Square Occupation vs Mobility / Square Control

[mcostalba]

First, your "two moves away" is simply wrong. When I enter q-search, I do a stand pat score, then try a capture, do a stand-pat, try a capture, etc. Always one intervening move except when I back up a couple of plies and start going forward again.

I was talking of the case where after stand pat you return immediately due to a cut-off. In this case the sequence is

do_move
eval with mobility caching
return
undo_move
do_move // the next one
eval

Now the two eval are 2 moves away (one undo and one do_move of the next one).

I don't remember the 99% hit rate quote since I have never measured it,

You talked about calculating again 1 time out of 100. It means 99% cache hit here.

[bob]

First, your "two moves away" is simply wrong. When I enter q-search, I do a stand pat score, then try a capture, do a stand-pat, try a capture, etc. Always one intervening move except when I back up a couple of plies and start going forward again.

I was talking of the case where after stand pat you return immediately due to a cut-off. In this case the sequence is

do_move
eval with mobility caching
return
undo_move
do_move // the next one
eval

Now the two eval are 2 moves away (one undo and one do_move of the next one).

OK. But how many ranks/files/diagonals can you change between the two eval calls? There are possibly 4 rooks, 4 bishops, 2 queens (knights, etc are irrelevant as far as this discussion goes since they don't have the sliding ability and extensive mobility). A single move will only occasionally affect more than one of those.

I don't remember the 99% hit rate quote since I have never measured it,

You talked about calculating again 1 time out of 100. It means 99% cache hit here.

[swami]

Anyways, my chess sense tells me that square occupation is a simplified way of representing more complex issues like square control or mobility. As a player I am more interested in square control and mobility, in that sense I see dealing with square occupation as something of a compromise to make programming easier. PST's make the programmatic assigning of values to pieces much easier, but it becomes more difficult to come up with meaningful values

While you're at it, you might want to try this test suite concerning "Square Vacancy" - It would be more fun if you solved it on your own and compared the scores with your engine.

http://sites.google.com/site/strategict ... re-vacancy