PV Search and Transposition Table
Posted: Thu Dec 20, 2012 1:30 am
I have worked out the kinks in my TT implementation and it seems to be working fine, until ...
My last version of the AB search (without TT) I had implemented PV Search, which also worked. I based this from CPW (http://chessprogramming.wikispaces.com/ ... ion+Search)
I created a new AB search which uses the TT and PV Search and bad moves happened. I tried to implement a separate zwSearch like in CPW but that was worse. I did not have the zwSearch examine the TT.
From debugging, it is clear to me that a TT entry is being used somehow when the smaller window is used but I am having trouble identifying where and why.
So before I dig deeper into what the TT is doing, I decided to modify the search process. Instead of of the search calling iteself with alpha-1,alpha I have it call the previous version of AB search which does not have a TT check. Yes, it works but I do not believe it is the correct path.
Is the a correct implementation or should the PVSearch with the window of alpha-1,alpha still check the TT for entries?
Also, in the CPW link, the zwSearch shows this: score = -zwSearch(1-beta, depth - 1). I am expecting 1-beta as a type-o.
Thank you
My last version of the AB search (without TT) I had implemented PV Search, which also worked. I based this from CPW (http://chessprogramming.wikispaces.com/ ... ion+Search)
I created a new AB search which uses the TT and PV Search and bad moves happened. I tried to implement a separate zwSearch like in CPW but that was worse. I did not have the zwSearch examine the TT.
From debugging, it is clear to me that a TT entry is being used somehow when the smaller window is used but I am having trouble identifying where and why.
So before I dig deeper into what the TT is doing, I decided to modify the search process. Instead of of the search calling iteself with alpha-1,alpha I have it call the previous version of AB search which does not have a TT check. Yes, it works but I do not believe it is the correct path.
Is the a correct implementation or should the PVSearch with the window of alpha-1,alpha still check the TT for entries?
Also, in the CPW link, the zwSearch shows this: score = -zwSearch(1-beta, depth - 1). I am expecting 1-beta as a type-o.
Thank you