Re: [gnugo-devel] Extremely bad mistake

[Gunnar Farneback]

I wrote:
> I was sent a game where GNU Go lost with 26 handicap stones due to a
> repeated mistake. This is extremely bad. I would even say a
> showstopper for 3.4. (I haven't checked it but would guess the semeai
> code is at fault.)

It turns out that this primarily is a tactical reading mistake. The
appended patch adds some relevant test cases.

A part of the solution is probably to make break_chain3_moves() more
ambitious. I know Paul has started on this. Do you have any results to
report?

/Gunnar

Index: reading.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/reading.tst,v
retrieving revision 1.54
diff -u -r1.54 reading.tst
--- reading.tst 22 Apr 2003 02:49:01 -0000      1.54
+++ reading.tst 11 Jun 2003 15:45:25 -0000
@@ -750,6 +750,31 @@
 181 defend E3
 #? [1 E4]*
 
+# A18 only attacks with bad ko (extremely bad since the whole semeai
+# is lost if white ignores the ko threat). A17 also attacks with bad
+# ko but here a lost ko means seki. A19 gives a good ko but losing it
+# costs the whole semeai.
+# The distinction between these results is too subtle for the tactical
+# reading code so for now we only check that it realizes that there is
+# no attack without ko.
+loadsgf games/reading41.sgf 118
+182 attack C19
+#? [(2|3) (A17|A19|A18)]
+183 defend C19
+#? [1 (A19|E11)]
+
+# Playing E11 immedidately gives a bad ko. A18 obviously is totally
+# ineffective. Necessary to start at A16.
+loadsgf games/reading41.sgf 130
+184 attack C19
+#? [2 A16]
+185 attack A19
+#? [2 A16]
+186 defend C19
+#? [1 E11]
+187 defend A19
+#? [1 E11]
+
 # Report number of nodes visited by the tactical reading
 10000 get_reading_node_counter
 #? [0]&
Index: semeai.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/semeai.tst,v
retrieving revision 1.40
diff -u -r1.40 semeai.tst
--- semeai.tst  8 Jun 2003 14:46:56 -0000       1.40
+++ semeai.tst  11 Jun 2003 15:45:28 -0000
@@ -258,6 +258,20 @@
 50 analyze_semeai O7 L7
 #? [1 1 (J5|J6|K7|J4|G5|G6|H7)]
 
+# See reading:182,183 for comments.
+loadsgf games/reading41.sgf 118
+51 analyze_semeai C19 B16
+#? [1 0 (A19|E11)]
+52 analyze_semeai B16 C19
+#? [1 3 A17]
+
+# See reading:184-187 for comments.
+loadsgf games/reading41.sgf 130
+53 analyze_semeai C19 B16
+#? [1 0 E11]
+54 analyze_semeai B16 C19
+#? [1 2 A16]
+
 
 ########### end of semeai tests #################
 
Index: strategy5.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/strategy5.tst,v
retrieving revision 1.27
diff -u -r1.27 strategy5.tst
--- strategy5.tst       12 Apr 2003 02:57:59 -0000      1.27
+++ strategy5.tst       11 Jun 2003 15:45:28 -0000
@@ -110,6 +110,11 @@
 278 gg_genmove black
 #? [!J11]*
 
+# Really, anything is better than A18. (B13 and B12 are quite
+# ineffective too but can at least not be repeated...)
+loadsgf games/reading41.sgf 130
+279 gg_genmove black
+#? [!A18]
 
