Centaur vs Unassisted Engine

[MonteCarlo]

Dedicated thread for the test proposed in http://talkchess.com/forum3/viewtopic.php?f=2&t=76407.

I'll have a response in a couple hours (maybe longer, depending on how work goes).

Current game state:

[pgn]
[White "jefk"]
[Black "unassisted engine"]

1. d4[/pgn]

[MonteCarlo]

Nf6, no surprise.

[pgn][White "jefk"]
[Black "unassisted engine"]

1. d4 Nf6[/pgn]

[jefk]

ok, thx for the quick reply,
(so we now have 1.d4 Nf6)

Next i probably will play 2.c4, from book (other
option is 2.Nf3 but this imho may have drawbacks)
Also from my experience, i'll wait a little before making
a decision, and similar then before posting the move;
but in this case i'll probably reply already tomorrow.

NB as agreed one or 2 hrs /move for your fast machine,
and i presume you don't use pondering (which would
give you much more drawing chances but also
will use a lot of electricity and i presume you also use
your comp for other purposes than this little game)

PS yes my play is like 'centaur' ; but again to clarify, more
to test centaur play for (ICCF type) [i]correspondence[/i] play,
not like 'Freestyle' Chess (as played for some years until
last year or so) because that had much faster time controls

[MonteCarlo]

Understood on all points!

I will not be pondering. I will be starting the engine from scratch for each move, so the engine will always have maximum 1-2 hours (or less, for things like only-legal-move) of compute for each reply.

I will probably pre-compute some replies in the interest of time, like for 2.c4 here, but I won't ever let it get more than 2 hours of compute for any position, always starting from scratch.

Cheers!

[jefk]

ok, yes it's 2.c4 indeed.
(2.Nf3 would transpose probably anyway after your 2...d5).

As for your comp, well reloading hash may take a few minutes,
there also were engines (i remember houdini were you could
save hashtables, your allowed to do that, of course, if you want !


PS and now i expect 2..e6 (but i'll wait)

[MonteCarlo]

I thought about saving hash tables or just snapping VM including memory, but figured I'd keep things simple.

I appreciate the sporting offer, but I'll just start from scratch each move

Reply to 2.c4 has already been pre-computed, and yes, it is the unsurprising e6.

[pgn][White "jefk"]
[Black "unassisted engine"]

1. d4 Nf6 2.c4 e6[/pgn]

[jefk]

next move is 3.g3 indeed
jef

[MonteCarlo]

3...d5

[pgn][White "jefk"]
[Black "unassisted engine"]

1. d4 Nf6 2.c4 e6 3.g3 d5
[/pgn]

[jefk]

4.Bg2

and then your engine probably will play 4...Bb4+
Precomputed or not

(although that's not absolute certain, if it recognizes
(or already has recognized) the awesome
probable next move, 5.Nd2! (NB not confirming this
yet, i want to be certain).
For your engine besides 4...Bb4+ there are three to four
possible, other decent moves, leading to Catalan (on
which i happened to have done quite some analysis
some 2/3 yrs ago, and btw your engine at one/two hrs
will not find such analysis in such a short time, Nnue or not).

[Milos]

4.Bg2

and then your engine probably will play 4...Bb4+
Precomputed or not

(although that's not absolute certain, if it recognizes
(or already has recognized) the awesome
probable next move, 5.Nd2! (NB not confirming this
yet, i want to be certain).
For your engine besides 4...Bb4+ there are three to four
possible, other decent moves, leading to Catalan (on
which i happened to have done quite some analysis
some 2/3 yrs ago, and btw your engine at one/two hrs
will not find such analysis in such a short time, Nnue or not).

I don't get you CC guys. Just look at last "merged" Cerebellum, it's stronger than 99% of your CC "custom" books. No need to invent the wheel.
And if you really wanna have any chance to beat SF you'd need to quickly move away from a "fishbook", but trying at least to keep some advantage for white. I'd say, impossible task, which you'd be proven very soon.
Btw. giving only 2h to unassisted SF on frankly subpar machine is really not fair. In 10 moves in 50 days typical CC TC, one can play 60 CC games in parallel with unassisted SF. I bet none of you CC pro's plays 60 games at the same time.
I bet with 1 day per move on the machine that has 50+ Mnps using SFdev with Cerebellum Merged up to move 15 in main lines, and then just unassisted SF is virtually unbeatable (unless the other side uses 4k+ cores cluster).