2 games between the strongest Go opponents

Discussion of computer chess matches and engine tournaments.

Moderator: Ras

User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 10:21 pm
Full name: Kai Laskos

2 games between the strongest Go opponents

Post by Laskos »

I played manually 2 games between the strongest publicly available Go opponents: Zenith 4 (Zen engine) and Crazy Stone 2012. The time control was 30 seconds per move, no handicap, komi 6.5. All four cores of the computer used.

The first game saw Zen with Black, and it crushed Crazy Stone by a whopping 87.5 points

SGF:
(;GM[1]FF[4]AP[Zenith:4.0]SZ[19]HA[0]KM[6.5]CA[UTF-8]
PB[Zenith Go 30 sec.]BR[+]PW[You]WR[]ZT[30]DT[2013-01-27]RE[B+87.5];B[pd];
W[pp];B[dd];W[dp];B[jd];W[md];B[pf];W[pc];B[qc];W[oc];B[od];W[nd];B[lc];
W[ld];B[kc];W[mg];B[pi];W[jp];B[qn];W[np];B[dj];W[qb];B[cn];W[qd];B[qe];
W[rc];B[ig];W[re];B[gn];W[gp];B[eo];W[ep];B[mi];W[ql];B[om];W[ol];B[nl];
W[ok];B[oo];W[nm];B[op];W[oq];B[no];W[qq];B[mm];W[nk];B[ml];W[lj];B[lh];
W[ni];B[nh];W[mh];B[li];W[oi];B[mp];W[nq];B[io];W[pm];B[nn];W[lk];B[kl];
W[ip];B[jj];W[pn];B[po];W[qo];B[lg];W[lf];B[ce];W[db];B[cc];W[jn];B[jm];
W[in];B[ho];W[kn];B[hp];W[hq];B[jo];W[ko];B[kp];W[kq];B[lp];W[ll];B[lm];
W[km];B[jl];W[hm];B[gm];W[hl];B[hn];W[im];B[hk];W[gl];B[lq];W[jq];B[ln];
W[lo];B[mo];W[on];B[em];W[el];B[dl];W[ek];B[fj];W[gk];B[lr];W[ik];B[jr];
W[jk];B[ir];W[gc];B[cq];W[ic];B[mf];W[ng];B[kf];W[nf];B[fb];W[fc];B[eb];
W[ec];B[dc];W[cb];B[bb];W[id];B[je];W[hf];B[fe];W[ff];B[ef];W[hg];B[hi];
W[gi];B[gj];W[iq];B[kr];W[hr];B[fr];W[gq];B[cp];W[do];B[dn];W[co];B[bo];
W[bp];B[bq];W[bn];B[dk];W[er];B[dr];W[ej];B[ei];W[hj];B[gh];W[ii];B[hh];
W[ih];B[ij];W[ji];B[hk];W[kj];B[hj];W[ki];B[fl];W[kk];B[jb];W[jc];B[kd];
W[ib];B[gd];W[if];B[ie];W[he];B[hd];W[ee];B[ed];W[jg];B[gb];W[kb];B[lb];
W[ja];B[ha];W[hb];B[fd];W[dq];B[es];W[eq];B[br];W[cs];B[fs];W[ga];B[fa];
W[hc];B[ca];W[la];B[mb];W[nb];B[ma];W[ka];B[da];W[ds];B[cr];W[db];B[cb];
W[mr];B[hs];W[jf];B[ga];W[gs];B[gr];W[is];B[js];W[fq];B[bs];W[ds];B[cs];
W[ao];B[bm];W[aq];B[ar];W[bl];B[cm];W[am];B[fk];W[fo];B[en];W[bj];B[bi];
W[ci];B[cj];W[bk];B[bh];W[cg];B[il];W[ch];B[bg];W[bf];B[cf];W[be];B[bd];
W[eg];B[gf];W[df];B[gg];W[de];B[cd];W[fg];B[le];W[aj];B[al];W[ag];B[ae];
W[ai];B[af];W[rf];B[me];W[pe];B[eh];W[ne];B[mj];W[mk];B[mq];W[ns];B[kg];
W[ge];B[nj];W[oj];B[ls];W[nm];B[kh];W[mc];B[ia];W[na];B[jb];W[jh];B[ms];
W[nr];B[ah];W[om];B[];W[an];B[];W[])

