Robert Jasiek
2016-03-04 14:22:23 UTC
Here are 4 SGF commentaries on dynamic aspects of AlphaGo - Fan Hui
games. Save each file within the [sgf] tag brackets (without them),
then view offline.
[sgf]
(;FF[4]ST[2]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GM[1]PW[AlphaGo]RE[W+2.50]RU[Chinese]GN[
]DT[2015-10-05]BR[2p]C[Commentary by
Robert Jasiek using theory of the book Positional Judgement 2 -
Dynamics
Best viewed with a local SGF editor that shows arbitrary labels. Note
that there are variations starting with a pass or the
same move as the major sequence's move. Variations contain essential
diagram SGF nodes starting new diagrams even when there
is no new move. Online viewing of this SGF may be insufficient for
seeing all the commentary, markup, SGF diagrams and
variations.]FG[259:]PB[Fan Hui]CP[(to comments)\: Antti Tormanen
1p]AN[Antti Tormanen 1p]PM[2]TM[3600]KM[7.50]
;B[pd]
;W[dd]
;B[pp]
;W[dq]
;B[co]
;W[dl]
;B[dn]
;W[fp]
;B[bq]
;W[jq]
;B[cf]
;W[ch]
;B[fd]
;W[df]
;B[dg]
;W[cg]
;B[dc]
;W[ce]
;B[cc]
;W[hc]
;B[fb]
;W[nc]
;B[qf]
;W[pb]
;B[bf]
;W[be]
;B[ef]
;W[de]
;B[qc]
;W[kc]
;B[qn]
;W[cm]
;B[cr]
;W[mq]
;B[oq]
;W[qm]
;B[pm]
;W[ql]
;B[rn]
;W[pl]
;B[om]
;W[qi]
;B[hq]
;W[hp]
;B[gq]
;W[gp]
;B[iq]
;W[ip]
;B[jr]
;W[kq]
;B[er]
;W[rg]
;B[qg]
;W[rf]
;B[re]
;TR[dl][dn][fp][gp][hp][ip][jq][kq][mq][om]W[qh]L[jo][jp]C[Influence
stone difference for the lower center\: 2 - 8 = -6 in
White's facour. Note that the exchange Black A - White B does not
alter this because afterwards the stone A is too weak to
count as a significant influence stone and the influence stone K3 is
replaced by K4.]
(
;PM[2]FG[259:]
;L[jm]C[As the study shows, this move results in the local territory
count 7 points. Instead Black A results in the local
territory count ca. 3 points. Hence, the black move increases the
local territory count by ca. 4 points in Black's favour.
Since White needs to answer, Black gains, in the purely territorial
judgement, ca. 4 points in sente.]B[kr]
;PM[2]FG[259:]
(
;W[ln]
;B[sf]
;W[rh]
;B[qb]
;W[fe]
;B[el]
;W[ek]
;B[fk]
;W[ej]
;B[fl]
;W[dm]
;B[fj]
(
;W[il]C[There have been comments expressing surprise about this move.
Why? In terms of influence stone difference, it is the
obvious move\: in the lower center, the move creates the influence
stone difference -4 in White's favour and, if instead
Black may make the next play in the lower center, prevents the
influence stone difference 4 in Black's favour.
The local change in influence stone difference between Black or White
making a play in the lower center is 8 influence
stones. This is a large number for a change in influence stone
difference!
Since the move has a great impact on the score, it is not so
surprising that AlphaGo could find the move, although the
program would assess it in terms of expected score rather than in the
- for humans - very much simpler terms of change in
influence stone difference.]
;B[fi]
;W[ei]
;B[fh]
;W[gd]
;B[dh]
;W[ci]
;B[di]
;W[cj]
(
;C[This move is a mistake. Surely, it has a small impact on the local
territories but otherwise the move is played in a
neutral region as a neutral stone. Instead, it should be played as a
valuable influence stone in the valuable (upper) center
region.]B[fn]
;W[em]
;B[hn]
;W[in]
;B[en]
;W[oo]
;B[np]
;W[nn]
;B[po]
;W[hm]
;B[fm]
;W[nl]
(
;B[og]
;W[nr]
;B[or]
;W[nh]
;B[oj]
;W[ol]
;B[oh]
;W[ni]
;B[oi]
;W[ng]
;B[nf]
;W[mf]
;B[ne]
;W[me]
;B[gc]
;W[hb]
;B[bd]
;W[ed]
;B[fc]
;W[ff]
;B[ae]
;W[bg]
;B[af]
;W[eh]
;B[fg]
;W[eg]
;B[ge]
;W[hd]
;B[gf]
;W[ee]
;B[if]
;W[fa]
;B[ga]
;W[gb]
;B[ea]
;W[ec]
;B[eb]
;W[nd]
;B[je]
;W[pe]
;B[oe]
;W[od]
;B[of]
;W[pc]
;B[qd]
;W[jh]
;B[kd]
;W[lc]
;B[nj]
;W[hh]
;B[hg]
;W[mj]
;B[mk]
;W[lj]
;B[nk]
(
;W[lk]
;B[pa]
;W[rm]
;B[mp]
;W[lp]
;B[lr]
;W[lq]
;B[bn]
;W[qq]
;B[rp]
;W[rq]
;B[qp]
;W[ag]
;B[ad]
;W[cp]
;B[bp]
;W[oa]
;B[qa]
;W[dr]
;B[ds]
;W[ob]
;B[ml]
;W[nm]
;B[pn]
;W[hj]
;B[kg]
;W[jg]
;B[kf]
;W[kh]
;B[bl]
;W[bm]
;B[am]
;W[bk]
;B[jc]
;W[jb]
;B[lm]
;W[km]
;B[mn]
;W[mo]
;B[mm]
;W[no]
;B[kl]
;W[jm]
;B[ll]
;W[lg]
;B[jk]
;W[cl]
;B[qj]
;W[rj]
;B[gm]
;W[ho]
;B[al]
;W[ak]
;B[an]
;W[ic]
;B[mr]
;W[nq]
;B[ns]
;W[op]
;B[pq]
;W[jd]
;B[pj]
;W[sg]
;B[ii]
;W[se]
;B[sd]
;W[ih]
;B[ji]
;W[hi]
;B[ie]
;W[ld]
;B[ke]
;W[he]
;B[gg]
;W[eq]
;B[fq]
;W[ep]
;B[cq]
;W[gn]
;B[ki]
;W[li]
;B[ik]
;W[sn]
;B[so]
;W[sm]
;B[dp]
;W[eo]
;B[id]
;W[jc]
;B[do]
;W[fo]
;B[hk]
;W[hl]
;B[cb]
;W[ph]
;B[pg]
;W[qk]
;B[ha]
;W[ia]
;B[fa]
;W[ca]
;B[ba]
;W[dj]
;B[cd]
;W[sf]
;B[gl]
;W[gj]
;B[gk]
;W[ij]
;B[kk]
;W[lh]
;B[ig]
;W[lf]
;B[le]
;W[gh]
;B[kj]
;W[jf]
;B[hf]
;W[jl]
;B[jj]
;W[gi]
;B[pi]
;W[cn]
;B[pk]
;W[ok]
;B[on]
;W[bb]MA[mg][mh][mi]C[White's territory excess in the center is 3
points in the scoring position. White wins by 2.5 points.
Not quite a coincidence. (Note that White sacrificed more points there
to increase territory in the lower center.)]
)
(
;PM[2]FG[259:]
;W[lk]
;TR[hh][jh][mf][ng][nh][ni][mj][lj][lk][ln][in][hm][il]PM[2]C[Result
of the earlier changes in influence stones and neutral
stones\: the marked white stones control the (upper) center so that
White makes a territory excess there.]FG[259:]
)
)
(
;PM[2]FG[259:]
;B[og]
;TR[fh][fi][fj][fk][il][hm][in][ln][nn][nl][pl][qi][qh][og][qg][qf][pd][nc][kc][hc][gd][fe]LB[hn:N][fl:N][fm:N][em:N][en:N]
[fn:N]PM[2]C[Influence stone difference in the upper center
(triangles)\: 8 - 14 = -6 in White's favour. That is, clearly
favourable for White.
The new neutral stones in the lower left center played since move 79
are denoted by N. The game sequence's neutral stone
difference (from move 79 to move 91), that is the difference of new
black neutral stones and new white neutral stones, is 5
- 1 = 4.
Each new excess neutral stone is a great disadvantage. (In the worst
case, already just one excess neutral stone equals a
pass.) The related principle says\: "\[...\] the neutral stone
difference should be zero". (Because neither player wants to
get new inefficient to unvaluable stones.)
White improved the influence stone difference from -1 to -6 (recall
that negative values favour White). The neutral stone
difference 4 (Black's disadvantage) is another view on expressing
White's great advantage during the sequence.
