble. You played fine here, getting great outside strength in return for letting W live inside. You say the left side attack did not work, but actually it worked very well, not for killing, but for winning the game. W swallowed two stones, making maybe 8 points of territory, while B got thickness totally controlling the center. After :b75:

, Leela thinks there are half a dozen good moves to choose from with 80% win rate. Simplest is probably defending at P7.

bernds · **#12**

I don't think anyone has commented on the following yet. The position in the lower right remained like this:

Click Here To Show Diagram Code: [go]$$Wc19 $$ . . . . . . . . . . | $$ . O . . . X . X . . | $$ . . . X . . . . . . | $$ . X . . O . O . . . | $$ O . X . . . . . . . | $$ . . . . . . . . . . | $$ . . . . . . . . . . | $$---------------------+[/go]

for a very long time. If the colors were reversed, it would look like a handicap game, one player getting surrounded in the corner - and a dan player would bully his opponent by playing on the 3-3 point. Why not try that? Best case the two white stones get captured.

mumps · **#13**

bernds wrote:

I don't think anyone has commented on the following yet. The position in the lower right remained like this:

Click Here To Show Diagram Code: [go]$$Wc19 $$ . . . . . . . . . . | $$ . O . . . X . X . . | $$ . . . X . . . . . . | $$ . X . . O . O . . . | $$ O . X . . . . . . . | $$ . . . . . . . . . . | $$ . . . . . . . . . . | $$---------------------+[/go]

for a very long time. If the colors were reversed, it would look like a handicap game, one player getting surrounded in the corner - and a dan player would bully his opponent by playing on the 3-3 point. Why not try that? Best case the two white stones get captured.

Actually, because of M3 and N5 stones playing at the 3-3 point gives White more chances to live or connect out.

Try S4 instead to kill White.

Fedya · **#14**

Bill: You suggested peeping on :b27:

. To me, that looks like the sort of move where, if I played it, I would get criticized for playing aji keshi. Indeed, I thought of that in your comment on move 103 up in the top left for strengthening White needlessly.

Is there any good way to figure out which such moves are aji keshi and which aren't short of having really good reading?

(And I'm not certain why Leela wouldn't want me to kill the White group around C6.)

Bill Spight · **#15**

Fedya wrote:

Bill: You suggested peeping on :b27:

. To me, that looks like the sort of move where, if I played it, I would get criticized for playing aji keshi. Indeed, I thought of that in your comment on move 103 up in the top left for strengthening White needlessly.

Is there any good way to figure out which such moves are aji keshi and which aren't short of having really good reading?

Well, there isn't any one way. I am attaching an edited SGF file with a couple of comments. Go to move 27 and also to move 104,

A knowledge of shape can help. Empty triangles figure into the question of :b27:

.

The main question of aji keshi has to do with alternate plays that you give up. A variation for :b27:

shows the main play that is given up. It's not a good play. In this case the peep can be called a no loss kikashi, for the reason.

Also, in general, when attacking it is a good idea to peep at the group you are attacking. Usually the connection will make those stone heavy, that is, too big to sacrifice easily. Often, as in this case, the peep will let you move your attack further ahead than otherwise, and strengthen your attacking group.

Quote:

(And I'm not certain why Leela wouldn't want me to kill the White group around C6.)

Because they read the whole board, today's bots can be weak at local reading, including ladders, semeai, and life and death. And Leela-11 is not all that strong, anyway. Bots do not incorporate knowledge like humans do. The sagari to the second line is something that humans learn, but does not occur that often in real games, so a bot might not learn it quickly by self play. So we do not know if Leela actually considered that tesuji, or the sequence shown by Joaz. Letting White live gives Black a good game, so Leela may have had no reason to search locally for a kill. I learned that tesuji when I was around your strength, so I saw it immediately.

dfan · **#16**

I don't know about Leela 11, but Leela Zero knows how to kill White on the left; it just prefers to let White live and make thickness in sente. (I tried a couple of different networks and the amount of preference varied.) I showed it the position after the kill, to be sure it was evaluating the right thing.

Edited to add: This does not mean that killing White on the left is "incorrect" or a bad idea!

Apparently, I still don't understand Leela

Who is online