 ############ End of Tests #################
 
Index: games/reading41.sgf
===================================================================
RCS file: games/reading41.sgf
diff -N games/reading41.sgf
--- /dev/null   1 Jan 1970 00:00:00 -0000
+++ games/reading41.sgf 11 Jun 2003 15:45:28 -0000
@@ -0,0 +1,59 @@
+(;GM[1]FF[4]
+SZ[19]HA[9]KM[0.5]
+PW[qwert]WR[NR]
+PB[ppmm (GNU Go 3.3.19)]BR[9k*]
+PC[THQY]DT[2003-06-08]AP[gnugoclient:2.0]
+RE[W+170.5]
+AB[dd][jd][pd][dj][jj][pj][dp][jp][pp]
+;W[];B[gj];W[];B[om];W[];B[go];W[];B[jg];W[];B[jm];W[];B[mj];W[]
+;B[jq];W[];B[dc];W[];B[cg];W[];B[pq];W[];B[cj];W[];B[qj];W[];B[cp]
+;W[];B[pc];W[];B[dg];W[];B[gg];W[];B[jc];W[fe];B[gd];W[ln];B[kn]
+;W[bl];B[bn];W[db];B[eb];W[cb];B[ge];W[ce];B[bd];W[ec];B[fb];W[bc]
+;B[be];W[cd];B[de];W[cc];B[cf];W[ef];B[ed];W[fh];B[ei];W[eh];B[dh]
+;W[hi];B[ih];W[hh];B[hg];W[fi];B[hj];W[ej];B[ii];W[di];B[ci];W[ek]
+;B[eg];W[fg];B[fo];W[dm];B[el];W[dl];B[mk];W[bh];B[bi];W[df];B[dk]
+;W[gk];B[fl];W[fk];B[ol];W[gm];B[ik];W[en];B[bm];W[cl];B[hn];W[ad]
+;B[af];W[bg];B[bf];W[aj];B[gf];W[ff];B[fd];W[ai];B[ei];W[fj];B[da]
+;W[ca];B[ea];W[di];B[ah];W[al];B[ch];W[bj];B[ba];W[bb];B[ae];W[ck]
+;B[ag];W[qn];B[ab];W[aa];B[ac];W[ad];B[ac];W[on];B[ro];W[rn];B[nn]
+;W[oo];B[no];W[op];B[ab];W[ad];B[ac];W[nr];B[ab];W[ad];B[ac];W[np]
+;B[mo];W[rq];B[ab];W[ad];B[ac];W[dr];B[cr];W[fr];B[mp];W[br];B[cs]
+;W[ir];B[jr];W[hq];B[es];W[er];B[ds];W[ep];B[eo];W[do];B[eq];W[cq]
+;B[bq];W[dq];B[fq];W[bp];B[gr];W[co];B[ab];W[ad];B[ac];W[qf];B[of]
+;W[qh];B[re];W[og];B[ri];W[nf];B[ng];W[oe];B[gq];W[nh];B[mg];W[pf]
+;B[rg];W[nc];B[ab];W[ad];B[ac];W[rc];B[qe];W[lc];B[bs];W[aq];B[fs]
+;W[mq];B[ab];W[ad];B[ac];W[ll];B[rl];W[km];B[ab];W[ad];B[ac];W[jl]
+;B[il];W[hm];B[ab];W[ad];B[ac];W[kp];B[lo];W[jn];B[im];W[ko];B[lq]
+;W[kq];B[mr];W[nq];B[lp];W[kr];B[lr];W[ks];B[hp];W[ms];B[rp];W[qr]
+;B[pr];W[rs];B[mm];W[ls];B[kk];W[qo];B[ab];W[ad];B[ac];W[mh];B[ab]
+;W[ad];B[ac];W[lg];B[js];W[io];B[ab];W[ad];B[ac];W[in];B[qp];W[sr]
+;B[fp];W[ql];B[mn];W[ml];B[kl];W[rk];B[ab];W[ad];B[ac];W[rf];B[nl]
+;W[lm];B[nm];W[rh];B[ab];W[ad];B[ac];W[rj];B[gn];W[fm];B[ab];W[ad]
+;B[ac];W[nj];B[oj];W[pb];B[ab];W[ad];B[ac];W[ob];B[ab];W[ad];B[ac]
+;W[ni];B[nk];W[ke];B[ab];W[ad];B[ac];W[je];B[qi];W[si];B[ab];W[ad]
+;B[ac];W[if];B[ab];W[ad];B[ac];W[id];B[ab];W[ad];B[ac];W[ig];B[ab]
+;W[ad];B[ac];W[hc];B[hb];W[gc];B[gb];W[fc];B[kb];W[hd];B[ei];W[bh]
+;B[ab];W[ad];B[ac];W[di];B[ie];W[he];B[ei];W[gh];B[mb];W[lb];B[la]
+;W[mc];B[ph];W[pg];B[ki];W[qk];B[lk];W[kn];B[pm];W[pk];B[ok];W[kh]
+;B[jh];W[li];B[qm];W[rm];B[dp];W[nb];B[ma];W[kd];B[ic];W[lj];B[kj]
+;W[oh];B[mi];W[lh];B[pn];W[po];B[na];W[ga];B[ha];W[kc];B[jb];W[ip]
+;B[iq];W[ho];B[hr];W[gs];B[hs];W[oa];B[ka];W[hk];B[hl];W[gl];B[pi]
+;W[jf];B[kg];W[kf];B[fn];W[gi];B[ij];W[ns];B[jo];W[ar];B[oi];W[cp]
+;B[ep];W[as];B[gs];W[];B[pl];W[];B[]
+C[R4 removed
+K8 removed
+J2 removed
+B6 removed
+E8 removed
+C11 removed
+E11 removed
+A17 removed
+D16 removed
+final_score: W+170.5
+G19 removed
+Q17 removed
+R15 removed
+N13 removed
+S13 removed
+S8 removed]
+)