Fan Hui lost the game because he made small premature endgame while
overlooking the very disadvantageous changes of
influence (measured by the change in the influence stone difference)
and neutral stones (measured by the neutral stone
difference). In other words, his positional judgement appreciated
territory while it neglected the dynamic positional
aspects of influence and new neutral stones.]FG[259:]
)
)
(
;PM[2]FG[259:]
;B[ji]
;TR[fh][fi][fj][fk][ji][il][ln][om][pl][qi][qh][qg][qf][pd][nc][kc][hc][gd][fe]PM[2]C[Influence
stone difference in the
upper center\: 9 - 10 = -1 in White's favour. That is, almost
equal.]FG[259:]
)
)
(
;PM[2]FG[259:]
;W[il]
;TR[fj][fk][fl][il][gp][hp][ip][jq][kq][mq][ln][om]PM[2]C[The
influence of unmarked stones is dominated by marked stones.
The influence stone difference in the lower center is 4 - 8 = -4 in
White's favour.
This presumes the life of the black center string. However, currently
it is weak.]FG[259:]
)
(
;PM[2]FG[259:]
;W[gd]C[Assuming some tenuki from the lower center.]
;C[Sample play by Black in the lower center.]B[jl]
;MA[em][en][eo][fo][go][ho][io][jp][kp][ko][kn][jm][im][hm][gm][fm]TR[fj][fk][fl][jl][ln][om]W[tt]C[Result\:
the marked
lower left center has become an almost neutral region. Only the marked
stones remain as significant influence stones on the
outside of the lower center. There, the influence stone difference is
5 - 1 = 4 in Black's favour.
Note that now the black center group is not (so) weak any more. It can
easily become strong.]
)
)
(
;PM[2]FG[259:]
;W[ln]
;TR[dl][dn][fp][gp][hp][ip][jq][kq][mq][ln][om]LB[oq:*][pl:*][qn:*][pp:*][pm:*]PM[2]C[Ignoring
the influence-dominated
stones marked with asterisks, counting the triangled stones as
significant influence stones and judging the lower center,
its influence stone difference is 2 - 9 = -7 in White's favour.
The black move L2 has gained ca. 4 points for Black while White has
gained the 1 extra influence stone M6. This influence
stone faces the wide open center and therefore is valuable. Surely it
is more valuable than an equivalent of 4 extra points
of territory. Hence, by exchanging L2 for M6, Black loses the
equivalence of a few points in sente, regardless of the exact
amount as a territory equivalent.
Note\: during the early opening, a move has the per move value 14
points. (See the book Joseki 2 - Strategy.) If an early
move has 50% territory and 50% influence value, its influence value
expressed as a territory equivalent is 7 points. In view
of the wide open center in the current position, it is reasonable to
assume that the value of one additional influence stone
is between 4 and 7 points if expressed as an estimated territory
equivalent.
One of the reasons why Fan Hui lost the game is having played the
premature endgame L2.]FG[259:]
)
)
(
;SQ[ln][kn][jn][in][hn][gn][go][ho][io][jo][jp][ko][kp][lo][lp][lq][lr][ls][kr][ks][js][ir][is][hr][hs][gr][gs][fs][es][fr]
PM[2]C[Locale in which current territory is counted.]FG[259:]
;C[This move results in the local current territory count 7
points.]B[kr]
;W[ln]
;C[Black has this as a follow-up privilege for a purely territorial
endgame.]B[lq]
;W[lp]
;B[lr]
;MA[es][fs][gr][gs][hr][hs][ir][is][js][ks][kp][jp]W[mp]L[io][lo]C[A
and B is ca. 1/2 point each.
The local territory count is 10 - 3 = 7 points.]
)
(
;PM[2]FG[259:]
;C[In the locale, if now White makes a territorial move near the edge,
we get the cases "Black makes the next local play"
with the current-territory count 5 and "White makes the next local
play" with the count 1.5. On average, the local count is
(5 + 1.5) / 2 = 6.5 / 2 = 3.25 ~= 3 points.]B[jm]
(
;W[kr]
;B[jo]
;W[fq]
;B[fr]
;W[jp]
(
;MA[es][fs][gs][gr][hr][hs][is][lr][ls]C[Case Black makes the next
local play.
The local territory count is 7 - 2 = 5 points.]B[ir]
)
(
;PM[2]FG[259:]
;C[Case White makes the next local play.
As the following study shows, this case results in the approximate
local territory count 1.5 points.]B[tt]
(
;W[hs]
(
;C[Black cannot answer here.]B[is]
;W[gr]
;B[ir]
;W[ks]
;B[fs]
;W[js]
)
(
;PM[2]FG[259:]
;C[Black cannot answer here.]B[ir]
(
;W[gr]C[Fails]
(
;B[fs]
;W[js]
;B[gs]
;W[is]
)
(
;PM[2]FG[259:]
;C[Black can answer here]B[hr]
;W[js]
;B[gs]
)
)
(
;PM[2]FG[259:]
;W[js]
(
;C[Black cannot answer here.]B[is]
;W[gr]
;B[ks]
;W[ls]
;B[fs]
;W[js]C[W success]
)
(
;PM[2]FG[259:]
;C[White endgame success but connection J1 is gote.]B[gr]
)
)
)
)
(
;PM[2]FG[259:]
;W[ir]C[Can be an endgame variation but W will end in gote, so this is
not considered for determining the current
territory.]
)
(
;PM[2]FG[259:]
;W[js]C[White does not threaten to cut and capture by G2. However, for
determining the current territory, we may assume
Black's peaceful reply.]
;B[ir]
;W[hs]
;C[Regardless of the move order, in which this position is reached,
this is the position to be judged when assessing the
current territory in the case "White makes the next local play". For
this purpose, we may not add another white play because
we assume White's sente. Therefore, counting current territory depends
on an endgame-like count with a case analysis of the
local follow-ups.
Black's follow-up results in the local territory count 5 points.
White's follow-up results in the local territory count -2
points. Between the two follow-up cases, there is an excess move
difference of 2 moves. The approximated average local move
value is
(5 - (-2)) / 2 = 7 / 2 = 3.5 points.
Therefore the approximated average local territory count (which can be
calculated by the change from Black's follow-up or
the change from White's follow-up) is 5 - 1 * 3.5 = -2 + 1 * 3.5 = 1.5
points.]B[gr]
(
;W[tt]C[Case Black follow-up]
;MA[es][fs][gs][hs][hr]L[lr]C[Black has 6 points. We might make extra
efforts to assess White's points around A exactly, but
let us simply make the approximate guess that White has ca. 1 point on
average there.
The approximated local territory count is 6 - 1 = 5 points.]B[is]
)
(
;PM[2]FG[259:]
;MA[lr][ls]W[is]L[cs]C[Case White follow-up.
For the sake of simplicity, let us assume that A still is territory.
The local territory count is 0 - 2 = -2 points in White's favour.]
)
)
)
(
;TR[dl][dn][jm][om]PM[2]L[hl]SQ[fp][gp][hp][ip][jp][kq][mq]C[Assessment
of the influence stone difference facing the lower
center.
Optimistic view\: squares denote significant influence stones. The
influence stone difference is 3 - 8 = -5 in White's
favour.
Pessimistic view\: just one additional black stone around A puts the
squared stones in a neutral region without significant
outside influence. There is a great risk of this happening so the
squared stones should not count as significant influence
stones. The influence stone difference is 3 - 1 = 2 in Black's favour.
Indeed, the risk of too many white wall stones ending as neutral
stones is so great that White should not have answered K7
at L2. L2 has only been assumed for the purely territorial judgement
anyway.]FG[259:]
)
)
(
;PM[2]FG[259:]
;W[jo]C[From an influence point of view, it is essential to maintain
the thickness of the white group.]
;TR[dl][dn][jm][om][fp][gp][hp][ip][jo][jq][kq][mq]PM[2]C[Influence
stone difference for the lower center\: 3 - 9 = -6 in
White's favour.]FG[259:]
)
)
)
[/sgf]
[sgf]
(;FF[4]ST[2]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GM[1]PW[Fan
Hui]RE[B+R]RU[Chinese]GN[ ]DT[2015-10-06]C[Commentary by Robert
Jasiek using theory of the book Positional Judgement 2 - Dynamics
Best viewed with a local SGF editor to see all the markup in SGF
diagrams of variations.]WR[2p]FG[259:]PB[AlphaGo]PM[2]TM
[3600]KM[7.5]OT[1 moves / 30 sec]
;B[pd]
;W[dd]
;B[pp]
;W[dq]
;B[do]
;W[co]
;B[dp]
;W[cp]
;B[eq]
;W[cn]
;B[dn]
;W[dm]
;B[cq]
;W[dr]
;B[bq]
;W[cr]
;B[cm]
;W[cl]
;B[br]
;W[er]
;B[bm]
;W[bn]
;B[bl]
;W[bp]
;B[fq]
;W[dl]
;B[ck]
;W[dk]
;B[cj]
;W[fr]
;B[ao]
;W[ap]
;B[fo]
;W[dj]
;B[ch]
;W[di]
;B[bg]
;W[am]
;B[bk]
;W[gn]
;B[hq]
;W[ip]
;B[iq]
;W[jp]
;B[kq]
;W[kp]
;B[lq]
;W[jq]
;B[jr]
;W[ir]
;B[hr]
;W[kr]
;B[is]
;W[mp]
;B[lp]
;W[lo]
;B[mo]
;W[np]
;B[ln]
;W[ko]
;B[lr]
;W[po]
;B[no]
;W[op]
;B[oo]
;W[pq]
;B[qp]
;W[qq]
;B[ro]
;W[pn]
;B[rq]
;W[mn]
;B[nn]
;W[mm]
;B[nm]
;W[nl]
;B[ml]
;W[lm]
;B[ol]
;W[nk]
;B[om]
;W[rr]
;B[oq]
;W[ok]
;B[nq]
;W[pk]
;TR[di][dj][dk][dl][dm][gn][ip][jp][kp][ko][lo][mn][mm][lm][nl][nk][ok][pk]L[eg]C[Probably
this move by AlphaGo is a
mistake.
