Re: [gnugo-devel] 3.3.12 vs 3.2

[bump]

> I figured a third opinion would help. I ran 2 sets of 50 games between gnugo
> 3.2 as black and 3.3.12 (actually cvs version from saturday 14:00 GMT) as
> white. Komi was 5.5.
> 
> I will let you interpret the results. In the first one 3.3.12 wins 35 games,
> and in the second one it wins 32 games. Is anybody interested in some/all of
> the sgf files ? The whole gzipped thing is about 43kB.

Yesterday Gunnar was running twogtp and last I heard he was getting
a win rate of 55% for 3.3.12.

As I mentioned to Gunnar and Evan I played a game of 3.3.12 against
2.0 and for much of the game it liked like 2.0 would win. 

Gunnar's comment was "let me guess, 2.0 played better in the
fuseki." That's true --- it's certainly bad that 3.3.12
takes 5 moves to capture one stone in the upper right while
2.0 stakes out the rest of the board. Moves 80 and 84 are
very bad. 90 at N5 would connect. 92 at O6 would be better.
Even after move 93 W can win the semeai in the lower right
by connecting under at S3. But if B moves first, B wins the
semeai. GNU Go 3.3.12 thinks the position is seki.

So many misteps led to this result.

After that, W is still ahead, so my comment that 2.0 might
win is probably wrong, but still I found this game a little
sobering.

My guess is that 3.3.12 is stronger than 3.2 but not as
much as I would like. We had two losses against Aya,
one against Katsunari, last summer, and it was posted
recently to rec.games.go that the revised WinHonte beats
GNU Go 3.2.

Dan

(;GM[1]FF[3]
RU[Japanese]SZ[19]HA[0]KM[5.5]
PW[White]
PB[Black]
GN[White (W) vs. Black (B)]
DT[2002-11-23]
SY[Cgoban 1.9.12]TM[-];B[ec];W[cd];B[dp];W[de];B[oc];W[fd];B[oq];W[fc]
;B[qp];W[eb];B[qd];W[fq];B[cn];W[dr];B[cq];W[iq];B[qk];W[cl];B[kd];
W[cr];B[bq];W[oe];B[pf];W[ci];B[br];W[lq];B[lo];W[mp];B[ok];W[no];
B[dc];W[hf];B[cc];W[bc];B[bb];W[bd];B[qh];W[jf];B[ng];W[jd];B[jc];
W[ke];B[ld];W[kh];B[id];W[el];B[je];W[le];B[ne];W[ig];B[of];W[kj];
B[oh];W[om];B[en];W[mm];B[bl];W[bk];B[bm];W[jo];B[mk];W[gm];B[lm];
W[ll];B[ml];W[kl];B[ln];W[nr];B[mh];W[or];B[pq];W[rs];B[pr];W[qn];
B[rm];W[rn];B[qm];W[ps];B[me];W[sm];B[sl];W[sn];B[rl];W[fo];B[oo];
W[on];B[np];W[po];B[op];W[lp];B[mo];W[nm];B[nn];W[lf];B[jn];W[in];
B[fn];W[hc];B[gn];W[ib];B[go];W[kb];B[jb];W[ja];B[io];W[ho];B[ip];
W[hp];B[jp];W[kp];B[ko];W[jq];B[jo];W[im];B[hn];W[hm];B[hq];W[gp];
B[hr];W[gq];B[hs];W[gr];B[lb];W[ak];B[ab];W[db];B[cb];W[la];B[kc];
W[na];B[ka];W[ia];B[ic];W[hb];B[ma];W[rp];B[nb];W[bs];B[hd];W[gd];
B[rf];W[dm];B[rr];W[eo];B[do];W[eq];B[dn];W[cm];B[qq];W[qs];B[cg];
W[rq];B[eg];W[bf];B[qr];W[sr];B[bg];W[he];B[fh];W[dh];B[dg];W[ff];
B[fj];W[ei];B[fi];W[bi];B[ej];W[gk];B[ii];W[ij];B[hi];W[dj];B[fk];
W[fl];B[da];W[fb];B[lg];W[kg];B[mf];W[kf];B[kr];W[lr];B[ar];W[ep];
B[rb];W[lh];B[li];W[lj];B[mj];W[ki];B[mi];W[ie];B[jd];W[al];B[bn];
W[kq];B[mg];W[am];B[lk];W[kk];B[an];W[nl];B[nk];W[pm];B[pl];W[dq];
B[ol];W[em];B[tt];W[fm];B[tt];W[cs];B[tt];W[as];B[tt];W[tt]
)