Search found 119 matches

by Tony P.
Mon Jul 13, 2020 9:19 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Data structure choice for TT
Replies: 27
Views: 1408

Re: Data structure choice for TT

hgm wrote:
Mon Jul 13, 2020 9:08 pm
So accesses occur in bursts. This can easily make a shallow entry temporarily pass a deep entry in your scheme, leading to overwriting of the deep entry.
Ah, so that's why Leela is using the LRU cache replacement policy instead of the LFU one. Thanks for enlightening me again! :idea:
by Tony P.
Mon Jul 13, 2020 3:52 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Data structure choice for TT
Replies: 27
Views: 1408

Re: Data structure choice for TT

Tony P. wrote:
Mon Jul 13, 2020 3:05 pm
What are some examples of fast and strong engines with non-Zobrist hashes?
Oops, Leela is an obvious example :oops: (of a strong one). Its hash function, based on a few additions, multiplications, bit shifts and concatenations, doesn't seem slow.
by Tony P.
Mon Jul 13, 2020 3:38 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Mind your language
Replies: 16
Views: 1032

Re: Mind your language

The Youtube video used various sources of data, outlined in its description. It didn't use TIOBE alone, fortunately.
by Tony P.
Mon Jul 13, 2020 3:23 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Data structure choice for TT
Replies: 27
Views: 1408

Re: Data structure choice for TT

I don't know what you mean by N-independent Oops, sorry for the confusion. Here's the definition in Wikipedia. Actually, the N-independence of a family of hash functions guarantees nothing about the performance of its particular member with fixed parameters :(, and it assumes nothing about the dist...
by Tony P.
Mon Jul 13, 2020 3:05 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Data structure choice for TT
Replies: 27
Views: 1408

Re: Data structure choice for TT

Yes, that's exactly why bucketing is needed. However, I can't easily construct a 3-collision (3 positions with the same hash) based on a 2-collision. Surely, a 3-collision may spawn further ones in the subtrees because if h(pos1)=h(pos2)=h(pos3), then h(pos1 xor move)=h(pos2 xor move)=h(pos3 xor mov...
by Tony P.
Mon Jul 13, 2020 6:27 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Data structure choice for TT
Replies: 27
Views: 1408

Re: Data structure choice for TT

And the distribution of the importance of the entries in a typical game tree is such that you will almost always find one of the most unimportant entries in a set of four, which is then suitable for overwriting. Thanks a lot for this insight! My mistake was that I was overestimating the ratio of th...
by Tony P.
Mon Jul 13, 2020 5:30 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Mind your language
Replies: 16
Views: 1032

Re: Mind your language

Hmm, you've prompted me to look up their methodology, and I'm not sure if it even discerns C++ related search hits from C related ones correctly, as that would require all the 25 polled engines to be sensitive to the '++' suffix.
by Tony P.
Mon Jul 13, 2020 3:51 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Mind your language
Replies: 16
Views: 1032

Re: Mind your language

Brace yourselves: Rust is coming :P I hope it's even going to overtake Scratch for the 17th place in the TIOBE index next month :lol: (you've read right - Scratch, a language aimed at children, is in the top 20 of that particular index - that's how tech-savvy the kids are becoming).
by Tony P.
Fri Jul 10, 2020 4:41 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Data structure choice for TT
Replies: 27
Views: 1408

Data structure choice for TT

Let's share thoughts on unusual data structures that may further improve the utility of a TT. To be clear, I'm not writing an engine, just theorycrafting yet :mrgreen: I've been considering what would happen if an engine had expensive eval or move ordering and visited 10-100 times fewer nps than the...
by Tony P.
Tue Jul 07, 2020 5:58 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Intel AMX with TMUL on Xeon Sapphire Rapids (2021?)
Replies: 6
Views: 440

Re: Intel AMX with TMUL on Xeon Sapphire Rapids (2021?)

Thanks for sharing the exciting news! Now I've started dreaming of competitions held on a single CPU server again to end the debate about the hardware bias. It will be interesting to see how long it will take anyone to beat SF (both classical and EUNN) on its home turf.