The influence stone difference (the significant influence stones
there are marked) in the center is 0 - 18 = -18 in White's
favour. In other words, White has a sphere of dominating influence.
Black should play at A, which reduces the white center influence,
helps the somewhat weak black group on the left side and
attacks D16. For comparison, F17 does the latter but mostly neglects
the other two aspects.
There sphere of dominating influence has the great local potential of
being expanded and transformed to a solidified moyo by
White A. Therefore, A is a valuable intersection for either player's
move.
Besides, this principle applies to possibly reducing or invading the
sphere of dominating influence now\: "Reduce, but do
not invade, a moyo with a large influence stone difference and without
space for easy life of an invading group." That is,
the reduction A is possible, but an immediate invasion would be a
mistake.
AlphaGo has learned knowledge from next move shape samples, so we can
assume that both A and F17 must have been among the
next move candidates. On the other hand, there would be only a few
samples supporting the exact shape of the boundary of the
white sphere of dominating influence. Now, it is speculation whether
AlphaGo suffered from insufficient samples for such
spheres of influence or whether subsequent Monte Carlo playouts hinted
at a preference for F17 over A. The latter is not
inconceivable because there is aji in the white walls and so a large
fraction of sample playouts might hit at successful
later exploitations of aji. In fact, this is exactly how the actual
game ended.]B[fc]
;PM[2]FG[259:]
;W[cf]
;B[dh]
;W[eh]
;B[eg]
;W[fh]
;B[df]
;W[bf]
;B[fg]
;W[gh]
;B[ce]
;W[de]
;B[be]
;W[ef]
;B[dg]
;W[bd]
;B[af]
;W[cd]
;B[cg]
;W[gd]
;B[hc]
;W[gc]
;B[gb]
;W[fd]
;B[eb]
;W[ec]
;B[fb]
;W[db]
;B[jc]
;W[ke]
;B[ld]
;W[le]
;B[md]
;W[nf]
;B[pf]
;W[pc]
;B[qc]
;W[od]
;B[oe]
;W[qd]
;B[pe]
;W[qb]
;B[rc]
;W[rb]
;B[oc]
;W[pb]
;B[rd]
;W[ob]
;B[mb]
;W[sb]
(
;B[ho]
;W[jm]
;B[oh]
;W[nh]
;B[ni]
;W[oi]
;B[pi]
;W[oj]
;B[hn]
;W[hl]
;B[hg]
;W[gg]
;B[ie]
;W[if]
;B[hf]
;W[ig]
;B[hh]
;W[hd]
;B[id]
;W[he]
;B[jf]
;W[ih]
;B[gf]
;W[ff]
;B[hi]
;W[kg]
;B[fj]
;W[ei]
;B[ii]
;W[kh]
;B[gl]
;W[gk]
;B[hk]
;W[ik]
;B[fk]
;W[hj]
;B[gi]
;W[en]
;B[fl]
;W[fn]
;B[eo]
;W[em]
;B[hm]
;W[gm]
;B[il]
;W[hk]
;B[go]
)
(
;MA[ng][nh][ni][nj][ho][em][fm][jf][ig][hg]PM[2]L[kn][km][ll][ei]C[The
big white moyo has major weaknesses at the crossed
gaps and the labelled minor weaknesses.
Reduction of a big moyo uses a method (see the book), of which I
mention some important aspects\: First, identify the major
weaknesses of the moyo. Combine imagination of reduction, positional
judgement and reduction strategies exploiting the major
weaknesses of the moyo.
And, from a human player's perspective, this is exactly what AlphaGo
did with moves 137 - 149 and afterwards, then combined
with reading and re-evaluation of the then more important weaknesses D
and the gap around E7, which initially is the least
important of the major weaknesses.
White 138 also decreased the degree of the minor weaknesses ABC. So in
the end, we can say that AlphaGo exploited every aji
to reduce the white moyo. In particular, the program started with the
most important weaknesses by devising a timing
enabling this - instead of failing to exploit some major weakness at
all.
Besides, we see AlphaGo's implicit application of these related
principles\:
- Perform a series of sente-like \[i.e., double sente or sente\]
actions. Optionally, conclude with a gote-like \[i.e.,
reverse sente or gote\] action.
- Do not let your earlier actions inhibit your planned later actions.
- \[A principle of first strengthening adjacent groups need not be
applied because Black's groups already are strong.\]
- \[...\] the attacker of a moyo must avoid suffering from severe
collateral damage \[...\] (Compare move 155.)
Hence AlphaGo's skill in reducing a big moyo agrees to the book's
theory of how to do so. However, the program achieves this
implicitly by relying on shape knowlede, neural nets and random sample
continuations while the book provides human-
applicable theory not needing database knowledge and excessive
calculation power.
Program and human play are related, however, when it comes to reading
of the created capturing race. Reading tactics is
necessary for both, although AlphaGo reads implicitly by means of
sufficiently good sampling while a human player might
prefer to read explicitly.]FG[259:]
)
)
[/sgf]
[sgf]
(;ST[2]GM[1]FF[4]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GN[
]PM[2]KM[6.50]RE[B+Resign]PB[Alpha Go]RU[Japanese]FG[259:]PW[Fan Hui]
C[Commentary by Robert Jasiek using theory of the book Positional
Judgement 2 - Dynamics
Best viewed with a local SGF editor that shows arbitrary labels. Note
that there are variations starting with a pass or the
same move as the major sequence's move. Variations contain essential
diagram SGF nodes starting new diagrams even when there
is no new move. Online viewing of this SGF may be insufficient for
seeing all the commentary, markup, SGF diagrams and
variations.]
;B[dp]
;W[pp]
;B[dd]
;W[pc]
;B[pe]
;W[qe]
;B[qf]
;W[qd]
;B[pg]
;W[nc]
;B[qk]
;W[qm]
;B[ok]
;W[oq]
;B[dj]
;W[fq]
;B[dn]
;C[How can White expect to win if he leaves at least one weak group
(F3 or F17), which Black can attack favourably? This is
not a shinogi position but a position of voluntarily giving the
opponent an unnecessary advantage.
Ok, to be fair, Fan Hui could not know yet just how strong AlphaGo
would be in profiting from attacking a weak group. After
having seen all games of the match, we know it better.]W[fc]
;B[df]
;W[ip]
;B[ec]
(
;W[fd]
;B[id]
;W[he]
;B[ie]
;W[hg]
;B[hf]
;W[gf]
;B[if]
;W[gg]
;B[jh]
;C[Although the fight from move 32 to 98 looks complicated, it can be
understood as a series of one-sided actions by Black
or White to apply this principle\:
"The positional judgement after each action determines the judgement
before it."
Well, this is the theory, but the fight is long and complicated.
Predicting it to its quiet end and working judgements
backwards from the judgement after move 98 to the judgement of move 32
is too hard. Fan Hui may as well have played his move
for this very reason\: to start an unpredictable fight. The related
dry principle confirms this\:
"If, within the scope of reading, both players prefer a complex fight
during their strategic choices, the positional
judgement assesses the initial position simply as having an
unpredictable winner."
The fight starts with the one-move action of White's invasion and
installation of a second weak group.
However, Fan Hui is being unreasonable to start such an unpredictable
fight here because 1) both his groups are weak and 2)
each nearby black group is strong. It is common knowledge that
starting a fight under such circumstances should only be a
means of cheating when clearly behind. Hm, but considering that all
black stones work well while White has had a weak group
of 6 stones in a neutral region (where neither player can make much
territory), White already is behind!
It is the reader's exercise to decompose the following fight into its
one-sided actions.]W[dh]
;B[eg]
;W[ih]
;B[ji]
;W[eh]
;B[fh]
;W[fi]
;B[fg]
;W[ej]
;B[gi]
;W[gj]
;B[hj]
;W[dk]
;B[gk]
;W[fj]
;B[hi]
;W[ib]
;B[cj]
;W[cg]
;B[ck]
;W[be]
;B[cc]
;W[bc]
;B[bb]
;W[cd]
;B[ac]
;W[bd]
;B[db]
;W[de]
;B[ee]
;W[cf]
;B[ed]
;W[jc]
;B[hc]
;W[gb]
;B[hb]
;W[ha]
;B[fb]
;W[gc]
;B[dl]
;W[bi]
;B[ci]
;W[bh]
;B[qo]
;W[pl]
;B[pk]
;W[qp]
;B[fa]
;W[ga]
;B[lc]C[It seems that AlphaGo could "read" to move 98.]
