Page 13 of 23

Re: TCEC 10

Posted: Mon Nov 20, 2017 2:46 pm
by schack
Are the changes in 6.03 limited to this stalemate issue?

Re: TCEC 10

Posted: Mon Nov 20, 2017 4:59 pm
by cc2150dx
Thanks for the update, Robert!

Re: TCEC 10

Posted: Mon Nov 20, 2017 5:19 pm
by Houdini
schack wrote:Are the changes in 6.03 limited to this stalemate issue?
Yes, it's a minor correction.
The fact that this problem has remained undetected for over a year suggests that it's not occurring very frequently.

Robert

Re: TCEC 10

Posted: Mon Nov 20, 2017 5:43 pm
by shrapnel
schack wrote:Are the changes in 6.03 limited to this stalemate issue?
Let's hope for a "Butterfly Effect " :)

Re: TCEC 10

Posted: Mon Nov 20, 2017 6:39 pm
by Nordlandia
shrapnel wrote:
schack wrote:Are the changes in 6.03 limited to this stalemate issue?
Let's hope for a "Butterfly Effect " :)
+1 🦋

Re: TCEC 10

Posted: Mon Nov 20, 2017 9:07 pm
by styx
Houdini wrote: Houdini 6.03 corrects a stalemate bug which was reported to me this weekend. It occurs when there are white pawn captures available, for example this position:
[d]8/3R1pk1/5p1p/5PbK/B2p2PP/3P4/1r6/8 w - - 0 1

Try this position with the previous Houdini 6, set multi-PV to 4. You will get some surprising results :).

The error was already present in Houdini 5, but apparently nobody has ever noticed it.

Robert
out of curiosity: what was the engine output? non-houdini user here

Re: TCEC 10

Posted: Mon Nov 20, 2017 9:59 pm
by Geonerd
egiovannotti wrote:
Dan Cooper wrote:
Jouni wrote:Stockfish is first engine, which goes WHOLE season UNBEATEN!! No word on that in chessdom page?
Let's compare.

Komodo goes undefeated in the chess.com speed championship but misses qualifying for the final by .5 points. This goes unreported in the chess.com website. No one cares.

Stockfish goes undefeated in TCEC but misses qualifying for the final by .5 points. This goes unreported on the chessdom.com website. People freak out.
+1
The contempt that made Stockfish remain unbeaten was the same that did not allow him to add the necessary points. Surely with the right contempt Komodo and Houdini would also be unbeaten.
There's no doubt a bit of Game Theory and second-guessing going on here between the various programmers. Setting contempt too high, relative to the other engines, can cost unnecessary losses, while setting it too low costs wins. Ya pay yer money and you take yer chances! :P

Re: TCEC 10

Posted: Tue Nov 21, 2017 3:40 am
by Geonerd
Quite an interesting game (#2, move 54) unfolding now. Wonderfully speculative, active play by H has given him space and very active pieces, but K is 2 solid pawns up. If K can untangle...

Re: TCEC 10

Posted: Tue Nov 21, 2017 2:42 pm
by Norm Pollock
I looked to see if the Fizbo fiasco had a ripple effect on the results in round 2.

The beneficiary Booot scored the same vs the Big Three (0.5/2.0). No effect on the results of round 2.

Re: TCEC 10

Posted: Tue Nov 21, 2017 2:59 pm
by Norm Pollock
correction:

the results were
Booot 3.0/12
K-S-H 3.0/4

same conclusion- Booot did not influence eventual outcome.