If you guys do this please count me in! I’ve read through it but feel like I’m missing something important or maybe just not appreciating the style. In any case it seems totally impractical for my games but that also makes it perfect material for the forum.
3 Likes
I use miai counting to get “temperature” and “count” as understood in Combinatorial Game Theory (a mathematical theory applicable to many games not just go). I happen to be a fan and have studied a lot of combinatorial game theory since long before getting into go, so biased “mathy approach” here 
As a personal recommendation, note that learning “endgame tesujis” and sequences is more important than “very precise calculations”, as once you have the basic principles of how to compute the value of moves, computing the value extremely carefully and precisely but using a “suboptimal game tree” that ignores the tesuji/best play is much worse than rough estimation but being tesuji-aware (as you will get a “very precise” but completely wrong answer). So endgame is tsumego all over again after all
If you are interested on how to understand “computing values and counts” (that is, miai counting), and you happen to understand Spanish, I gave two lectures about it for the Argentine Go Association Youtube Channel long ago: Lecture 1, Lecture 2
I have that book and I have read it complete once and understood I guess 95% of all the main ideas (has a lot of details and specific examples, many of those I checked just superficially or “trusted the main result without any detailed verification of the example shown”, so I remember little of those detailed rules / case analysis). Indeed, for the purposes of “getting stronger at go”, my advise is to completely ignore that one. That is a book primarily about math/combinatorial game theory as a very interesting theory for itself, and not about getting better at go practically.
More specifically, almost all of the book is concerned with Tedomari, and in fact only with Tedomari for the very last point (as the Tedomari concept appears anytime that “the value of the largest move suddenly decreases”, as in “the last big fuseki point” which is a Tedomari possibility not covered in the book). The main thing studied in the book is the situation when there are a lot of possible moves on the board, ALL of which have a miai value of EXACTLY ONE, so we will be gaining 1 point per move whatever we play anyway, and we thus only care about being THE LAST to do a 1-point move, as that way we are 1-point better than our opponent. Thus almost the whole book delves into a lot of CGT very technical stuff and extremely special-case studies of how to get exactly 1 more point in special situations (no more, no less, ONE point at most of advantage by perfectly understanding the theory in 95% of the book). So extremely interesting math, but beware not efficient at all for getting stronger at go .
The Tedomari concept itself is useful though so check that one. Interestingly, in tedomari problems it is very often the case that “my best move is NOT the opponent best move”, unlike “normal” game situations. This is because in normal situations, “play largest area” is best play, and “largest area” does not depend on whose turn it is, so both players want to play there.
Most of the time the orthodox strategy is optimal play (but not always: exceptions are mainly because of ko or tedomari). Orthodox strategy [in endgame situations] is basically:
“On your turn, compute the miai value (CGT temperature) of each local independent area of the board, and play in the largest one. However, if the opponent plays any local-sente move, answer those immediately instead of following previous rule.”
If there is no ko, it can be proved mathematically that this “simple” strategy deviates little from optimal play: if the largest miai value is currently X, we are guaranteed to get a score no worse than X points from perfect play. Note that this strategy completely ignores tedomari considerations: if there are many same-value moves, just play any, don’t care about tedomari. So if current largest miai value on the board is 3.8, this strategy guarantees a score at most 3 less than optimal. Even if say, 40 moves remain until the end of the game, meaning at most 3 of those 40 moves will not be perfect moves if we follow orthodox strategy.
That is why it is most of the time the correct thing to do, and why “after our count is ahead by some points, we win even when playing very safe very submissive answering everything”. This justifies in a sense that “accurate and fast estimation of miai values” is orders of magnitude more useful than tedomari study.
Interestingly, since judging values exactly is hard / doing actual math computation is very time consuming, my understanding is that pros actually rely a lot on just reading (which they have to massively train anyway for life and death/middle game), known patterns, and intuition. From what I read in a nordic go dojo article by Antti long ago, pros often use a combo of (tsumego-like-reading + intuitive approximate evaluation + tedomari) in place of precise miai calculations:
- First the “big moves” are identified intuitively and the rest discarded (these will usually already have the largest miai values if we compute them, and be very close in actual value say maybe 2.7 to 3 miai value, otherwise the pro pattern practice/intuition + a little bit of reading would notice the size difference easily).
- Then the pro reads different orderings of how to play those “currently big moves”, as if it were a big global tsumego, until “the value drops” and only “smaller moves” remain.
- For this global tsumego, the pro will try to play a sequence that ensures they get tedomari: they get to be the last player to move while there are still “big moves”. So it is a tsumego where winning is not “group lives” but “I got the last move (from the big ones)”.
In positions where there are reasonable “drop points” in move value, this pro “judge-big-vs-small + read-variations” style leads to very strong endgame, possibly beating the orthodox strategy by actually getting tedomari. If instead the value decreases very very slowly and continuously, it is not natural to separate “big moves” from “small moves” anywhere. In those endgames, the orthodox strategy and precise miai values find best play.
5 Likes
I think that I found the article that I mentioned (although I am not quite sure it is the same that I read before, but it certainly touches upon the same main idea)
It includes a simple illustration of how in tedomari situations, it is very often the case that optimal move areas for different players differ.
3 Likes