Search found 748 matches

by smatovic
Mon Mar 12, 2012 12:34 pm
Forum: Computer Chess Club: General Topics
Topic: Zeta OpenCL Chess - Alpha Release 0960
Replies: 15
Views: 2508

Re: Zeta OpenCL Chess - Alpha Release 0960

I type: new and then go, it seem windows hangs, after that a black windows comes and 2 seconds later windows goes normal and windows says that there was GPU driver problem
Maybe i will setup a Windows-Box and test it with the next Beta Version...

--
Srdja
by smatovic
Sun Mar 11, 2012 8:54 pm
Forum: Computer Chess Club: General Topics
Topic: Zeta OpenCL Chess - Alpha Release 0960
Replies: 15
Views: 2508

Re: Zeta OpenCL Chess - Alpha Release 0960

I run zeta-cl-64-ja.exe I type go or test and it just output move a1a1 nothing more and this is the zeta_cl.debug content:
hm, strange, it should print some errors if it couldnt init the OpenCL device...

ah, before the go or test command you need to type "new" to set up the board!

--
Srdja
by smatovic
Sun Mar 11, 2012 2:30 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: MCTS without random playout?
Replies: 26
Views: 5107

Re: MCTS without random playout?

correct me if i am wrong, but afaik in CNS you have to store at every expanded node the C-Numbers for every Score....so 16 Bit means you need to store 65536 C Numbers at every node...this could work with 10k expanded nodes, but not with 100k. ##edit### Oh wait maybe i should tell something what i di...
by smatovic
Sun Mar 11, 2012 2:13 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: MCTS without random playout?
Replies: 26
Views: 5107

Re: MCTS without random playout?

Additional another thing i find not so interesting is the 2 ply fullwidth searches introduced somewhere. If you simply don't know how to parallellize your software, then it won't help to 'rape' your CNS algorithm by doing 2 ply searches everywhere near the leaves. I agree. Basically CNS is most int...
by smatovic
Sun Mar 11, 2012 12:54 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: MCTS without random playout?
Replies: 26
Views: 5107

Re: MCTS without random playout?

Hi Vincent,

i ve read some papers about CNS and CCNS,
these algorihtms look very promising,
especially CCNS with its alphabeta bounds for the leaf node search.

Unfortunately the CCNS is not trivial....will need some time to get into this...

--
Srdja
by smatovic
Sun Mar 11, 2012 11:22 am
Forum: Computer Chess Club: General Topics
Topic: Zeta OpenCL Chess - Alpha Release 0960
Replies: 15
Views: 2508

Re: Zeta OpenCL Chess - Alpha Release 0960

Interesting project. Are you planning to release a Windows version any time soon? Jim made an windows 64 bit compile but it crashed on his system, i added the windows exe into the bin folder on github: https://github.com/smatovic/Zeta Be sure to place the config.cfg and zeta.cl files in the same fo...
by smatovic
Sat Mar 10, 2012 3:14 pm
Forum: Computer Chess Club: General Topics
Topic: Zeta OpenCL Chess - Alpha Release 0960
Replies: 15
Views: 2508

Zeta OpenCL Chess - Alpha Release 0960

Heyho, here my new OpenCL Chess approach: https://github.com/smatovic/Zeta/tree/zeta_cl_0960 Same Version runs on AMD and NV, needs settings in config.cfg, currently only Linux 64 supported. Changes: - Randomized Best First Minimax Algorithm with Qsearch at Leaf Nodes - fixes in Move Generator Issue...
by smatovic
Sat Mar 03, 2012 4:41 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: MCTS without random playout?
Replies: 26
Views: 5107

Re: MCTS without random playout?

thanks,

i still play with these values....interesting how the engine changes its play...

Currently i got with this formula the best results:

Code: Select all

uct_score =  child.wins / child.visits  + 0.01 *  sqrt( (2000 * log(parent.visits))  / child.visits ) ; 
--
Srdja
by smatovic
Sat Mar 03, 2012 4:32 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Stockfish port to C#
Replies: 14
Views: 2360

Re: Stockfish port to C#

Any ideas on work arounds for template and pointer arithmetic would be helpful.
...hope you dont copy hole boards to do/undo a move...

--
Srdja
by smatovic
Fri Mar 02, 2012 7:15 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: MCTS without random playout?
Replies: 26
Views: 5107

Re: MCTS without random playout?

Hi Antonio, i tried to implement this win count based best first minimax and have to say that my version can not compete with an AB depth 4 searcher. But it plays a lot of better than an plain MC or UCT approach...so may i ask what uct score formula you use? my looks like this: uct_score = child.win...