Mobility iin evaluation functions- how much is it worth?
Tom King
All,
A while back I remember reading (maybe here, maybe ICCA journal) about
the value of having a direct measure of mobility in an evaluation
function of a chess program. I think I read that Schaeffer had estimated
that the adding a measure of mobility to an evaluation function was
worth "about a ply". In other words a chess program with a mobility term
in its evaluation function, and searching to depth N, would score quite
well against the same program, without a mobility term searching to
depth N+1.
Anyone got any views on this? How much is mobility worth in terms? Does
"about a ply" still hold at "interesting" depths that PC programs can
now reach, say depth 9 or 10?
I ask because I've recently been playing with a mobility term in my
chess program, Francesca. I'm not 100% convinced that it's worth "about
a ply", but Francesca's playing more interesting chess with the mobility
term.
--
Tom King
Robert Hyatt
Tom King (t...@hatbulb.demon.co.uk) wrote:
: All,
Opinion only of course. But here's what I personally believe: Mobility
is simply *a* scoring term you can use. I don't believe that maximizing
mobility results in a won game. I do believe that winning a game will
result in high mobility, but it's an issue of cause and effect. I think
mobility is an effect of playing well. Not that if you maximize mobility
you will be playing well as a result.
Second, I use mobility in bishop scoring in Crafty. I have used it in all
piece scoring but did not like the result. I'm in a fortunate position in
that mobility is basically "free" in crafty if I want it. I think that
other eval terms can make this less important and replace mobility. For
example, rook on open file tends to increase the mobility, but it is a
better "quality" of mobility.
Most GM players seem to talk about "quality" mobility, which is more difficult
to explain.
In any case, if you are fiddling with crafty and want to try it, it is easy
and won't cost any speed... just look at the "BishopMobility()" function...
and call RookMobility() in the same way... and for queens, use 'em both
together...
Valavan Manohararajah
In article <d4VXaEAL...@hatbulb.demon.co.uk>,
Tom King <t...@hatbulb.demon.co.uk> wrote:
>All,
>
>A while back I remember reading (maybe here, maybe ICCA journal) about
>the value of having a direct measure of mobility in an evaluation
>function of a chess program. I think I read that Schaeffer had estimated
>that the adding a measure of mobility to an evaluation function was
>worth "about a ply". In other words a chess program with a mobility term
>in its evaluation function, and searching to depth N, would score quite
>well against the same program, without a mobility term searching to
>depth N+1.
>
>Anyone got any views on this? How much is mobility worth in terms? Does
>"about a ply" still hold at "interesting" depths that PC programs can
>now reach, say depth 9 or 10?
>
>I ask because I've recently been playing with a mobility term in my
>chess program, Francesca. I'm not 100% convinced that it's worth "about
>a ply", but Francesca's playing more interesting chess with the mobility
>term.
>--
>Tom King
I use only a bishop mobility term.... at one time, in an old version of Rajah
I did use mobility terms for queens, and rooks as well. I didn't like what
I saw... in certain positions rooks don't have that much mobility but are
quite well placed. The same applies for queens as well. Besides computing
mobility for many pieces will take up a large amount of time.
The only reason why I left the bishop mobility term in was that it might
help with bad bishop cases.... I haven't seen any weird bishop moves lately,
while when I had rook and queen mobility I used to see some real weird moves
that turn out to be maximizing the mobility of pieces.
Tom King
In article <EBGs3...@ecf.toronto.edu>, Valavan Manohararajah
<man...@ecf.toronto.edu> writes
Maybe you get weird moves if you try and maximize mobility. But are the
weird moves always bad moves? I suppose it depends on your own chess
playing ability. Often I find it difficult to judge whether a move that
looks weird is really bad or not. This is probably because my own chess
playing ability is so limited (1500 ELO).
With the mobility term my program does lose games, but they're fun to
watch. Without the mobility term it would often get slowly squeezed to
death. :-(
Interestingly, the BT2450 test suite (a favorite of mine) shows
practically no difference between the version with, and the version
without mobility.
--
Tom King