Search found 3459 matches

by Daniel Shawul
Sun Feb 23, 2020 3:35 pm
Forum: Computer Chess Club: Tournaments and Matches
Topic: Scorpio 3.0.7 NN 42850
Replies: 7
Views: 1706

Re: Scorpio 3.0.7 NN 42850

Hi Werner, Thanks for the test. I just wanted to be sure T40 nets work with INT8. The default network with medium Ender is used for CCCC. @Steppenwolf, the installation is pretty easy. All you have to do is download the install script for windows (install.bat) or for linux (install.sh). Run that to ...
by Daniel Shawul
Sat Feb 22, 2020 12:22 am
Forum: Computer Chess Club: Tournaments and Matches
Topic: Scorpio 3.0.7 NN 42850
Replies: 7
Views: 1706

Re: Scorpio 3.0.7 NN 42850

Hi Werner, What changes did you make? My guess is a fresh install was needed instead of editing scorpio.ini. Anyway, I am glad that T40 nets work with INT8 now -- there was some major bug with INT8 calibration using EPD files. My tests show that Scorpio with T40 net is slightly weaker than with Lc0 ...
by Daniel Shawul
Sun Feb 16, 2020 4:36 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Engines playing Musketeer Chess, good price
Replies: 142
Views: 19870

Re: Engines playing Musketeer Chess, good price

Isn't losing the gating piece when king is in check already punishment enough? It would be simpler to just allow gating every time the piece in front moves, and then let the consequences (e.g. gating piece being immediately captured being one of them) prevail. If the king is incheck, obviously the g...
by Daniel Shawul
Sun Feb 16, 2020 2:07 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Engines playing Musketeer Chess, good price
Replies: 142
Views: 19870

Re: Engines playing Musketeer Chess, good price

For the record, nebiyu misses this rule too. e9f9 is a bad move but I don't see why it is not allowed? Gating always occurs the first time the piece in front moves, right? The idea is something similar to castling rights where if the king is under attack and is forced to move, it loses the right to ...
by Daniel Shawul
Mon Feb 10, 2020 5:49 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: enjoy the paper - :-)
Replies: 8
Views: 2238

Re: enjoy the paper - :-)

Do you live in a cave so as to discover the existence of this paper in 2020 ?!
I am sure people here have talked about it the day after it was published on arxiv ...
by Daniel Shawul
Thu Feb 06, 2020 1:44 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: hash collisions
Replies: 370
Views: 29055

Re: hash collisions

You can actually catch collisions with extra 64 bits stored in TT entries. This will be another hashkey computed using a different hashing formula. The chances of two positions colliding twice when using two different hashing formulas is almost non-existent. I wonder if using two 64-bit zobrist hash...
by Daniel Shawul
Fri Jan 31, 2020 4:39 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: hash collisions
Replies: 370
Views: 29055

Re: hash collisions

There are two different things: 1. The possibility that distinct positions produce the same hash. 2. The possibility that two threads compete, one writing the hash entry while another is reading the same entry. The lockless hash method prevents 2. I don't think it prevents 1 or resolves the possibl...
by Daniel Shawul
Fri Jan 31, 2020 2:16 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: hash collisions
Replies: 370
Views: 29055

Re: hash collisions

This set of fixes appears to have removed the issues - I have a lot of asserts I can turn on and now can run with the artificially high collision rate and full debugging on, and there is no crash or assertion hit. But any kind of issue from moves the hash was already very rare. Arasan competed in t...
by Daniel Shawul
Wed Jan 29, 2020 1:18 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: hash collisions
Replies: 370
Views: 29055

Re: hash collisions

32-bit is not enough for me -- it crashes with a single thread search in less than 10 seconds and it is a real hash collision too, but maybe I used a shitty RNG for generating my hashes. I used to do legality checking before switching to relying on lockless hash, but I guess it is time turn legality...
by Daniel Shawul
Tue Jan 28, 2020 4:38 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: hash collisions
Replies: 370
Views: 29055

Re: hash collisions

I think lockless hashing was pretty safe at modest number of threads most people have removed the need for legality checking of the hash move. Now it may be necessary to bring that back in.