Image

In the second game Crazy Stone had black, and it beat Zen by 14.5 points

SGF:
(;GM[1]FF[4]AP[Zenith:4.0]SZ[19]HA[0]KM[6.5]CA[UTF-8]PB[You]BR[]
PW[Zenith Go 30 sec.]WR[+]ZT[30]DT[2013-01-27]RE[B+Resign];B[dp];W[dd];
B[pp];W[pd];B[jp];W[dj];B[dm];W[pj];B[pl];W[nj];B[nc];W[jc];B[qc];W[pc];
B[qd];W[pe];B[pb];W[ob];B[qb];W[oc];B[qf];W[pf];B[qg];W[pg];B[dg];W[hg];
B[gh];W[eh];B[dh];W[gj];B[ei];W[ej];B[fi];W[gl];B[hi];W[ff];B[di];W[cj];
B[cl];W[el];B[ik];W[im];B[fn];W[hn];B[kl];W[km];B[lm];W[ko];B[kp];W[jl];
B[kn];W[kk];B[jm];W[jn];B[ee];W[ef];B[df];W[de];B[ed];W[ec];B[fd];W[cf];
B[bg];W[ih];B[gf];W[gg];B[dc];W[fc];B[gd];W[cc];B[db];W[cb];B[fg];W[ge];
B[hf];W[he];B[if];W[fe];B[gc];W[hh];B[fb];W[da];B[eb];W[id];B[ie];W[hd];
B[be];W[bd];B[fj];W[hb];B[fk];W[ln];B[lo];W[km];B[mn];W[bf];B[gk];W[eg];
B[fh];W[ji];B[ll];W[jk];B[fl];W[hp];B[go];W[qq];B[pq];W[qp];B[ho];W[po];
B[qo];W[qn];B[ro];W[pr];B[or];W[qr];B[mj];W[mi];B[lj];W[li];B[pn];W[oo];
B[np];W[rn];B[af];W[ce];B[ae];W[ad];B[ag];W[cg];B[ch];W[cd];B[bh];W[bb];
B[ab];W[ac];B[kd];W[kc];B[lc];W[ke];B[nb];W[oa];B[ph];W[qe];B[re];W[qh];
B[qi];W[rh];B[rg];W[sg];B[rf];W[ri];B[oh];W[pi];B[ng];W[mf];B[le];W[lb];
B[nf];W[me];B[ne];W[nh];B[md];W[mb];B[kf];W[mg];B[og];W[mc];B[na];W[ld];
B[nd];W[mo];B[kn];W[lf];B[on];W[no];B[oi];W[oj];B[qk];W[qj];B[jo];W[mp];
B[mq];W[nq];B[ni];W[rk];B[mh];W[lh];B[ii];W[kh];B[rl];W[pk];B[sk];W[ql];
B[qm];W[rm];B[qk];W[sj];B[rj];W[lk];B[si];W[nk];B[sn];W[rp];B[sm];W[lq];
B[mk];W[ml];B[kr];W[mm];B[mr];W[ip];B[nr];W[nn];B[ln];W[lp];B[jm];W[lr];
B[in];W[jr];B[ir];W[kq];B[jq];W[ks];B[is];W[om];B[il];W[cq];B[dq];W[ij];
B[hj];W[iq];B[gr];W[hr];B[gs];W[js];B[hs];W[fr];B[gq];W[cp];B[dr];W[ma];
B[kj];W[sh];B[hc];W[ib];B[ic];W[jd];B[jj];W[ol];B[sd];W[pm];B[qn];W[cn];
B[bo];W[sj];B[rk];W[gp];B[fq];W[cm];B[co];W[dn];B[bl];W[io];B[bn];W[ap];
B[bp];W[gn];B[en];W[eo];B[bm];W[nh];B[pa])