;W[lb]
;B[kb]
;W[kc]
;B[mb]
;W[ld]
;B[mc]
;W[md]
;B[nb]
;W[ke]
;B[oc]
;W[od]
;B[nd]
;W[ne]
;B[pb]
;W[nc]
;B[ob]
;C[With the next move, a "complex fight" (neither a "one-sided fight"
nor a simple "exchange") starts and cannot be easily
decomposed into a series of one-sided fights. Therefore, in the now
created fighting region on the left lower side, every
relevant dynamic strategic aspect must be evaluated and re-evaluated
after every move to choose the best dynamic change for
the moving player. This should be done in a context of a, here
omitted, static positional judgement for the rest of the
board excluding the fighting region.
Initially, the left lower white group is alive and stable, White
threatens to enclose the right lower side and either player
has the option of making territory in the lower left corner.]W[pd]
;B[kq]C[The move destabilises the white group, creates an unstable
black group and prevents White from enclosing territory
on the right lower side.
For the sake of simplicity, let us assume that this invasion is more
attractive than the reduction L5.
Alternatively, Black could take the lower left corner with, e.g., D3,
which White might answer at L4.
Is L3 or D3 better? White answers either move, again stabilising or
increasing the stability of the white group. L3 means
that White can make a few excess points, D3 - L4 means that the
players make similar amounts of new points. There is,
however, a major difference between D3 and L3\: Black L3 or otherwise
White L4 greatly affect the potential development in
the lower center, which is wide and adjacent to strong black and white
stones so that either player threatens to build a
moyo there or prevent the opponent's moyo. Hence, L3 is more valuable
than D3.]
;C[White stabilises the white group and destabilises the black
stone.]W[jq]
;B[kp]C[Black stabilises his and destabilises the white group.]
(
;W[dr]C[White stabilises F3 and prepares taking the corner territory
but leaves behind the weakness A, which White B might
eliminate. As the variation suggests, D2 is better than B. A similar
argument could be made against C.]L[hq][kr][in]
;FG[259:]PM[2]
;B[hq]C[Black destabilises all white stones and stabilises his other
string by creating the option of connecting at A.]L[jr]
;FG[259:]PM[2]
;C[and so on]W[ir]
;B[gq]
;W[fp]
;B[fr]
;W[er]
;B[ho]
;W[gr]
;B[hr]
;W[fs]
;B[go]
;C[Result\: White has taken the corner and reduced the black territory
on the left side a bit. Black has stabilised his
strings on the lower side, claims territory in the lower left center,
threatens to expand the moyo rightwards, has reduced
White's center potential and prevents white territory on the right
lower side.
Territorially, probably this is fair but Black has gained additional
center potential while White has lost quite some center
potential. Hence, considering all, Black has gained from the fight\:
the center potential has been shifted greatly from
"shared" to "dominated by Black".]W[cq]
;B[rp]
;W[rq]
;B[ig]
;W[qb]
;B[rb]
;W[qa]
;B[rc]
;W[qc]
;B[la]
;W[on]
;B[rf]
;W[sb]
;B[mg]
;W[of]
;B[og]
;W[mf]
;B[nm]
;W[lg]
;B[mh]
;W[lo]
;B[ko]
;W[ln]
;B[kn]
;W[lm]
;B[km]
;W[ll]
;B[om]
;W[pn]
;B[mp]
;W[mq]
;B[np]
;W[nn]
;B[nq]
;W[kr]
;B[mr]
;W[hs]
;B[or]
;W[pr]
;B[op]
;W[pq]
;B[po]
;W[oo]
;B[ro]
;W[sq]
;B[qr]
;W[ps]
;B[rr]
;W[sr]
;B[rm]
;W[rl]
;B[ql]
)
(
;FG[259:]PM[2]
;W[kr]
;B[lr]
;W[jr]
;B[rp]
;W[rq]
;B[er]
;W[jn]
;B[no]C[Black took the lower left corner and now threatens to defend A
or B. White would have a hard time catching up with
Black's excess of new territory gained during the fight.]L[kp][qo]
)
)
(
;FG[259:]PM[2]
;W[eb]
;B[db]
;C[This variation can lead to a long endgame, something Fan Hui tried
and lost in game 1. Besides, the low and narrow
formation on the upper side is unattractive.]W[ic]
)
)
[/sgf]
[sgf]
(;ST[2]GM[1]FF[4]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GN[
]PM[2]KM[6.50]RE[W+Resign]PB[Fan Hui]RU[Japanese]FG[259:]PW[Alpha Go]
C[Commentary by Robert Jasiek using theory of the book Positional
Judgement 2 - Dynamics
Best viewed with a local SGF editor.]
;B[dp]
;W[pp]
;B[cd]
;W[pd]
;B[nc]
;W[qf]
;B[ic]
;W[cj]
;B[ci]
;W[di]
;B[ch]
;W[dj]
;B[dh]
;W[cn]
;B[ei]
;W[fq]
;B[eo]
;W[cq]
;B[cl]
;W[dl]
;B[cp]
;W[dq]
;B[dm]
;W[cm]
;B[el]
;W[dk]
;B[bl]
;W[bk]
;B[dn]
;W[bm]
;B[gp]
;W[gq]
;B[hp]
;W[iq]
;B[pb]
;W[qc]
;B[ld]
;W[ec]
;B[dc]
;W[db]
;B[ed]
;W[cc]
;B[dd]
;W[bb]
;B[eb]
;W[fc]
;B[fb]
;W[bd]
;B[da]
;W[cb]
;B[be]
;W[ac]
;B[ek]
;W[bi]
;B[bh]
;W[fd]
;B[gb]
;W[id]
;B[jd]
;W[hc]
;B[hb]
;W[jc]
;B[ib]
;W[md]
;B[je]
;W[mc]
;B[me]
;W[nd]
;B[lb]
;W[lc]
;B[kc]
;W[le]
;B[ai]
;W[bo]
;B[lf]
;W[kd]
;B[jb]
;W[ej]
;B[fj]
;W[fi]
;B[bj]
;W[eh]
;B[ak]
;W[gj]
;B[fk]
;W[ef]
;B[fe]
;W[gk]
;B[ck]
;W[ei]
;B[fm]
;W[hh]
;B[ji]C[The territorial deire value of A is ca. 22 points. Besides, A
greatly strengthens the black group in the lower
center. If Black's move is gote because the white group lives, it is
hard to justify the move as more valuable than A.
Maybe Fan Hui expected his move to be sente. AlphaGo's "reading"
during the subsequent fight is impressive\: the program
defends life.]L[bp]
;FG[259:]PM[2]
;W[bp]
;B[ij]
;W[ff]
;B[ge]
;W[hl]
;B[jl]
;W[im]
;B[jo]
;C[Except for intervening forcing sequences before and afterwards
during the fight, now its phase of sente actions by Black
to attack the life of the white group ends.]W[jm]
;B[km]C[This move starts the "reward action", with which Black rewards
himself at the end of the fight.
It is a general nature of one-sided fights that the attacker plays a
series of sente actions and optionally concludes with a
reward action.
]
;W[kn]
;B[jn]
;W[kl]
;B[lm]
;W[jk]
;B[ll]
;W[kk]
;B[ik]
;W[il]
;B[lk]
;W[kj]
;B[mi]
;W[kh]
;B[jh]
;W[lj]
;B[mj]
;W[lh]
;B[ke]
;W[mg]
;B[lo]C[This concludes the reward action. Here, as usual, the reward
action is a gote action. Black's reward is the creation
of his group of thickness on the outside. Still it is unconceivable
that Black can compensate his loss at B4.
The position before the fight (before move 93) inherits the judgement
of the position after the fight (after move 123).
Instead, Fan Hui seems to have expected killing the group. In this
case, the judgement before the fight would be "win
because I kill you".]
;W[qm]
;B[ne]
;W[oe]
;B[ld]
;W[fp]
;B[hn]
;W[fo]
;B[gn]
;W[og]
;B[mb]
;W[od]
;B[qq]
;W[qp]
;B[pq]
;W[oq]
;B[or]
;W[nr]
;B[rp]
;W[ro]
;B[op]
;W[nq]
;B[rr]
;W[sp]
;B[sq]
;W[rq]
;B[gl]
;W[hk]
;B[rp]
;W[pr]
;B[qr]
;W[rq]
;B[nh]
;W[ng]
;B[rp]
;W[mn]
;B[ln]
;W[rq]
;B[gi]
;W[hi]
;B[rp]
;W[ce]
;B[so]
;W[ee]
;B[oo]
;W[pj]
;B[lr]
;W[pn]
;B[rn]
;W[on]
;B[mo]
;W[nk]
;B[mk]
;W[jr]
;B[qk]
;W[pk]
;B[qj]
;W[qi]
;B[rl]
;W[ri]
;B[kr]
;W[kq]
;B[lq]
;W[kp]
;B[lp]
;W[rm]
;B[sm]
;W[ql]
;B[po]
;W[rk]
;B[qb]
;W[rb]
;B[ra]
;W[rc]
;B[oc]
;W[hq]
;B[gc]
;W[de]
;B[hf]
;W[gh]
;B[bf]
;W[ig]
;B[jg]
;W[if]
;B[ie]
;W[ep]
;B[do]
;W[fn]
;B[ip]
;W[jq]
;B[ko]
;C[Black resigns.]W[no]
)
[/sgf]
games. Save each file within the [sgf] tag brackets (without them),
then view offline.
