Re: Finding the best chess move
- From: chess player <Gulleiot@xxxxxxxxxxxxxx>
- Date: Fri, 15 Jun 2007 10:05:33 -0700
On 15 Jun, 16:32, "Nick" <tulse04-ne...@xxxxxxxxxxx> wrote:
"chessplayer" <Gulle...@xxxxxxxxxxxxxx> wrote in message
news:1181919556.944148.99380@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
I'm trying to find astatisticalmethod to find thebestmove in a
chessgame, based on past results of playing that move in that
particular position. Often a particular sequence of moves, perhaps 12
or so for each side, have been played many times before and so in a
large database of games, one can determine the soundness of a move by
the results that have been achieved with it.
First, the score of achessmove is defined as
100*(number_of_wins_by_white + 0.5*number_of_draws) / total_games
so if there are 100 games, with white winning 50, black winning 45 and
5 drawn, the score is 52.5%. (At high levesl ofchess, white is
considered to have a small advantage, so the score is usually 54% or
so. Among weaker players, having white is not really an advantage).
Looking at a database of games, I see for the very first move, a
particular move (1.Na3, moving the knight to a very odd position)
scores thebestin my database, with a score of 70%. So whilst that
might look a good move, the fact is that it was only played 4 times in
638401 games, so I doubt it would be safe to draw any conclusions
about the soundness of that move, based on such a small sample. In
fact, mychesskn9owledge tells me it would be a poor first move).
I'm trying to find a way of seeing if its possible to determine the
bestmove from a number of moves, based on how well it scores and the
number of times it was played. F
To take a larger sample, the following scores and frequencies were
observed, where the first column is the move (1.e4, 1.d4 etc), the
second the number of times it was played in 638401 games, and the 3rd
column is the score.
1.Na3, 4, 75%
1.Nf3, 54361, 55.6%
1.d4, 211948, 55.0%
1.g3, 4429, 54.5%
1.c4, 38305, 54.4%
1.e4, 320507, 53.4%
1.b3, 2474, 51.0%
1.Nc3, 1365, 49.0%
1.d3, 276, 48.1%
Bobby Fischer said once, that opening with the move 1.e4 was the "best
by test", yet based on my database, the move 1.d4 scored a little
better (55.0%) than 1.e4, which scored 53.4%. So was Fischer wrong?
I'm looking for a method that can determine if a move A is better than
move B, with a certain confidence limit.
Can anyone suggest astatisticaltest that might be appropiate in this
case?
To give a further example, one sequence of opening moves is 1.e4 e6
2.d4 d5, then white chooses one of 4 moves for his 3rd move
3.Nc3 (main line)
3.Nd2 (Tarrasch variation)
3.e5 (advance variation)
4.exd5 (exchange variation).
in my database, which has 36280 games reachng that position, the
scores and fequencies are:
3.Nf3, 14435, 56.6%
3.Nd2, 11233, 57.3%
3.e5, 5348, 51.7%
3.exd5, 3042, 8.3%
other rarer moves, only total 222, of the 36280 moves.
Any thoughts?
Surely it depends on the skills of the player. It might be statistically thebest- but if I don't know the strategy after moving there, then it is not
much good.
Nick
Yes, I don't deny that. Clearly to make best use of a move, you need
to know the strategy, but I don't see much point in learning the
strategy of a move order that performs poorly. It is *generally*
better to find a move that scores well and learn the strategy of
that.
There are exceptions to that. I often play a gambit known as the Smith
Morra Gambit, where white gives away a pawn, but gets some
compensation in that his pieces are easily and quickly developed, and
the slightest error by black can result in him being crushed. I know
the Smith Morra Gambit is unsound at the highest levels of chess, so
few grandmasters ever play it. But I score pretty well with it. If I
ever become a grandmaster (which does not look very likely), I'm sure
I would stop playing it.
The move 3.exd5 above (known as the French Exchange), is often chosen
by white when he/she only needs a draw to win a competition, since it
is very 'drawish'. In my database of games, 42% are drawn, so if white
just wants to avoid loosing, using the French Exchange is quite a
good choice.
Rather oddly, the most popular move in a particular position does not
always give the best results, although in many cases the move chosen
most often scores the best. But obviously when the data set is very
small, a very poor move can score very highly.
So I don't deny there are various reasons for using a particular move,
but I feel that if a move A is statistically better than move B, it is
probably worth learning the strategy behind move A, rather than devote
time to the inferior move B. Having decided to use move A in any given
position, there is usually a book devoted to it, so one can learn the
strategy.
Anyway, those things aside, is there a statistical test that could
indicate when a higher score is likely to be indicative of a better
move, rather than the results of an inadequate sample size?
.
- Follow-Ups:
- Re: Finding the best chess move
- From: Nick
- Re: Finding the best chess move
- References:
- Finding the best chess move
- From: chess player
- Re: Finding the best chess move
- From: Nick
- Finding the best chess move
- Prev by Date: Re: Help with probability&stat problem
- Next by Date: Computing overall rank
- Previous by thread: Re: Finding the best chess move
- Next by thread: Re: Finding the best chess move
- Index(es):
Relevant Pages
|