Image

These results seem to confirm what is seen on KGS (maybe a slight superiority of Zen), but on a personal computer with publicly available software. I cannot say much about the quality of play, as my level is some 15 stones below these two marvels. My impression is that they have very strong fuseki and endgame, first by some pattern recognition, second by sheer calculating power.
User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 10:21 pm
Full name: Kai Laskos

Re: 2 games between the strongest Go opponents

Post by Laskos »

Another clear win of Zen (as black) against Crazy Stone, a crushing 130.5 points. I don't know how Crazy Stone managed to lose that badly, even I loose gentler.

SGF:
(;GM[1]FF[4]AP[Zenith:4.0]SZ[19]HA[0]KM[6.5]CA[UTF-8]
PB[Zenith Go 30 sec.]BR[+]PW[You]WR[]ZT[30]DT[2013-01-27]RE[B+130.5];
B[pd];W[dp];B[dd];W[pp];B[dj];W[jp];B[cn];W[fp];B[jd];W[qf];B[pj];W[ql];
B[of];W[og];B[pg];W[pf];B[qg];W[rg];B[rh];W[re];B[mj];W[oe];B[nf];W[pe];
B[me];W[md];B[ld];W[mc];B[lc];W[lb];B[kb];W[dl];B[eo];W[ep];B[cl];W[dm];
B[sg];W[rf];B[ck];W[en];B[ij];W[gj];B[fk];W[gk];B[gh];W[fi];B[il];W[gm];
B[do];W[co];B[bn];W[hi];B[cp];W[fo];B[gi];W[fj];B[hj];W[fh];B[qq];W[pq];
B[qp];W[qo];B[ro];W[qn];B[gg];W[fg];B[gf];W[ee];B[cf];W[cg];B[df];W[dg];
B[ef];W[ff];B[fe];W[ed];B[fd];W[ec];B[dc];W[ge];B[fc];W[eb];B[he];W[de];
B[ce];W[dn];B[bp];W[cm];B[bm];W[ne];B[mf];W[fb];B[db];W[bo];B[ao];W[do];
B[cr];W[cq];B[bq];W[ap];B[aq];W[an];B[am];W[dq];B[pr];W[or];B[mb];W[ci];
B[dr];W[er];B[bj];W[nb];B[qi];W[ok];B[la];W[nj];B[ni];W[oc];B[qr];W[mk];
B[lj];W[lk];B[kj];W[jk];B[kk];W[jj];B[ji];W[oj];B[oi];W[kl];B[jl];W[ph];
B[qh];W[cd];B[gb];W[be];B[bf];W[bg];B[bd];W[eg];B[ae];W[cj];B[nr];W[cc];
B[ea];W[gc];B[fa];W[oq];B[os];W[ps];B[ns];W[nq];B[mr];W[lr];B[mq];W[lq];
B[mp];W[rn];B[so];W[mn];B[km];W[lo];B[pk];W[lm];B[no];W[pl];B[ol];W[om];
B[nl];W[nm];B[nk];W[jm];B[ll];W[jr];B[dk];W[sq];B[sr];W[rr];B[rq];W[ss];
B[rs];W[qs];B[rs];W[qs];B[ps];W[sr];B[kn];W[sn];B[rp];W[bi];B[ak];W[aj];
B[ai];W[br];B[bs];W[ar];B[as];W[ds];B[io];W[br];B[ar];W[jo];B[jn];W[qk];
B[ag];W[es];B[cs];W[bh];B[ah];W[ip];B[ho];W[hp];B[hm];W[hl];B[gn];W[fn];
B[ek];W[go];B[hn];W[fl];B[hk];W[gl];B[on];W[pn];B[oo];W[po];B[rj];W[rk];
B[lp];W[sj];B[ko];W[kp];B[pm];W[mo];B[np];W[op];B[sp];W[nn];B[rs];W[ls];
B[qj];W[di];B[ej];W[ei];B[el];W[na];B[ma];W[qd];B[kh];W[sf];B[sh];W[em];
B[bc];W[ms];B[];W[])

Image
User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 4:27 pm

Re: 2 games between the strongest Go opponents

Post by Don »

Computer go has come an amazingly long way in just the past 5 or 10 years. More than even chess has.