[sgf]
(;FF[4]ST[2]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GM[1]PW[AlphaGo]RE[W+2.50]RU[Chinese]GN[
]DT[2015-10-05]BR[2p]C[Commentary by
Robert Jasiek using theory of the book Positional Judgement 2 -
Dynamics
Best viewed with a local SGF editor that shows arbitrary labels. Note
that there are variations starting with a pass or the
same move as the major sequence's move. Variations contain essential
diagram SGF nodes starting new diagrams even when there
is no new move. Online viewing of this SGF may be insufficient for
seeing all the commentary, markup, SGF diagrams and
variations.]FG[259:]PB[Fan Hui]CP[(to comments)\: Antti Tormanen
1p]AN[Antti Tormanen 1p]PM[2]TM[3600]KM[7.50]
;B[pd]
;W[dd]
;B[pp]
;W[dq]
;B[co]
;W[dl]
;B[dn]
;W[fp]
;B[bq]
;W[jq]
;B[cf]
;W[ch]
;B[fd]
;W[df]
;B[dg]
;W[cg]
;B[dc]
;W[ce]
;B[cc]
;W[hc]
;B[fb]
;W[nc]
;B[qf]
;W[pb]
;B[bf]
;W[be]
;B[ef]
;W[de]
;B[qc]
;W[kc]
;B[qn]
;W[cm]
;B[cr]
;W[mq]
;B[oq]
;W[qm]
;B[pm]
;W[ql]
;B[rn]
;W[pl]
;B[om]
;W[qi]
;B[hq]
;W[hp]
;B[gq]
;W[gp]
;B[iq]
;W[ip]
;B[jr]
;W[kq]
;B[er]
;W[rg]
;B[qg]
;W[rf]
;B[re]
;TR[dl][dn][fp][gp][hp][ip][jq][kq][mq][om]W[qh]L[jo][jp]C[Influence
stone difference for the lower center\: 2 - 8 = -6 in
White's facour. Note that the exchange Black A - White B does not
alter this because afterwards the stone A is too weak to
count as a significant influence stone and the influence stone K3 is
replaced by K4.]
(
;PM[2]FG[259:]
;L[jm]C[As the study shows, this move results in the local territory
count 7 points. Instead Black A results in the local
territory count ca. 3 points. Hence, the black move increases the
local territory count by ca. 4 points in Black's favour.
Since White needs to answer, Black gains, in the purely territorial
judgement, ca. 4 points in sente.]B[kr]
;PM[2]FG[259:]
(
;W[ln]
;B[sf]
;W[rh]
;B[qb]
;W[fe]
;B[el]
;W[ek]
;B[fk]
;W[ej]
;B[fl]
;W[dm]
;B[fj]
(
;W[il]C[There have been comments expressing surprise about this move.
Why? In terms of influence stone difference, it is the
obvious move\: in the lower center, the move creates the influence
stone difference -4 in White's favour and, if instead
Black may make the next play in the lower center, prevents the
influence stone difference 4 in Black's favour.
The local change in influence stone difference between Black or White
making a play in the lower center is 8 influence
stones. This is a large number for a change in influence stone
difference!
Since the move has a great impact on the score, it is not so
surprising that AlphaGo could find the move, although the
program would assess it in terms of expected score rather than in the
- for humans - very much simpler terms of change in
influence stone difference.]
;B[fi]
;W[ei]
;B[fh]
;W[gd]
;B[dh]
;W[ci]
;B[di]
;W[cj]
(
;C[This move is a mistake. Surely, it has a small impact on the local
territories but otherwise the move is played in a
neutral region as a neutral stone. Instead, it should be played as a
valuable influence stone in the valuable (upper) center
region.]B[fn]
;W[em]
;B[hn]
;W[in]
;B[en]
;W[oo]
;B[np]
;W[nn]
;B[po]
;W[hm]
;B[fm]
;W[nl]
(
;B[og]
;W[nr]
;B[or]
;W[nh]
;B[oj]
;W[ol]
;B[oh]
;W[ni]
;B[oi]
;W[ng]
;B[nf]
;W[mf]
;B[ne]
;W[me]
;B[gc]
;W[hb]
;B[bd]
;W[ed]
;B[fc]
;W[ff]
;B[ae]
;W[bg]
;B[af]
;W[eh]
;B[fg]
;W[eg]
;B[ge]
;W[hd]
;B[gf]
;W[ee]
;B[if]
;W[fa]
;B[ga]
;W[gb]
;B[ea]
;W[ec]
;B[eb]
;W[nd]
;B[je]
;W[pe]
;B[oe]
;W[od]
;B[of]
;W[pc]
;B[qd]
;W[jh]
;B[kd]
;W[lc]
;B[nj]
;W[hh]
;B[hg]
;W[mj]
;B[mk]
;W[lj]
;B[nk]
(
;W[lk]
;B[pa]
;W[rm]
;B[mp]
;W[lp]
;B[lr]
;W[lq]
;B[bn]
;W[qq]
;B[rp]
;W[rq]
;B[qp]
;W[ag]
;B[ad]
;W[cp]
;B[bp]
;W[oa]
;B[qa]
;W[dr]
;B[ds]
;W[ob]
;B[ml]
;W[nm]
;B[pn]
;W[hj]
;B[kg]
;W[jg]
;B[kf]
;W[kh]
;B[bl]
;W[bm]
;B[am]
;W[bk]
;B[jc]
;W[jb]
;B[lm]
;W[km]
;B[mn]
;W[mo]
;B[mm]
;W[no]
;B[kl]
;W[jm]
;B[ll]
;W[lg]
;B[jk]
;W[cl]
;B[qj]
;W[rj]
;B[gm]
;W[ho]
;B[al]
;W[ak]
;B[an]
;W[ic]
;B[mr]
;W[nq]
;B[ns]
;W[op]
;B[pq]
;W[jd]
;B[pj]
;W[sg]
;B[ii]
;W[se]
;B[sd]
;W[ih]
;B[ji]
;W[hi]
;B[ie]
;W[ld]
;B[ke]
;W[he]
;B[gg]
;W[eq]
;B[fq]
;W[ep]
;B[cq]
;W[gn]
;B[ki]
;W[li]
;B[ik]
;W[sn]
;B[so]
;W[sm]
;B[dp]
;W[eo]
;B[id]
;W[jc]
;B[do]
;W[fo]
;B[hk]
;W[hl]
;B[cb]
;W[ph]
;B[pg]
;W[qk]
;B[ha]
;W[ia]
;B[fa]
;W[ca]
;B[ba]
;W[dj]
;B[cd]
;W[sf]
;B[gl]
;W[gj]
;B[gk]
;W[ij]
;B[kk]
;W[lh]
;B[ig]
;W[lf]
;B[le]
;W[gh]
;B[kj]
;W[jf]
;B[hf]
;W[jl]
;B[jj]
;W[gi]
;B[pi]
;W[cn]
;B[pk]
;W[ok]
;B[on]
;W[bb]MA[mg][mh][mi]C[White's territory excess in the center is 3
points in the scoring position. White wins by 2.5 points.
Not quite a coincidence. (Note that White sacrificed more points there
to increase territory in the lower center.)]
)
(
;PM[2]FG[259:]
;W[lk]
;TR[hh][jh][mf][ng][nh][ni][mj][lj][lk][ln][in][hm][il]PM[2]C[Result
of the earlier changes in influence stones and neutral
stones\: the marked white stones control the (upper) center so that
White makes a territory excess there.]FG[259:]
)
)
(
;PM[2]FG[259:]
;B[og]
;TR[fh][fi][fj][fk][il][hm][in][ln][nn][nl][pl][qi][qh][og][qg][qf][pd][nc][kc][hc][gd][fe]LB[hn:N][fl:N][fm:N][em:N][en:N]
[fn:N]PM[2]C[Influence stone difference in the upper center
(triangles)\: 8 - 14 = -6 in White's favour. That is, clearly
favourable for White.
The new neutral stones in the lower left center played since move 79
are denoted by N. The game sequence's neutral stone
difference (from move 79 to move 91), that is the difference of new
black neutral stones and new white neutral stones, is 5
- 1 = 4.
Each new excess neutral stone is a great disadvantage. (In the worst
case, already just one excess neutral stone equals a
pass.) The related principle says\: "\[...\] the neutral stone
difference should be zero". (Because neither player wants to
get new inefficient to unvaluable stones.)
White improved the influence stone difference from -1 to -6 (recall
that negative values favour White). The neutral stone
difference 4 (Black's disadvantage) is another view on expressing
White's great advantage during the sequence.
Fan Hui lost the game because he made small premature endgame while
overlooking the very disadvantageous changes of
influence (measured by the change in the influence stone difference)
and neutral stones (measured by the neutral stone
difference). In other words, his positional judgement appreciated
territory while it neglected the dynamic positional
aspects of influence and new neutral stones.]FG[259:]
)
)
(
;PM[2]FG[259:]
;B[ji]
;TR[fh][fi][fj][fk][ji][il][ln][om][pl][qi][qh][qg][qf][pd][nc][kc][hc][gd][fe]PM[2]C[Influence
stone difference in the
upper center\: 9 - 10 = -1 in White's favour. That is, almost
equal.]FG[259:]
)
)
(
;PM[2]FG[259:]
;W[il]
;TR[fj][fk][fl][il][gp][hp][ip][jq][kq][mq][ln][om]PM[2]C[The
influence of unmarked stones is dominated by marked stones.
