Site review - ChessLab database
Mark Weeks
million games online' by ChessLab. The home page is at
http://www.chesslab.com/, where we learn that the service provides 'Java
based interactive search, analysis of 2 million chess games [...] move
chess pieces on Java chess board; search and analyse any position for
free [...] search games by position, players names, years, etc. over the
Internet for free [...] interactive study of chess games, generation of
PGN'.
It's not clear if this is a commercial site. My guess is 'no', because
there are almost no banner ads & I see no other way that it generates
revenue. The bottom of the page says 'Copyright Š 1999, GameColony.com',
which links to a page offering 'Free Online Board and Card Games'. The
Usenet posts announcing the service started around end-1998. Network
Solutions 'Web Interface to Whois' says that the record for the site was
created on 31-Aug-1998. The site is registered to BPS Software, Inc. &
all contacts are listed as Boris Shneyderman.
The Java fun starts on http://www.chesslab.com/PositionSearch.html. This
search interface has two frames -- on the left is a chess board with
some controls; on the right are four tabs. The default 'Search games'
tab controls a database search using parameters & buttons. The other
three tabs are 'Analyse', which suggests moves for a given position;
'Magazine 12/99', which opens up the 'ChessLab Interactive Magazine';
and 'Board', which sets some options for the left frame.
On the 'Search games' tab there are five buttons for 'Help', 'Search
games', 'Openings', 'Position stats', and 'Download'. The help button
opens a new window which explains the parameters and buttons. It links
to other help pages, of which the most useful is the index
http://www.chesslab.com/Help/h_indexchess.htm.
The 'Search games' button launches a database search based on the search
parameters:-
- White Player
- Black Player
- Both colors [checkbox]
- City/Location *
- Game Result ['White won', 'Black won', & 'Draw' checkboxes]
- Date range ['Latest games (1991-present)' & 'Historical archive
(1485-1990)']
- Year
- Rating *
The parameters marked '*' are 'Accurate for chess games after 1997'.
There is also a link to set some 'Advanced options', which are:-
- Hits per page
- With checking ['Castling' & 'En passant captures']
- Player name scan ['Fast', 'Similar, & 'Full scan']
- Max. moves
- Reverse sort by year
The 'Openings' button allows a selection of ECO codes. When one is
selected, the moves are played automatically on the board in the left
frame.
The 'Position stats' button searches the database and returns statistics
[% won by White, % won by Black, % Draw] for games in the selected
database. The initial position of a chess game gives:-
- Historical archive 40%-28%-31%
- Latest games 38%-30%-30%
whih means that White has won 38% of the games in the database played
since the beginning of 1991.
There is a roundoff problem here -- the numbers don't add up to 100%. I
apportioned the missing 1-2% over each of the results using the same
ratio, e.g. in the 'Historical archive', the result 'White won' gets
40.40%, 'Black won' gets 28.28%, etc. Then I calculated the expected
point value of a game for the two databases. This gives:-
- Historical archive W-0.561 B-0.439
- Latest games W-0.541 B-0.459
which indicates that White's expected point value is decreasing over
time. I did some more tests of this interesting statistics function,
which I'll post on the discussion group as a response to this article.
The 'Download' button opens up a new window, which offers to create a
file either from the game on the board -or- from the games returned in
the latest search. The file is written to the same new window, where it
can be saved as a PGN file. I initially had some trouble with this
because the buttons were never reset correctly after a download. I
discovered in the help pages that this is because I run my browser
without a local cache. The problem disappeared when the cache was
reactivated.
Unfortunately, the PGN headers don't respect the PGN standard. The help
page 'Please send your chess games' describes the PGN format used by
ChessLab...
'In one email (or attached text file) you can send one or multiple
games, where the format conforms to the following example:
[Event "8th Amber Blindfold"]
[Site "Monte Carlo MNC"]
[Date "1999.??.??"]
[White "Karpov, A"]
[Black "Ljubojevic, L"]
[Result "1-0"]
[WhiteElo "2710"]
[BlackElo "2571"]
[Round "6"]
1. d4 Nf6 [...] 30. Rd8 1-0
'Please note that standard export PGN format has no comments or diagram
indicators. Also, please note that there is one space between the move
number and the actual chess move.'
...This is the same format which is produced on a download. The
specification for the PGN format is available on the site at
http://www.chesslab.com/PGNDescription.txt. This is the standard dated
1994.03.12, by Steven J. Edwards, where the relevant section is...
'8.1.1: Seven Tag Roster
'There is a set of tags defined for mandatory use for archival storage
of PGN data. This is the STR (Seven Tag Roster). The interpretation of
these tags is fixed as is the order in which they appear. Although the
definition and use of additional tag names and semantics is permitted
and encouraged when needed, the STR is the common ground that all
programs should follow for public data interchange.
