Persistant hash concept

[Hood]

Hello,

why that concept was broken ?
Any ideas.

Rgds
Hood

[Uly]

-It had too many bugs and no time for fixing them
-The code no longer worked (it was broken just like time management)
-Low priority

[Hood]

It does not explain why other engines are not using it?

[Uly]

Ah. Well, Vas coined the term, but it's confusing because it could also mean "preserve hash". I thought you were referring specifically to Rybka 3 Persistent Hash.

I'd like to call it "position learning", to differentiate it from "result learning" and "book learning".

Well, one thing is that it doesn't really help on games, unless you play the same openings over and over.

Also, the rating lists always have any kind of learning off, so implementing it will not make the engine look better anywhere.

The engines that have position learning are:

Shredder (gold standard)
Rybka 3 (super-buggy)
Spike (not really helpful)
Hiarcs (have never gotten it to work)

So I think it's something really hard to implement. People at the Friendly Stockfish thread said a dirty implementation would be easy, but nobody has suggested code yet.