The influence stone difference in the lower center is 4 - 8 = -4 in
White's favour.
This presumes the life of the black center string. However, currently
it is weak.]FG[259:]
)
(
;PM[2]FG[259:]
;W[gd]C[Assuming some tenuki from the lower center.]
;C[Sample play by Black in the lower center.]B[jl]
;MA[em][en][eo][fo][go][ho][io][jp][kp][ko][kn][jm][im][hm][gm][fm]TR[fj][fk][fl][jl][ln][om]W[tt]C[Result\:
the marked
lower left center has become an almost neutral region. Only the marked
stones remain as significant influence stones on the
outside of the lower center. There, the influence stone difference is
5 - 1 = 4 in Black's favour.
Note that now the black center group is not (so) weak any more. It can
easily become strong.]
)
)
(
;PM[2]FG[259:]
;W[ln]
;TR[dl][dn][fp][gp][hp][ip][jq][kq][mq][ln][om]LB[oq:*][pl:*][qn:*][pp:*][pm:*]PM[2]C[Ignoring
the influence-dominated
stones marked with asterisks, counting the triangled stones as
significant influence stones and judging the lower center,
its influence stone difference is 2 - 9 = -7 in White's favour.
The black move L2 has gained ca. 4 points for Black while White has
gained the 1 extra influence stone M6. This influence
stone faces the wide open center and therefore is valuable. Surely it
is more valuable than an equivalent of 4 extra points
of territory. Hence, by exchanging L2 for M6, Black loses the
equivalence of a few points in sente, regardless of the exact
amount as a territory equivalent.
Note\: during the early opening, a move has the per move value 14
points. (See the book Joseki 2 - Strategy.) If an early
move has 50% territory and 50% influence value, its influence value
expressed as a territory equivalent is 7 points. In view
of the wide open center in the current position, it is reasonable to
assume that the value of one additional influence stone
is between 4 and 7 points if expressed as an estimated territory
equivalent.
One of the reasons why Fan Hui lost the game is having played the
premature endgame L2.]FG[259:]
)
)
(
;SQ[ln][kn][jn][in][hn][gn][go][ho][io][jo][jp][ko][kp][lo][lp][lq][lr][ls][kr][ks][js][ir][is][hr][hs][gr][gs][fs][es][fr]
PM[2]C[Locale in which current territory is counted.]FG[259:]
;C[This move results in the local current territory count 7
points.]B[kr]
;W[ln]
;C[Black has this as a follow-up privilege for a purely territorial
endgame.]B[lq]
;W[lp]
;B[lr]
;MA[es][fs][gr][gs][hr][hs][ir][is][js][ks][kp][jp]W[mp]L[io][lo]C[A
and B is ca. 1/2 point each.
The local territory count is 10 - 3 = 7 points.]
)
(
;PM[2]FG[259:]
;C[In the locale, if now White makes a territorial move near the edge,
we get the cases "Black makes the next local play"
with the current-territory count 5 and "White makes the next local
play" with the count 1.5. On average, the local count is
(5 + 1.5) / 2 = 6.5 / 2 = 3.25 ~= 3 points.]B[jm]
(
;W[kr]
;B[jo]
;W[fq]
;B[fr]
;W[jp]
(
;MA[es][fs][gs][gr][hr][hs][is][lr][ls]C[Case Black makes the next
local play.
The local territory count is 7 - 2 = 5 points.]B[ir]
)
(
;PM[2]FG[259:]
;C[Case White makes the next local play.
As the following study shows, this case results in the approximate
local territory count 1.5 points.]B[tt]
(
;W[hs]
(
;C[Black cannot answer here.]B[is]
;W[gr]
;B[ir]
;W[ks]
;B[fs]
;W[js]
)
(
;PM[2]FG[259:]
;C[Black cannot answer here.]B[ir]
(
;W[gr]C[Fails]
(
;B[fs]
;W[js]
;B[gs]
;W[is]
)
(
;PM[2]FG[259:]
;C[Black can answer here]B[hr]
;W[js]
;B[gs]
)
)
(
;PM[2]FG[259:]
;W[js]
(
;C[Black cannot answer here.]B[is]
;W[gr]
;B[ks]
;W[ls]
;B[fs]
;W[js]C[W success]
)
(
;PM[2]FG[259:]
;C[White endgame success but connection J1 is gote.]B[gr]
)
)
)
)
(
;PM[2]FG[259:]
;W[ir]C[Can be an endgame variation but W will end in gote, so this is
not considered for determining the current
territory.]
)
(
;PM[2]FG[259:]
;W[js]C[White does not threaten to cut and capture by G2. However, for
determining the current territory, we may assume
Black's peaceful reply.]
;B[ir]
;W[hs]
;C[Regardless of the move order, in which this position is reached,
this is the position to be judged when assessing the
current territory in the case "White makes the next local play". For
this purpose, we may not add another white play because
we assume White's sente. Therefore, counting current territory depends
on an endgame-like count with a case analysis of the
local follow-ups.
Black's follow-up results in the local territory count 5 points.
White's follow-up results in the local territory count -2
points. Between the two follow-up cases, there is an excess move
difference of 2 moves. The approximated average local move
value is
(5 - (-2)) / 2 = 7 / 2 = 3.5 points.
Therefore the approximated average local territory count (which can be
calculated by the change from Black's follow-up or
the change from White's follow-up) is 5 - 1 * 3.5 = -2 + 1 * 3.5 = 1.5
points.]B[gr]
(
;W[tt]C[Case Black follow-up]
;MA[es][fs][gs][hs][hr]L[lr]C[Black has 6 points. We might make extra
efforts to assess White's points around A exactly, but
let us simply make the approximate guess that White has ca. 1 point on
average there.
The approximated local territory count is 6 - 1 = 5 points.]B[is]
)
(
;PM[2]FG[259:]
;MA[lr][ls]W[is]L[cs]C[Case White follow-up.
For the sake of simplicity, let us assume that A still is territory.
The local territory count is 0 - 2 = -2 points in White's favour.]
)
)
)
(
;TR[dl][dn][jm][om]PM[2]L[hl]SQ[fp][gp][hp][ip][jp][kq][mq]C[Assessment
of the influence stone difference facing the lower
center.
Optimistic view\: squares denote significant influence stones. The
influence stone difference is 3 - 8 = -5 in White's
favour.
Pessimistic view\: just one additional black stone around A puts the
squared stones in a neutral region without significant
outside influence. There is a great risk of this happening so the
squared stones should not count as significant influence
stones. The influence stone difference is 3 - 1 = 2 in Black's favour.
Indeed, the risk of too many white wall stones ending as neutral
stones is so great that White should not have answered K7
at L2. L2 has only been assumed for the purely territorial judgement
anyway.]FG[259:]
)
)
(
;PM[2]FG[259:]
;W[jo]C[From an influence point of view, it is essential to maintain
the thickness of the white group.]
;TR[dl][dn][jm][om][fp][gp][hp][ip][jo][jq][kq][mq]PM[2]C[Influence
stone difference for the lower center\: 3 - 9 = -6 in
White's favour.]FG[259:]
)
)
)
[/sgf]
[sgf]
(;FF[4]ST[2]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GM[1]PW[Fan
Hui]RE[B+R]RU[Chinese]GN[ ]DT[2015-10-06]C[Commentary by Robert
Jasiek using theory of the book Positional Judgement 2 - Dynamics
Best viewed with a local SGF editor to see all the markup in SGF
diagrams of variations.]WR[2p]FG[259:]PB[AlphaGo]PM[2]TM
[3600]KM[7.5]OT[1 moves / 30 sec]
;B[pd]
;W[dd]
;B[pp]
;W[dq]
;B[do]
;W[co]
;B[dp]
;W[cp]
;B[eq]
;W[cn]
;B[dn]
;W[dm]
;B[cq]
;W[dr]
;B[bq]
;W[cr]
;B[cm]
;W[cl]
;B[br]
;W[er]
;B[bm]
;W[bn]
;B[bl]
;W[bp]
;B[fq]
;W[dl]
;B[ck]
;W[dk]
;B[cj]
;W[fr]
;B[ao]
;W[ap]
;B[fo]
;W[dj]
;B[ch]
;W[di]
;B[bg]
;W[am]
;B[bk]
;W[gn]
;B[hq]
;W[ip]
;B[iq]
;W[jp]
;B[kq]
;W[kp]
;B[lq]
;W[jq]
;B[jr]
;W[ir]
;B[hr]
;W[kr]
;B[is]
;W[mp]
;B[lp]
;W[lo]
;B[mo]
;W[np]
;B[ln]
;W[ko]
;B[lr]
;W[po]
;B[no]
;W[op]
;B[oo]
;W[pq]
;B[qp]
;W[qq]
;B[ro]
;W[pn]
;B[rq]
;W[mn]
;B[nn]
;W[mm]
;B[nm]
;W[nl]
;B[ml]
;W[lm]
;B[ol]
;W[nk]
;B[om]
;W[rr]
;B[oq]
;W[ok]
;B[nq]
;W[pk]
;TR[di][dj][dk][dl][dm][gn][ip][jp][kp][ko][lo][mn][mm][lm][nl][nk][ok][pk]L[eg]C[Probably
this move by AlphaGo is a
mistake.