It's all due to MCTS or Monte Carlo Tree Search. Until this came along most programs just looked at patterns and did local searches to prove a group was alive or try to kill a group but no real global strategy other than patterns to suggest moves.

But with MCTS the program are doing fairly deep searches - using playouts as a kind of evaluation function. A playout is like a randomly played game but the games are not completely random. The tree search is highly selective and done best first fashion. Lines that look good statistically are expanded. It was really a MAJOR breakthrough in computer Go and now every year they are getting better.

I think Zen is the top program (not sure about this) but Crazy Stone is no slouch.

If you play go and didn't know about this development you should check it out, there are a few open source programs that are incredibly strong. At 9x9 they are almost unbeatable unless you are a high Dan pro.
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 10:21 pm
Full name: Kai Laskos

Re: 2 games between the strongest Go opponents

Post by Laskos »

Don wrote:Computer go has come an amazingly long way in just the past 5 or 10 years. More than even chess has.

It's all due to MCTS or Monte Carlo Tree Search. Until this came along most programs just looked at patterns and did local searches to prove a group was alive or try to kill a group but no real global strategy other than patterns to suggest moves.

But with MCTS the program are doing fairly deep searches - using playouts as a kind of evaluation function. A playout is like a randomly played game but the games are not completely random. The tree search is highly selective and done best first fashion. Lines that look good statistically are expanded. It was really a MAJOR breakthrough in computer Go and now every year they are getting better.

I think Zen is the top program (not sure about this) but Crazy Stone is no slouch.

If you play go and didn't know about this development you should check it out, there are a few open source programs that are incredibly strong. At 9x9 they are almost unbeatable unless you are a high Dan pro.
I was aware of the progress, in fact I felt it directly. While I am a weak Go player, I have some friends who play in the 3 kyu to 1 amateur Dan range. 7-10 years ago even I was beating Gnu Go at 9x9 or 19x19, to my friends the games against computers were a waste of time, and it looked like Go will be a tough nut to crack. Then, with the advent of MCTS (MoGo was the first?) my friends started complaining that they loose on 9x9 against bots. And recently (the last 2-3 years) they started to loose on 19x19. A tremendous progress right in front of my eyes in some 7 years, from being disparaged as 20 kyu by good amateur players to the point that these good players loose to bots. I think even the Chess developed slower.
I pitted here the best two based on KGS rankings (Zen 6d and Crazy Stone 5d), personally I play them only hitting very often "hint" button, otherwise I have no chance, and I don't like the idea of 10-stone handicap. It seems that the KGS rankings differences translate directly on a personal computer (they use many cores on KGS), probably Zen is a bit better than Crazy Stone.
User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 10:21 pm
Full name: Kai Laskos

Re: 2 games between the strongest Go opponents

Post by Laskos »

The fourth and last game I am posting, another win by Zen with white, for a total score of 3:1 for Zen. It seems that 1 Dan difference on KGS is translated to the same difference on a quad core PC.

