Last night I had a rather fun/interesting game in the European league.
Quote:
The opening was Alpha/Leela Zero inspired (I've recently been studying with the Lizzie interface, highly recommended!), with a kiai-filled tenuki fest in which I ignored a double approach on my 4-4 to press and surround his group. We then had a bit of fuseki on the top side, I took sente instead of the honte defence when he invaded my 4-4, plus a big knight to resurrect my 4-4 stone that was surrounded. Rather than pulling out his cutting stone to start a fight, he extended on the side so I loosely netted it. We had a bit more opening type stuff with him playing territorialy, so I played two attachments to develop the centre, he plopped in a reduction stone on tengen and I probed to see if he'd give me territory or keep centre access to help his reduction. He opted for the former, so he jumped in further to the moyo, but with a big knight move instead of the one-space jump I expected. So I cut it, he jumped out with half the group to safety, but I got a thick shape in the centre and felt good as I'd switched my moyo from upper to lower side. When I grew this, he played a surprising yose move (s15) so I surrounded more, he played a nice shoulder hit reduction but extended one too many times meaning I played a thin net, which ended up capturing him thanks to a cute little tesuji I just managed to find in overtime. He then failed to kill my corner and resigned.
Some detailed commentary and AG/LZ review:
The lower right approach instead of patient knight move answer in lower left is hardly a new idea, but it being clearly superior is an AlphaGo idea I've embraced (maybe some individual pros thought it, checking a pattern search Kim Suchang and Cho Hunhyun liked it), LeelaZero #139 agrees. The idea is obviously that white thinks the press is bigger and more painful to ignore than the double approach to 4-4 (the natural "Hey, you tenukid my move!" idea) which AlphaGo Teach thinks is a significant mistake (wants to answer approach with knight move or high one space pincer). So when black double approached I pressed as AGT suggests (if he covers at e5 I'm happy to cover at p3), LZ #139 prefers to answer the double approach but I suspect as she improves she will come to agree with AG about press. AG's suggested answer to the press when you really want sente, as here, is interesting: the r4 kosumi; if crawl and jump white gets sente to play with the 4-4. So in my game black resisted that and crawled twice and then surrounded at e5, a shape you normally don't want to allow (but which my wife with her contrary style likes to allow, more on that later). AGT doesn't give this line but I presume it prefers white, it's similar to a
Master vs pro game in which Master nonchalantly allowed the same surround (by move 27 AGT says black is down to 30% win). The idea seems to be the e5 surround isn't so amazing give white's nice profit elsewhere (even though black still lived) and that you can still live in the corner fairly large later and the black wall's value might not be so great (which is why I then played c13, corner-obsessed LZ wants to live in upper right or lower left). (LZ thinks s3 should be s4, it did make it easy for me to tenuki.)
Unsurprisingly LZ doesn't like my side move at k17 (-2%), I expect AG wouldn't either but I still have some fondness for such weak human moves. Black's immediate 3-3 struck me as premature (I thought he might finish off lower left 3-3, LZ wants to n17) as k17 means I can reasonably omit the honte capture and take sente to live in the lower left, LZ agrees -4%. LZ says f4 should be e4 which is what I expected and have seen in bot games when they live in this corner later. I thought black's c10 might be soft (vs e15, though I figured I could handle that fight) as it allowed me to "have my cake and eat it" of living lower left and capture d15 cutting stone, but LZ says it's ok, but not a huge fan of my bad-aji net (d11 press best for efficient swallowing up, or e15 honte).
p17 is a traditional human territory-closing move, but LZ doesn't like it, prefers r7 to ask eye questions of my wall and make use of s4; and then wants me to r13 attach instead of simple q10 extension. s5 surprised me as a premature yose move (my s4 isn't sente, could make sense if it was) that strengthened me (he lost invasion aji) but maybe he thought it overconcentrates me, LZ -5%. LZ agrees with m17, seems pretty traditional. I thought about g3 cut here but didn't like tenuki or h4 answer, but went for d10 attachment because those are cool these days (LZ likes it). I thought he would probably hane but he was soft/solid/patient again with extend, LZ initially extends then hanes. The plus for him was he got sente when I pulled back and I expected him to reduce at l15 or k16 as this seemed miai, but LZ likes his h3 defence. So I attached again for the lols and there didn't seem much else to do (LZ agrees).
LZ agrees strategically with the expected reduction, but thinks tengen was too shallow, but even if deeper thinks I can patiently answer one space behind to defend the territory. I thought about k11 attach to wall off the top territory (expect lots of extends, if he hanes I cut, LZ likes) but wasn't confident of being so passive given all black's solid territory (I suppose LZ is confident of keeping a 5 points lead, I'm not) so r13 attached to see if he would give me some centre kikashi to cut off that escape route in which case I might try to attack tengen (but not so easy as not deep). LZ says r15 was soft but he got sente to go further into the moyo. LZ likes his big knight move even though I can cut it, if one point jump it says just kosumi answer to defend the territory is enough! LZ says L13 dodge was good (I focused on his k11 block and then cut being good for me, considered L12 a bit, should read broader but it's not like dodging refutes my move) but n13 soft, should hane even though I then wedge and cut and fight. Interestingly it initially (2k playouts liked my thick l10 but then at 10k prefers h14 (by 1%) to defend the top territory; even though black did then reduce the top a fair bit I stayed connected and I prefer the centre thickness.
LZ doesn't think of my m12 peep (prefers g11 which I considered but decided too dame even though good topologically), which I played to avoid a messy fight if L11 cut, she shows this is fine for me but 50+ moves long so I don't mind playing slightly lossy peeps to avoid such variations. c5 was decent but she prefers a move in lower centre: j7 rather than the k7 I played later. I think c5 is good for me because now that h3 is defended black b4, which morally should be a gote sequence, could lead to me messing up and dying in the corner if I tenuki afterwards (and giving b4 in sente is too sad), so reducing opportunities for me to blunder in overtime is a good game strategy. I expected black d5 as sente for corner pain so was pleased he gave me sente again with e8 but LZ says it's a fine move.
I'm pleased I identified the right area of the board with k7, though LZ prefers j7 by about 2%; I was being cautious of the k10 aji. Black then played an inexplicable yose move at s15 (but only -4%, I'm at 90% now and did feel happy in the game), I suppose I have the t15 kosumi sente yose tesuji (didn't see this at time) but I expected something around o11 (LZ's move) or k4, maybe he thought these were still miai? L6 was indeed a nice reduction point, but my k4 was indeed a good move (I could have played it instead of m7 push, but I did that to make my 2 stones into a stronger 3 and give black a chance to go out, but his natural local shape n6 was too much and let me pounce with k4. I was in overtime and didn't expect this to be a clean kill, and LZ says he can get out with L4 but I get some nice profit on lower side. In the game l2 tesuji means j4 wasn't sente on lower 2 stones (LZ prefers firmer h7 atari over h8 net) so then black dies and it's game over.
So overall I'm pleased with how the game went, though black's rather slack play meant he didn't challenge me to prove my moves were ok (e.g. not activate d15, or try to punish loose f14, s5, s15). This weekend is the British Challenger's league, I'm not sure yet how seriously I'll play and how much to fool around with crazy bot ideas, but this game has encouraged the latter. Some of the players though I know would put up more of a fight (even if they are overall weaker than Martin Li) so maybe slow and steady conventional play has more win chance. I'm there to have fun though...