The influence stone difference (the significant influence stones
there are marked) in the center is 0 - 18 = -18 in White's
favour. In other words, White has a sphere of dominating influence.
Black should play at A, which reduces the white center influence,
helps the somewhat weak black group on the left side and
attacks D16. For comparison, F17 does the latter but mostly neglects
the other two aspects.
There sphere of dominating influence has the great local potential of
being expanded and transformed to a solidified moyo by
White A. Therefore, A is a valuable intersection for either player's
move.
Besides, this principle applies to possibly reducing or invading the
sphere of dominating influence now\: "Reduce, but do
not invade, a moyo with a large influence stone difference and without
space for easy life of an invading group." That is,
the reduction A is possible, but an immediate invasion would be a
mistake.
AlphaGo has learned knowledge from next move shape samples, so we can
assume that both A and F17 must have been among the
next move candidates. On the other hand, there would be only a few
samples supporting the exact shape of the boundary of the
white sphere of dominating influence. Now, it is speculation whether
AlphaGo suffered from insufficient samples for such
spheres of influence or whether subsequent Monte Carlo playouts hinted
at a preference for F17 over A. The latter is not
inconceivable because there is aji in the white walls and so a large
fraction of sample playouts might hit at successful
later exploitations of aji. In fact, this is exactly how the actual
game ended.]B[fc]
;PM[2]FG[259:]
;W[cf]
;B[dh]
;W[eh]
;B[eg]
;W[fh]
;B[df]
;W[bf]
;B[fg]
;W[gh]
;B[ce]
;W[de]
;B[be]
;W[ef]
;B[dg]
;W[bd]
;B[af]
;W[cd]
;B[cg]
;W[gd]
;B[hc]
;W[gc]
;B[gb]
;W[fd]
;B[eb]
;W[ec]
;B[fb]
;W[db]
;B[jc]
;W[ke]
;B[ld]
;W[le]
;B[md]
;W[nf]
;B[pf]
;W[pc]
;B[qc]
;W[od]
;B[oe]
;W[qd]
;B[pe]
;W[qb]
;B[rc]
;W[rb]
;B[oc]
;W[pb]
;B[rd]
;W[ob]
;B[mb]
;W[sb]
(
;B[ho]
;W[jm]
;B[oh]
;W[nh]
;B[ni]
;W[oi]
;B[pi]
;W[oj]
;B[hn]
;W[hl]
;B[hg]
;W[gg]
;B[ie]
;W[if]
;B[hf]
;W[ig]
;B[hh]
;W[hd]
;B[id]
;W[he]
;B[jf]
;W[ih]
;B[gf]
;W[ff]
;B[hi]
;W[kg]
;B[fj]
;W[ei]
;B[ii]
;W[kh]
;B[gl]
;W[gk]
;B[hk]
;W[ik]
;B[fk]
;W[hj]
;B[gi]
;W[en]
;B[fl]
;W[fn]
;B[eo]
;W[em]
;B[hm]
;W[gm]
;B[il]
;W[hk]
;B[go]
)
(
;MA[ng][nh][ni][nj][ho][em][fm][jf][ig][hg]PM[2]L[kn][km][ll][ei]C[The
big white moyo has major weaknesses at the crossed
gaps and the labelled minor weaknesses.
Reduction of a big moyo uses a method (see the book), of which I
mention some important aspects\: First, identify the major
weaknesses of the moyo. Combine imagination of reduction, positional
judgement and reduction strategies exploiting the major
weaknesses of the moyo.
And, from a human player's perspective, this is exactly what AlphaGo
did with moves 137 - 149 and afterwards, then combined
with reading and re-evaluation of the then more important weaknesses D
and the gap around E7, which initially is the least
important of the major weaknesses.
White 138 also decreased the degree of the minor weaknesses ABC. So in
the end, we can say that AlphaGo exploited every aji
to reduce the white moyo. In particular, the program started with the
most important weaknesses by devising a timing
enabling this - instead of failing to exploit some major weakness at
all.
Besides, we see AlphaGo's implicit application of these related
principles\:
- Perform a series of sente-like \[i.e., double sente or sente\]
actions. Optionally, conclude with a gote-like \[i.e.,
reverse sente or gote\] action.
- Do not let your earlier actions inhibit your planned later actions.
- \[A principle of first strengthening adjacent groups need not be
applied because Black's groups already are strong.\]
- \[...\] the attacker of a moyo must avoid suffering from severe
collateral damage \[...\] (Compare move 155.)
Hence AlphaGo's skill in reducing a big moyo agrees to the book's
theory of how to do so. However, the program achieves this
implicitly by relying on shape knowlede, neural nets and random sample
continuations while the book provides human-
applicable theory not needing database knowledge and excessive
calculation power.
Program and human play are related, however, when it comes to reading
of the created capturing race. Reading tactics is
necessary for both, although AlphaGo reads implicitly by means of
sufficiently good sampling while a human player might
prefer to read explicitly.]FG[259:]
)
)
[/sgf]
[sgf]
(;ST[2]GM[1]FF[4]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GN[
]PM[2]KM[6.50]RE[B+Resign]PB[Alpha Go]RU[Japanese]FG[259:]PW[Fan Hui]
C[Commentary by Robert Jasiek using theory of the book Positional
Judgement 2 - Dynamics
Best viewed with a local SGF editor that shows arbitrary labels. Note
that there are variations starting with a pass or the
same move as the major sequence's move. Variations contain essential
diagram SGF nodes starting new diagrams even when there
is no new move. Online viewing of this SGF may be insufficient for
seeing all the commentary, markup, SGF diagrams and
variations.]
;B[dp]
;W[pp]
;B[dd]
;W[pc]
;B[pe]
;W[qe]
;B[qf]
;W[qd]
;B[pg]
;W[nc]
;B[qk]
;W[qm]
;B[ok]
;W[oq]
;B[dj]
;W[fq]
;B[dn]
;C[How can White expect to win if he leaves at least one weak group
(F3 or F17), which Black can attack favourably? This is
not a shinogi position but a position of voluntarily giving the
opponent an unnecessary advantage.
Ok, to be fair, Fan Hui could not know yet just how strong AlphaGo
would be in profiting from attacking a weak group. After
having seen all games of the match, we know it better.]W[fc]
;B[df]
;W[ip]
;B[ec]
(
;W[fd]
;B[id]
;W[he]
;B[ie]
;W[hg]
;B[hf]
;W[gf]
;B[if]
;W[gg]
;B[jh]
;C[Although the fight from move 32 to 98 looks complicated, it can be
understood as a series of one-sided actions by Black
or White to apply this principle\:
"The positional judgement after each action determines the judgement
before it."
Well, this is the theory, but the fight is long and complicated.
Predicting it to its quiet end and working judgements
backwards from the judgement after move 98 to the judgement of move 32
is too hard. Fan Hui may as well have played his move
for this very reason\: to start an unpredictable fight. The related
dry principle confirms this\:
"If, within the scope of reading, both players prefer a complex fight
during their strategic choices, the positional
judgement assesses the initial position simply as having an
unpredictable winner."
The fight starts with the one-move action of White's invasion and
installation of a second weak group.
However, Fan Hui is being unreasonable to start such an unpredictable
fight here because 1) both his groups are weak and 2)
each nearby black group is strong. It is common knowledge that
starting a fight under such circumstances should only be a
means of cheating when clearly behind. Hm, but considering that all
black stones work well while White has had a weak group
of 6 stones in a neutral region (where neither player can make much
territory), White already is behind!
It is the reader's exercise to decompose the following fight into its
one-sided actions.]W[dh]
;B[eg]
;W[ih]
;B[ji]
;W[eh]
;B[fh]
;W[fi]
;B[fg]
;W[ej]
;B[gi]
;W[gj]
;B[hj]
;W[dk]
;B[gk]
;W[fj]
;B[hi]
;W[ib]
;B[cj]
;W[cg]
;B[ck]
;W[be]
;B[cc]
;W[bc]
;B[bb]
;W[cd]
;B[ac]
;W[bd]
;B[db]
;W[de]
;B[ee]
;W[cf]
;B[ed]
;W[jc]
;B[hc]
;W[gb]
;B[hb]
;W[ha]
;B[fb]
;W[gc]
;B[dl]
;W[bi]
;B[ci]
;W[bh]
;B[qo]
;W[pl]
;B[pk]
;W[qp]
;B[fa]
;W[ga]
;B[lc]C[It seems that AlphaGo could "read" to move 98.]