SGF:
(;GM[1]FF[4]AP[Zenith:4.0]SZ[19]HA[0]KM[6.5]CA[UTF-8]PB[You]BR[]
PW[Zenith Go 30 sec.]WR[+]ZT[30]DT[2013-01-28]RE[W+61.5];B[dp];W[pd];
B[dc];W[pp];B[qn];W[de];B[nq];W[pn];B[pm];W[on];B[qp];W[qo];B[ro];W[po];
B[qq];W[rn];B[qm];W[rp];B[rm];W[so];B[pq];W[rq];B[om];W[mp];B[nn];W[no];
B[op];W[mn];B[nm];W[oo];B[lq];W[mq];B[mr];W[np];B[oq];W[nr];B[kr];W[or];
B[qf];W[fd];B[ce];W[cf];B[cd];W[dg];B[ch];W[me];B[of];W[qe];B[od];W[nd];
B[oc];W[oe];B[pf];W[pe];B[nf];W[ne];B[qr];W[rr];B[bf];W[dh];B[di];W[ei];
B[dj];W[ej];B[dk];W[cq];B[dq];W[lg];B[mh];W[ek];B[mb];W[kc];B[kb];W[lc];
B[lb];W[nc];B[jc];W[hc];B[nb];W[ob];B[pc];W[pb];B[qc];W[qb];B[ic];W[cp];
B[rc];W[rb];B[co];W[hb];B[ie];W[ib];B[jb];W[jg];B[ke];W[mf];B[mg];W[rf];
B[rg];W[do];B[cn];W[fq];B[cr];W[iq];B[ir];W[jq];B[hq];W[hp];B[gq];W[gp];
B[fr];W[er];B[fp];W[eq];B[ep];W[jr];B[gr];W[jf];B[je];W[hf];B[hd];W[mk];
B[mm];W[lm];B[ll];W[kk];B[lk];W[fo];B[dr];W[lj];B[kl];W[nj];B[pj];W[km];
B[ml];W[dn];B[kj];W[li];B[lf];W[cm];B[if];W[ig];B[bm];W[bl];B[dm];W[cl];
B[eo];W[en];B[bo];W[dl];B[fn];W[em];B[lp];W[lo];B[ko];W[jm];B[in];W[kp];
B[hg];W[hh];B[gg];W[jk];B[jl];W[jo];B[ik];W[jj];B[gh];W[ij];B[hi];W[ih];
B[hj];W[kh];B[eh];W[cg];B[bg];W[eg];B[gd];W[bh];B[ci];W[bi];B[sc];W[rd];
B[sf];W[re];B[js];W[fc];B[ff];W[dd];B[sb];W[sd];B[oh];W[cc];B[bc];W[cb];
B[ns];W[qs];B[os];W[pr];B[kq];W[kn];B[db];W[bb];B[ip];W[jp];B[io];W[eb];
B[ab];W[ec];B[ba];W[da];B[jn];W[ko];B[bj];W[ck];B[aj];W[cj];B[ji];W[ii];
B[hl];W[gk];B[ki];W[hk];B[fk];W[gi];B[fi];W[gj];B[al];W[qd];B[fl];W[fh];
B[il];W[fm];B[gn];W[ak];B[bk];W[ai];B[fe];W[fg];B[gf];W[ak];B[gl];W[im];
B[nk];W[hn];B[gm];W[mj];B[oj];W[ps];B[ls];W[sm];B[rl];W[sl];B[sk];W[sn];
B[rk];W[ld];B[lh];W[kg];B[ni];W[jh];B[mc];W[na];B[ma];W[fj];B[oa];W[am];
B[bn];W[al];B[el];W[dm];B[gc];W[gb];B[ef];W[an];B[bp];W[ao];B[bq];W[pa];
B[na];W[le];B[kf];W[kd];B[jd];W[ja];B[ka];W[ia];B[md];W[sa];B[df];W[ee];
B[he];W[ap];B[aq];W[ho];B[go];W[se];B[sg];W[ca];B[mi];W[hm];B[ms];W[bd];
B[];W[])

Image

It's amazing how well MCTS works, with Win-Loss information, knowledge in the tree and in the pseudo-random simulations, automatic acquisition of patterns, etc. At this pace, I guess in 5 years MCTS based Go bots will be at good pro Dan level on 19x19. They are already there on 9x9.