Hakkapeliitta TCEC, dev relase

[velmarin]

Binary TCEC, for those who like
try everything.

Author Mikko Aarnos
https://github.com/mAarnos/Hakkapeliitta/tree/dev
folder /bin Added binary for TCEC

[Graham Banks]

Really looking forward to the next official release.

[neelbasant]

Binary TCEC, for those who like
try everything.

Author Mikko Aarnos
https://github.com/mAarnos/Hakkapeliitta/tree/dev
folder /bin Added binary for TCEC

When will the MP version be out ?

[Ozymandias]

No work has been done since last summer, this is probably the mysterious 1-core engine of "considerable strength", mentioned earlier as yet another teaser.

[MikeB]

Binary TCEC, for those who like
try everything.

Author Mikko Aarnos
https://github.com/mAarnos/Hakkapeliitta/tree/dev
folder /bin Added binary for TCEC

very nice - for those who have a Mac, enjoy...

Pseudo Bench:

Code: Select all

Hakkapeliitta 3.22 (C) 2013-2015 Mikko Aarnos
Detected 24 CPU core(s)
Detected hardware POPCNT
uci 
id name Hakkapeliitta 3.22
id author Mikko Aarnos
option name Hash type spin default 32 min 1 max 65536
option name Pawn Hash type spin default 4 min 1 max 8192
option name Clear Hash type button
option name Contempt type spin default 0 min -75 max 75
option name Ponder type check default true
option name SyzygyPath type string default <empty>
option name SyzygyProbeDepth type spin default 1 min 1 max 100
option name SyzygyProbeLimit type spin default 6 min 0 max 6
option name Syzygy50MoveRule type check default true
uciok
go depth 20
...
info depth 20 currmove f2f4 currmovenumber 20
info depth 20 seldepth 20 score cp 27 time 2793 nodes 11693242 nps 4185000 tbhits 0 hashfull 767 pv e2e4 d7d5 e4d5 g8f6 d2d4 c7c6 b1c3 c6d5 c1f4 h7h5 g1f3 a7a6 h2h4 b8c6 f3e5 c8g4 f2f3 g4f5 e5c6 b7c6 
info time 2793 nodes 11693242 nps 4185000 tbhits 0 hashfull 767
bestmove e2e4 ponder d7d5

[Ralf Müller]

It's developed for TCEC, but doesn't support MP or plays there a MP-Version?

[JJJ]

I m testing it against Hakkapeliitta master in 10 sec + 50ms.

So far, I see +80 elo.

[IanO]

Thanks for the Mac build, works pretty well for me, giving me some crazy entertaining games! (If any engine needed a gambit opening book, it is Hakkepeliitta!)

It doesn't work well with short increments, though. I'm finding that it loses half of its drawn endgames on time when the increment is only 1 second. It makes me wonder how you are getting it to work in your super-short TC tournament. Here is a profile of its thinking times for the last ten moves in one such 5+1 game:

1.0 1.0 1.0 1.1 1.0 1.2 1.3 1.1 2.2 1.2

Hakkepeliitta needs to make its time allocation more conservative, apparently.

[Bloodbane]

Binary TCEC, for those who like
try everything.

Author Mikko Aarnos
https://github.com/mAarnos/Hakkapeliitta/tree/dev
folder /bin Added binary for TCEC

When will the MP version be out ?

I am completely tied up for the next couple of weeks but after that I intend to complete the MP version, no matter what.

Thanks for the Mac build, works pretty well for me, giving me some crazy entertaining games! (If any engine needed a gambit opening book, it is Hakkepeliitta!)

It doesn't work well with short increments, though. I'm finding that it loses half of its drawn endgames on time when the increment is only 1 second. It makes me wonder how you are getting it to work in your super-short TC tournament. Here is a profile of its thinking times for the last ten moves in one such 5+1 game:

1.0 1.0 1.0 1.1 1.0 1.2 1.3 1.1 2.2 1.2

Hakkepeliitta needs to make its time allocation more conservative, apparently.

I haven't had any time losses at short TC, or any time control for that matter. What you are experiencing is a bug which I have been trying to find for two years. People have reported this behaviour a few times, but I have never been able to replicate it. If you could send me a few PGNs of such games, or some way to cause such timeouts it would be much appreciated.

[Bloodbane]

Aaaaand right after I write that I am able to replicate the bug. Typical. Basically, it is a combination of high search depths, the hash table working as it should and checking for time only after 10000 nodes have been searched, and not the time allocation. The fix seems to be rather simple as well. I'll add a new binary soon.