On Apr 20, 11:29=A0am, help bot <nomorech...@[EMAIL PROTECTED]
> wrote:
> On Apr 20, 5:36 am, Martin Brown <|||newspam...@[EMAIL PROTECTED]
>
> wrote:
>
> > > > > > > 6. d2-d3{20} Ke8-g8{16}
>
> > > > > Although 7. BxN dxB 8. Nxe5 snaffles a pawn it isn't such a
massiv=
e
> > > > > advantage. However, since it is a purely materialistic gain it
is
> > > > > surprising that Getclub did not grab it with both hands.
>
> > I take that back. I had been assuming (incorrectly) that if it was a
> > significant advantage to white that blundercheck would have added the
> > relevant annotation. But since it didn't... I assumed there was
> > something about it that I couldn't quite see.
>
> > However, it seems looking more carefully at the opening book database
> > move order 7. BxN is the *only* continuation given at that point. The
> > game falls back into book territory further on by transposition.
>
> =A0 When I try to utilize the online databases, they
> only rarely give *quality* information. =A0Sometimes
> the problem is that losing moves are show with a
> high winning percentage, and sometimes the
> number of games reaching a particular position
> are few, and of low standard. =A0I expect the
> commercial databases are much better, but I
> neither wish to pay for something I rarely use,
> nor do I like buying an older version to save
> money, because I /know/ it's out-of-date!
An out of date one is still interesting. My database is from 2006 (and
I don't really feel compelled to update it). However, the 2006
bigopening book still contains quite a few dodgy "winning" moves in
the leaves of the tree. It would be interesting to check one for all
logical inconsistencies with an engine set to 20s/move and thereby
winnow out all the chaff. I suspect some of it is there for proof of
copyright reasons as watermarks.
>
> > A proper ab initio analysis of this position with Shredder quickly
> > concludes that BxN is 60cp better than the next best move Nc3, c4, h3,
> > Bd2 (all at 0 +/-3). Rybka is a bit more conservative and gives a 40cp
> > advantage with a preferred ordering of toher moves h3, Bd2, c4, Nc3,
> > Bb3a3, Bg5, Re1 (all at 6+/-2).
>
> > Rybka typically has a flatter evaluation function (like Fritz and
> > Crafty). Shredder has an exaggerated sense of some structural features
> > (which I think tends to guide it towards more human like play).
>
> =A0 The very latest version of Shredder has toned
> this down and they are breathing down Rybka's
> neck on the ratings chart. =A0But since Rybka is
> not well-honed in the endgame and Shredder
> (or Deep Shredder) has the fastest table-base
> search around, perhaps both programs are
> worth buying?
I have both. I prefer to spar against Shredder10. It feels mouch more
like playing against a human and it occassionally leaves things on
that someone playing at the nominal playing level might make. Rybka is
a bit too much like playing against a brick wall - although I have
once beaten it on time control (the only engine I have ever managed to
beat that way).
>
> > > > =A0 After thinking about this a little more, I realized
> > > > that the Knight need not be snatched before
> > > > taking the pawn, since it is, after all, pinned to
> > > > the King.
>
> > > =A0 Whoa! =A0Taking it just a tad slower, I now see that
> > > 5. Ke1-g1 *is* castling, so the a1-e8 pin is no more.
>
> > Oops. I really wish he would fix the =A3$#^$&% useless game notation.
I
> > find that the captures being shown as ordinary moves make life all but
> > impossible for following the games without a board or a computer.
>
> =A0 Only for us weak players. =A0I hear (but I would
> have no way of knowing) that strong players
> can do all this stuff blindfolded, in their heads!
I expect the blindfold might help provided someone reads you the
moves.
> > The problem is that I used blundercheck and trusted it to annotate any
> > dumb moves. Since it didn't annotate this I assumed incorrectly that
> > it was sound (and inside the usually robust book horizon). I confess I
> > didn't lok very hard at first.
>
> =A0 Maybe "blundercheck" has a set limit, which
> was just barely too high? =A0Some of these
No. Blundercheck effectively assumes that the book is correct and
stops when it encounters an in book position.
I have the blundercheck window set fairly narrow 8cp to analyse my own
games and don't alter it for Getclub. I do find that it sometimes
misses fairly obvious improvements that are found by full annotation.
I am not sure why. No-one from CM has ever bothered to explain why
blundercheck sometimes blunders.
> > The flaw is that when a move that is out of book is played (ie but the
> > wrong order) blundercheck should run the engine and see if the
> > sequence order alters the outcome significantly. Hitting a position
> > that is in the opening book repertoire is not a sufficient test -
> > there should be an option to cross check all the opening moves too
> > (especially any where the move played is not on the approved book list
> > for that ply - played out of order).
>
> =A0 Thus far, I have managed to avoid any of
> those kinds of problems by simply not
> bothering to install an openings book. =A0The
> programs are all out-of-book, all the time.
I generally play against engines using a deep opening book to try
ideas out in the opening. Most engines would take ages and not
necessarily find the right sorts of moves without a book (something I
actually agree with Phil Innes on).
>
> =A0 The problem is that every game analysis
> wastes time looking at oh, 1. e4 versus
> 1. d4 versus 1. Na3.
You could always build a custom shallow opening book from a trivial
set of games... I do have a null opening book somewhere. I think the
free version CB lite can be used to generate them.
>
> > I have noticed blundercheck sometimes misses clearly winning lines
> > when used to annotate certain types of game. I presume it stems from
> > having the cache polluted with moves from the suboptimal line actually
> > played when the decsisive move is encountered in the backwards search
> > down the tree.
>
> =A0 I have been using "Arena" (instead of the
> Fritz GUI), and it seems to help if I analyze
> the game backwards, so the stupid program
> can see how a true genius handled the position.
Populating the cache with moves seen from deeper plies works very well
provided that the moves played were close to optimal. It sometimes
gives odd results when a bad move has been played. I suspect thtat is
the root cause of my annoyances with blundercheck intermittently
failing to spot critical move in real games.
Regards,
Martin Brown
|
