I would like to know which ones are the Stockfish derivatives that allow you to save the hash table to disk. Ideally, I am looking for the most reecnt stockfish nnue version clones.
Thanks in advance.
Most recent implementationa of Persistent Hash Table
Moderators: hgm, Rebel, chrisw
-
- Posts: 41
- Joined: Sun Apr 07, 2019 1:10 am
- Full name: Giovanni Lavorgna
-
- Posts: 6
- Joined: Mon Jun 28, 2021 4:04 pm
- Full name: Hugh Whelan
Re: Most recent implementationa of Persistent Hash Table
Although it's base is Cfish, this clone (which uses its own net) allows a persistent transposition table: https://github.com/Sopel97/FatTitz This engine shows similar strength to Stockfish (see
See also: Brainlearn (https://github.com/amchess/BrainLearn)
See also: Brainlearn (https://github.com/amchess/BrainLearn)
-
- Posts: 41
- Joined: Sun Apr 07, 2019 1:10 am
- Full name: Giovanni Lavorgna
Re: Most recent implementationa of Persistent Hash Table
Thank you! BTW, nice projects.
Also, I was intrigued by the Persistent Learning Algoritm in Brain learn. Do you know how it differs from the classical hash table saving in FatTitz?
Thanks again for your help and assistance,
Also, I was intrigued by the Persistent Learning Algoritm in Brain learn. Do you know how it differs from the classical hash table saving in FatTitz?
Thanks again for your help and assistance,
-
- Posts: 389
- Joined: Tue Oct 08, 2019 11:39 pm
- Full name: Tomasz Sobczyk
Re: Most recent implementationa of Persistent Hash Table
Brainlearn's "persistent learning algorithm" is more intelligent in what it stores, has no size limit, and works all the time. FatTitz's is just a way to store the TT, so has finite size, a simple replace policy, and is reset between games. I'd say Brainlearn's approach is better if you want to analyze mostly opening positions in large quantities and have the engine "learn" seemlessly; FatTitz better if you want to preserve the hash table of the current search of a specific position and resume it later (though it doesn't preserve everything, for example histories or low ply TT entries)
dangi12012 wrote:No one wants to touch anything you have posted. That proves you now have negative reputations since everyone knows already you are a forum troll.
Maybe you copied your stockfish commits from someone else too?
I will look into that.
-
- Posts: 41
- Joined: Sun Apr 07, 2019 1:10 am
- Full name: Giovanni Lavorgna
Re: Most recent implementationa of Persistent Hash Table
Thanks for your explanation. So, if I read you correctly, you can achieve with the BrainLearn Persistent learning everything you can do with the FatTitz hash saving capability. I mean in terms of a multi-session position analysis. Moreover, you don't need to have a separate hash table for each root position, making its use much less demanding in terms of disk-space.Brainlearn's "persistent learning algorithm" is more intelligent in what it stores, has no size limit, and works all the time. FatTitz's is just a way to store the TT, so has finite size, a simple replace policy, and is reset between games. I'd say Brainlearn's approach is better if you want to analyze mostly opening positions in large quantities and have the engine "learn" seemlessly; FatTitz better if you want to preserve the hash table of the current search of a specific position and resume it later (though it doesn't preserve everything, for example histories or low ply TT entries)
Thanks again for eleborating on this important topic.