Uri Blass wrote:Another option is that rybka does some selective search and does not trust the score of the selective search so it does some type of average between the score and the score after selective search and I guess that people can find what happens by looking at the code.
This idea seems very strange...
Nothing strange with this idea
I think that this is exactly what strong human players do in chess.
They may think not in term of numbers but they can still prefer seeing mate based on selective search relative to getting a small positional advantage but not evaluate the mate that they see based on selective search as mate and if they can see simpler forced win of a bishop they are going to prefer to win without risks
There is a bonus in the eval that I did for forks, but it's only something like half a pawn or so. Vas did tell me that the program picks up some tactics but not others, so I guess this is one that it does pick up. This at least explains some of the huge gap at one ply.
Uri Blass wrote:I think that you can look at the pv of robbolito at 1 ply and learn about the moves that robbolito extends at this depth.
Sometimes the pv is more than one ply and not because of checks or captures.
Uri
It's difficult to tell, because the hash table apparently has a large effect on what PV comes back. Maybe it's not so bad at 1 ply but if you do a deep search, then a 1 ply search again the PV changes.
Uri Blass wrote:I think that you can look at the pv of robbolito at 1 ply and learn about the moves that robbolito extends at this depth.
Sometimes the pv is more than one ply and not because of checks or captures.
Uri
It's difficult to tell, because the hash table apparently has a large effect on what PV comes back. Maybe it's not so bad at 1 ply but if you do a deep search, then a 1 ply search again the PV changes.
You can simply start from clear hash when you look at the pv at depth 1.
Uri Blass wrote:I think that you can look at the pv of robbolito at 1 ply and learn about the moves that robbolito extends at this depth.
Sometimes the pv is more than one ply and not because of checks or captures.
Uri
It's difficult to tell, because the hash table apparently has a large effect on what PV comes back. Maybe it's not so bad at 1 ply but if you do a deep search, then a 1 ply search again the PV changes.
You can simply start from clear hash when you look at the pv at depth 1.
Uri
Since robbolito comes with source code, nobody has to guess. We can just trace what it is doing in a debugger.
