gnugo-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] Metamachine's first game


From: bump
Subject: Re: [gnugo-devel] Metamachine's first game
Date: Wed, 13 Nov 2002 06:14:20 -0800

> Congratulations! I hope this makes it into Gnu Go 3.4 (with GTP, please).

I would think so in some form unless we get something
better. Otherwise the entire file oracle.c is commented out
unless you compile it with --enable-oracle. So even though it
requires *ix to run the oracle there should be no portability issue.

The patch isn't really ready yet. In fact the patch oracle_3_12.1d
 is a little broke.  The load and analyze mode which was the main
thing in oracle_3_12.1c doesn't work, and I got a mysterious
crash in the second game.

I've played 3 games including the one I posted, and only one
finished, but the metamachine seemed ahead in all three games.
In the game that just finished, the metamachine played W and
won by 5.5 points. In the second game where it crashed the
metamachine (black) was ahead by about 10 points at the end.

Perhaps it's stronger than GNU Go but we need a few more games to
say that. 

When playing a game, this version is doing global search. I think
the strength of this idea is going to be in semilocal search, and
this experiment, though interesting, is not what I'm really
trying to get at here. Dreihirn experiments show that you can't
really make a strong engine just selecting among GNU Go's top 3
moves which is what this program is doing.

But assuming we want to pursue this, there is a way to make it
faster, which is to use the score reported by top_moves to prune
the tree rather than just taking the top 3 moves or so. We'd
still search a tree if there are several big moves and this
might suffice to prevent some mistakes. I think can make a
version of this that plays fast enough for tournament play, then
run a gtp series and see if it's really stronger.

> It would certainly be interesting to have a 20-game match between GNU Go 
> with and without the metamachine, without time limits (and perhaps different 
> board sizes?). What was the CPU and  speed  when you tested it?

The game posted was played on a 1.2 Gh P4. 

Dan

(;GM[1]FF[4]
RU[Japanese]SZ[19]HA[0]KM[5.5]
PW[GNU Go 3.3.12 pre 2]
PB[Metamachine]
GN[White (W) vs. Black (B)]
DT[2002-11-12]
SY[Cgoban 1.9.12]TM[-];B[qd];W[dp];B[dc];W[pp];B[cf];W[oc];B[ld];W[of]
;B[oe];W[ne];B[nq];W[qb];B[cn];W[pn];B[ck];W[qf];B[kq];W[pe];B[ic];
W[fp];B[ql];W[nn];B[qi];W[qn];B[pr];W[hp];B[qq];W[co];B[dm];W[kg];
B[ef];W[ig];B[gf];W[ol];B[hr];W[dn];B[bn];W[en];B[jq];W[fl];B[ek];
W[hi];B[jn];W[hf];B[ge];W[qp];B[gh];W[hh];B[jl];W[gn];B[rg];W[rf];
B[qh];W[gr];B[hq];W[gq];B[mc];W[bo];B[oh];W[an];B[bm];W[hs];B[rm];
W[js];B[rp];W[ro];B[ip];W[rq];B[ho];W[rr];B[sf];W[se];B[sg];W[rd];
B[gi];W[jd];B[jc];W[kd];B[kc];W[am];B[al];W[ao];B[cl];W[nb];B[le];
W[lb];B[je];W[kb];B[jb];W[ki];B[mm];W[nj];B[mf];W[mi];B[mn];W[pj];
B[qj];W[np];B[mq];W[pq];B[oq];W[qr];B[ng];W[or];B[nr];W[ps];B[mo];
W[gp];B[pk];W[ll];B[no];W[op];B[ir];W[hn];B[io];W[pi];B[ph];W[ok];
B[ka];W[la];B[mb];W[ma];B[nm];W[pl];B[ja];W[qk];B[rk];W[em];B[fj];
W[kk];B[el];W[hk];B[gk];W[gl];B[jk];W[jj];B[hj];W[ij];B[ns];W[im];
B[km];W[ik];B[qm];W[kl];B[jm];W[on];B[nl];W[nk];B[nc];W[pa];B[os];
W[nd];B[lc];W[na];B[rn];W[sn];B[sm];W[so];B[ml];W[mk];B[go];W[fo];
B[lg];W[kf];B[hd];W[ie];B[ke];W[id];B[he];W[if];B[mh];W[lh];B[mp];
W[qg];B[si];W[lm];B[ln];W[il];B[pr];W[qs];B[oo];W[oi];B[jf];W[jg];
B[gg];W[po];B[pg];W[gj]
)

This is the game that just finished:

(;GM[1]FF[4]
RU[Japanese]SZ[19]HA[0]KM[5.5]
PW[Metamachine]
PB[GNU Go 3.3.12 pre 2]
GN[White (W) vs. Black (B)]
DT[2002-11-12]
SY[Cgoban
1.9.12]TM[-];B[qd];W[ec];B[ep];W[oc];B[pe];W[lc];B[nd];W[cd]
;B[op];W[qq];B[cp];W[gq];B[dj];W[ch];B[qj];W[jq];B[eq];W[bj];B[le];
W[ob];B[je];W[rc];B[lq];W[po];B[qm];W[ic];B[rd];W[ge];B[oj];W[eh];
B[cm];W[ej];B[pr];W[pm];B[pl];W[qr];B[qc];W[qn];B[om];W[ih];B[id];
W[hc];B[pn];W[rn];B[oo];W[pp];B[oq];W[ki];B[ek];W[mi];B[rm];W[fj];
B[fk];W[lh];B[oh];W[sm];B[sl];W[sn];B[gj];W[ll];B[rl];W[fr];B[er];
W[go];B[fn];W[di];B[jc];W[jb];B[jd];W[kc];B[hl];W[gi];B[fi];W[ei];
B[cj];W[gh];B[bi];W[bh];B[bk];W[kr];B[lr];W[es];B[ds];W[fs];B[cr];
W[ls];B[ah];W[ag];B[ai];W[bg];B[ms];W[hj];B[qs];W[rs];B[ps];W[pq];
B[nr];W[jo];B[mg];W[lo];B[jl];W[rr];B[gk];W[kq];B[lp];W[kp];B[mo];
W[kk];B[nc];W[nb];B[hk];W[ij];B[rb];W[km];B[ln];W[kn];B[lm];W[jk];
B[fo];W[ml];B[gp];W[hp];B[fp];W[ho];B[nm];W[nh];B[ng];W[im];B[il];
W[oi];B[pi];W[kl];B[jm];W[hm];B[fq];W[hr];B[gm];W[od];B[oe];W[mm];
B[mn];W[ni];B[hn];W[mc];B[in];W[md];B[ph];W[ne];B[me];W[nf];B[lg];
W[mf];B[kg];W[lf];B[kf];W[of];B[og];W[pf];B[qf];W[hd];B[ig];W[hg];
B[if];W[hf];B[jh];W[ji];B[pb];W[kh];B[jg];W[io];B[ks];W[nk];B[js];
W[ok];B[pk];W[nj];B[ir];W[iq];B[pg];W[nd];B[kd];W[hs];B[ld];W[pa];
B[qa];W[oa];B[jr];W[pc];B[pd];W[pj];B[qb];W[ci];B[aj];W[gn];B[ko];
W[is];B[ls];W[jn];B[lo];W[ik];B[oj];W[ak];B[bj];W[mh];B[pj];W[nl];
B[ol];W[he];B[ie]
)




reply via email to

[Prev in Thread] Current Thread [Next in Thread]