;W[lb]
;B[kb]
;W[kc]
;B[mb]
;W[ld]
;B[mc]
;W[md]
;B[nb]
;W[ke]
;B[oc]
;W[od]
;B[nd]
;W[ne]
;B[pb]
;W[nc]
;B[ob]
;C[With the next move, a "complex fight" (neither a "one-sided fight"
nor a simple "exchange") starts and cannot be easily
decomposed into a series of one-sided fights. Therefore, in the now
created fighting region on the left lower side, every
relevant dynamic strategic aspect must be evaluated and re-evaluated
after every move to choose the best dynamic change for
the moving player. This should be done in a context of a, here
omitted, static positional judgement for the rest of the
board excluding the fighting region.
Initially, the left lower white group is alive and stable, White
threatens to enclose the right lower side and either player
has the option of making territory in the lower left corner.]W[pd]
;B[kq]C[The move destabilises the white group, creates an unstable
black group and prevents White from enclosing territory
on the right lower side.
For the sake of simplicity, let us assume that this invasion is more
attractive than the reduction L5.
Alternatively, Black could take the lower left corner with, e.g., D3,
which White might answer at L4.
Is L3 or D3 better? White answers either move, again stabilising or
increasing the stability of the white group. L3 means
that White can make a few excess points, D3 - L4 means that the
players make similar amounts of new points. There is,
however, a major difference between D3 and L3\: Black L3 or otherwise
White L4 greatly affect the potential development in
the lower center, which is wide and adjacent to strong black and white
stones so that either player threatens to build a
moyo there or prevent the opponent's moyo. Hence, L3 is more valuable
than D3.]
;C[White stabilises the white group and destabilises the black
stone.]W[jq]
;B[kp]C[Black stabilises his and destabilises the white group.]
(
;W[dr]C[White stabilises F3 and prepares taking the corner territory
but leaves behind the weakness A, which White B might
eliminate. As the variation suggests, D2 is better than B. A similar
argument could be made against C.]L[hq][kr][in]
;FG[259:]PM[2]
;B[hq]C[Black destabilises all white stones and stabilises his other
string by creating the option of connecting at A.]L[jr]
;FG[259:]PM[2]
;C[and so on]W[ir]
;B[gq]
;W[fp]
;B[fr]
;W[er]
;B[ho]
;W[gr]
;B[hr]
;W[fs]
;B[go]
;C[Result\: White has taken the corner and reduced the black territory
on the left side a bit. Black has stabilised his
strings on the lower side, claims territory in the lower left center,
threatens to expand the moyo rightwards, has reduced
White's center potential and prevents white territory on the right
lower side.
Territorially, probably this is fair but Black has gained additional
center potential while White has lost quite some center
potential. Hence, considering all, Black has gained from the fight\:
the center potential has been shifted greatly from
"shared" to "dominated by Black".]W[cq]
;B[rp]
;W[rq]
;B[ig]
;W[qb]
;B[rb]
;W[qa]
;B[rc]
;W[qc]
;B[la]
;W[on]
;B[rf]
;W[sb]
;B[mg]
;W[of]
;B[og]
;W[mf]
;B[nm]
;W[lg]
;B[mh]
;W[lo]
;B[ko]
;W[ln]
;B[kn]
;W[lm]
;B[km]
;W[ll]
;B[om]
;W[pn]
;B[mp]
;W[mq]
;B[np]
;W[nn]
;B[nq]
;W[kr]
;B[mr]
;W[hs]
;B[or]
;W[pr]
;B[op]
;W[pq]
;B[po]
;W[oo]
;B[ro]
;W[sq]
;B[qr]
;W[ps]
;B[rr]
;W[sr]
;B[rm]
;W[rl]
;B[ql]
)
(
;FG[259:]PM[2]
;W[kr]
;B[lr]
;W[jr]
;B[rp]
;W[rq]
;B[er]
;W[jn]
;B[no]C[Black took the lower left corner and now threatens to defend A
or B. White would have a hard time catching up with
Black's excess of new territory gained during the fight.]L[kp][qo]
)
)
(
;FG[259:]PM[2]
;W[eb]
;B[db]
;C[This variation can lead to a long endgame, something Fan Hui tried
and lost in game 1. Besides, the low and narrow
formation on the upper side is unattractive.]W[ic]
)
)
[/sgf]
[sgf]
(;ST[2]GM[1]FF[4]AP[GOWrite:2.3.48]SZ[19]CA[UTF-8]GN[
]PM[2]KM[6.50]RE[W+Resign]PB[Fan Hui]RU[Japanese]FG[259:]PW[Alpha Go]
C[Commentary by Robert Jasiek using theory of the book Positional
Judgement 2 - Dynamics
Best viewed with a local SGF editor.]
;B[dp]
;W[pp]
;B[cd]
;W[pd]
;B[nc]
;W[qf]
;B[ic]
;W[cj]
;B[ci]
;W[di]
;B[ch]
;W[dj]
;B[dh]
;W[cn]
;B[ei]
;W[fq]
;B[eo]
;W[cq]
;B[cl]
;W[dl]
;B[cp]
;W[dq]
;B[dm]
;W[cm]
;B[el]
;W[dk]
;B[bl]
;W[bk]
;B[dn]
;W[bm]
;B[gp]
;W[gq]
;B[hp]
;W[iq]
;B[pb]
;W[qc]
;B[ld]
;W[ec]
;B[dc]
;W[db]
;B[ed]
;W[cc]
;B[dd]
;W[bb]
;B[eb]
;W[fc]
;B[fb]
;W[bd]
;B[da]
;W[cb]
;B[be]
;W[ac]
;B[ek]
;W[bi]
;B[bh]
;W[fd]
;B[gb]
;W[id]
;B[jd]
;W[hc]
;B[hb]
;W[jc]
;B[ib]
;W[md]
;B[je]
;W[mc]
;B[me]
;W[nd]
;B[lb]
;W[lc]
;B[kc]
;W[le]
;B[ai]
;W[bo]
;B[lf]
;W[kd]
;B[jb]
;W[ej]
;B[fj]
;W[fi]
;B[bj]
;W[eh]
;B[ak]
;W[gj]
;B[fk]
;W[ef]
;B[fe]
;W[gk]
;B[ck]
;W[ei]
;B[fm]
;W[hh]
;B[ji]C[The territorial deire value of A is ca. 22 points. Besides, A
greatly strengthens the black group in the lower
center. If Black's move is gote because the white group lives, it is
hard to justify the move as more valuable than A.
Maybe Fan Hui expected his move to be sente. AlphaGo's "reading"
during the subsequent fight is impressive\: the program
defends life.]L[bp]
;FG[259:]PM[2]
;W[bp]
;B[ij]
;W[ff]
;B[ge]
;W[hl]
;B[jl]
;W[im]
;B[jo]
;C[Except for intervening forcing sequences before and afterwards
during the fight, now its phase of sente actions by Black
to attack the life of the white group ends.]W[jm]
;B[km]C[This move starts the "reward action", with which Black rewards
himself at the end of the fight.
It is a general nature of one-sided fights that the attacker plays a
series of sente actions and optionally concludes with a
reward action.
]
;W[kn]
;B[jn]
;W[kl]
;B[lm]
;W[jk]
;B[ll]
;W[kk]
;B[ik]
;W[il]
;B[lk]
;W[kj]
;B[mi]
;W[kh]
;B[jh]
;W[lj]
;B[mj]
;W[lh]
;B[ke]
;W[mg]
;B[lo]C[This concludes the reward action. Here, as usual, the reward
action is a gote action. Black's reward is the creation
of his group of thickness on the outside. Still it is unconceivable
that Black can compensate his loss at B4.
The position before the fight (before move 93) inherits the judgement
of the position after the fight (after move 123).
Instead, Fan Hui seems to have expected killing the group. In this
case, the judgement before the fight would be "win
because I kill you".]
;W[qm]
;B[ne]
;W[oe]
;B[ld]
;W[fp]
;B[hn]
;W[fo]
;B[gn]
;W[og]
;B[mb]
;W[od]
;B[qq]
;W[qp]
;B[pq]
;W[oq]
;B[or]
;W[nr]
;B[rp]
;W[ro]
;B[op]
;W[nq]
;B[rr]
;W[sp]
;B[sq]
;W[rq]
;B[gl]
;W[hk]
;B[rp]
;W[pr]
;B[qr]
;W[rq]
;B[nh]
;W[ng]
;B[rp]
;W[mn]
;B[ln]
;W[rq]
;B[gi]
;W[hi]
;B[rp]
;W[ce]
;B[so]
;W[ee]
;B[oo]
;W[pj]
;B[lr]
;W[pn]
;B[rn]
;W[on]
;B[mo]
;W[nk]
;B[mk]
;W[jr]
;B[qk]
;W[pk]
;B[qj]
;W[qi]
;B[rl]
;W[ri]
;B[kr]
;W[kq]
;B[lq]
;W[kp]
;B[lp]
;W[rm]
;B[sm]
;W[ql]
;B[po]
;W[rk]
;B[qb]
;W[rb]
;B[ra]
;W[rc]
;B[oc]
;W[hq]
;B[gc]
;W[de]
;B[hf]
;W[gh]
;B[bf]
;W[ig]
;B[jg]
;W[if]
;B[ie]
;W[ep]
;B[do]
;W[fn]
;B[ip]
;W[jq]
;B[ko]
;C[Black resigns.]W[no]
)
[/sgf]