Feel-good moves

Yes

The OGS KataGo analysis has few playouts. Do not assume it is infallible, KataGo can miss sharp tactics that strong humans can find at these settings. I expect KataGo would find it with hundreds of thousands of playouts. Likewise my +50% game winning sequence. It took an EGF 7d about 5 minutes reading in their head. It took me longer analysing variations.

5 Likes

But here KataGo knows about the move and has evaluated it as stronger than its recommendation. Why doesn’t it recommend the stronger move?

1 Like

It didn’t realise the power of that move until it was played in the game, forcing it to evaluate it more accurately than it did one move earlier. But that doesn’t update the inaccurate evaluation on the earlier move.

3 Likes

My 15-block Katago found Mark’s move after about 10000 playouts.

7 Likes

An EGF 7d found the same move when you asked them?

Neural network feels which moves are worth exploring. Other moves are mostly ignored. It spends a lot of playouts testing moves that look good and and spends very few playouts on other moves. So normal AI is good only at finding at least 1 good enough to win move. To find all good moves, you need modification like this: Releases · AncalagonX/leela-zero · GitHub
where all possible coordinates of next move are explored with equal number of playouts.

1 Like

Yes, I showed that problem to Kim Youngsam. I suppose he’s EGF 8d nowadays.

1 Like

He is more like an 8d EGF :slight_smile:

Would have been at 2011 London open when he was still “only” 7d. London Open | Tournament card | E.G.D. - European Go Database

There’s a setting for KataGo to make it explore more widely which should be used when using it for analysis rather than playing.

Quote from KataGo author:

if you’re doing any form of analysis, I recommend setting “analysisWideRootNoise” to a value like 0.05, or 0.10. This setting makes KataGo explore much more widely at the root, judging a lot more moves, and makes it a little less likely that a move is missed in the way you described.

You can also easily configure KataGo to search wider in general, not just at the root. For example, “nnPolicyTemperature=1.1” and “cpuctExploration=1.5” and “cpuctExplorationLog=0.6”. This will also slightly reduce the risk of missing improved moves within the tree. Or try yet-more extreme numbers, play with it and see.

The reason these settings aren’t default, of course, is that they make play worse on average, holding compute fixed. Because for play, you care a little bit more about finding a move that’s good enough and being a bit more sure of the move you’re planning to play that there is nothing wrong with it when you go deep. Whereas for analysis, depending on the goal of your analysis, maybe you care more about comparing alternatives, or finding the most likely optimal move, rather than a good-enough move, at the cost of searching less deeply and being less sure about any individual move that you might pick.

5 Likes

This move was so pretty, I built a hexagon / benzene ring around tengen. I was tempted to omit 126 to make it a perfect hexagon, but then Black’s clamp at L11 would have been trouble and I decided I wanted to win and not just draw pretty shapes.

4 Likes

Well maybe this one is just a nose tesuji but it felt good anyway.

It also ended the game :slight_smile: (it was a while ago now, but I remember enjoying it)

3 Likes

Not sure if this counts, but I’m quite proud of the sequence starting with move 65 in this game: karatego vs. Wade (which was played on IGS in the european team go championship last season).

I just checked with AI and it is approved by my low-playout KaTrain. At move 68 my opponent should have retreated with J3 but didn’t, which gave me quite an advantage. Soon after I unfortunately lost everything back by a series of bad moves :sweat_smile:

1 Like

A nice atekomi tesuji indeed. Replaying one of my games I linked, there is another atekomi tesuji hidden in there that went unplayed, but was crucial to me being able to push from the outside (thus weakening my m17 top group) along line 10 and then take the lower side profit with e8 Kuksu 5th Cycle Title Match, Board 1, namely j17 as insurance if black attacks from the outside or tries something like n18.

3 Likes

And on that subject, here is a very nice problem, black to play:

2 Likes

Against jlt in the OGS Team League (Terrific’s–Martin’s Board 2).

This position developed from a running fight which began in the opening on (42) and had been in progress, excluding the play on the top side, for something like forty moves.

White had just played X, which is not directly sente against either of my groups on the top side. I may have had time to play M3, but it’s hard to say. There’s a lot of L&D in this position.

In the game, I was very happy to get this move in, since I had felt I was behind (I was, but only – according to the bot – by a small and varying margin of 1–5 points since (55)).

The AI calls it a mistake and says that it loses five points. Possibly it does think I can exchange M3 in sente; perhaps it just wants the greedier K12.

I love this move, though, because it:

  1. coordinates with all my thirteen stones in the top right to claim a large area of the centre

  2. defends my A group

  3. defends my B group

  4. attacks White’s C group

Note that there is still aji with the Δ wedges, though, and the B group is also not completely healthy. The game was eventually decided in the life and death of that B group.

The moyo is also insolid, with some play available to White around O16 and R10. jlt was able to play very skilfully there and pull off a large reduction.

IndieSn didn’t have anything to say about the move in his otherwise excellent review of the game, though.

3 Likes

I didn’t know a video of our game was out there on YouTube. He missed some points in our game, but his style is quite entertaining, he could be a sports commentator.

2 Likes

Yeah I only had a look at the start and the commentary definitely has a lot of energy. Would like to watch more of that video later on.

Wait til we have some kind of e-sports/mind-sports type channel for Go, that’ll be the future of Western Go :stuck_out_tongue:

1 Like

This is a game on KGS and I am black.

Feel good and guilty move, I was crushed before this:

Not that difficult, but I did read out this sequence at move 99, 11 moves earlier.

Then I did not expect my opponent was so convinced:

But anyway, I take what I can get:

6 Likes

Played several games with 4-5D last month. One common comment from them is my shape is not good. So I started paying attention. Was behind in this game, but was able to catch the weakness of white shape. Pretty happy.

White resigned after this move. Very much playable still per AI.

5 Likes