'For import format, the order of tag pairs is not important. For export
format, the STR tag pairs appear before any other tag pairs. (The STR
tag pairs must also appear in order; this order is described below).
Also for export format, any additional tag pairs appear in ASCII order
by tag name.
'The seven tag names of the STR are (in order):
1) Event (the name of the tournament or match event)
2) Site (the location of the event)
3) Date (the starting date of the game)
4) Round (the playing round ordinal of the game)
5) White (the player of the white pieces)
6) Black (the player of the black pieces)
7) Result (the result of the game)'
... Note that 'Round' should be the fourth tag, but it is the last in
ChessLab's header. Fortunately, this is only a minor annoyance, which is
probably overcome by most software. I've never had a real problem with
it.
My main interest is in the value of this site to the chess historian. It
can also be used to research openings and positions, and it appears that
the tool was designed primarily for this reason. I decided to perform
some practical tests related to chess history.
---
I) How many Bobby Fischer games are in the 'Historical archive'?
1) A search on 'fischer' returned many games played by other than Robert
James, so I had to narrow the search on name. The PGN standard says, 'If
a first name or first initial is available, it is separated from the
family name by a comma and a space.'
2) A search on 'fischer, r' returned 'Sorry, no games found for this
position with current parameters'. Maybe the name is represented
otherwise. A search on 'fischer,r' returned the same message. I decided
to locate some Fischer games to see exactly how the name is spelled in
the database.
The following searches were all limited to 1972.
3) A search on 'fischer' & 'spassky' returned 21 games where Fischer's
name is spelled 'Fischer, R', following the PGN standard. A search on
'fischer' [without Spassky] returned 22 games. There was one additional
game against 'Parham, F'.
4) A search on 'fischer,' returned 21 games, including the Parham game,
so I lost one game somewhere. A search on 'fischer, ' [note the
trailing space] also returned 21 games.
5) A search on 'fischer r' [as well as on 'fischer, r' & 'fischer, r ']
returned 'Sorry, no games found...'.
At this point, I gave up -- I don't know how to narrow the search on
name. Unless I'm overlooking something obvious, the search doesn't
handle players with the same family name -- searching for one of the
Polgar sisters would be a problem.
Finally, I looked at the results of the searches where 21 games and 22
games were returned. For some reason, in some searches, game 13 of the
1972 match is returned twice with a different number of moves in the two
games. The PGN headers seem identical. The extra game is returned on the
search for 'fischer' & 'spassky', but not on the search for 'fischer'
alone. There is something wrong here.
---
II) How many Morphy games are there?
If I can't get the number of Fischer games in one search, I'll have to
go through the database year by year to get all of Fischer's games. I
didn't want to do that for this article -- it is, after all, only a
test. There should be little trouble with Morphy's name, so I decided to
try that instead.
The results of a search are normally returned by descending year &
ascending player names. The first game returned for 'Morphy, P' was
played in 1898 (Morphy died in 1884, so it must be another Morphy!)
against 'NN', the second was played in 1869.
I can only get a maximum of 50 games each time. After clicking on 'Next'
a few times I arrived at 'games from 251 to 280' & no 'Next' button. The
last game in the list was for 'Morphy, E-Ford' in 1840. The next to last
was for 'Morphy, P-Morphy, A' in 1847. There were two games in 1848.
I downloaded the 280 Morphy games. I also downloaded the corresponding
file from UPITT...
http://www.pitt.edu/~schach/Archives/index2.html
ftp://136.142.185.47/group/student-activities/chess/PGN/Players/morphypg
.zip
... and noted that morphypg.zip is a 56 Kbyte file, was created on 'Sun
Jul 9 00:00:00 1995', and contains 400 games. After comparing the files
by game date, I discovered that most games in both files were played in
three consecutive calendar years...
Year ChessLab UPITT
1857 57 86
1858 118 131
1859 53 91
...and that UPITT consistently has more games in any one year. I know
from experience that the UPITT player files contain many duplicate
games, with small variations in the moves of the duplicate games. Can
that account for the difference of 120 games between the two files?
I compared the files on Morphy's opponents in 1858. After making a small
correction on the UPITT file, where the PGN standard for names is not
respected, I discovered many similarities, but also some differences.
Here's a sample of both...
Player ChessLab UPITT
Barnes, T 8
Barnes,T 8
Bird, H 4
Bird,H 4
Boden, S 9
Boden,S 10
Comparing the Boden games from 1858, I found the game missing in the
ChessLab collection. It is not listed under another year in the ChessLab
collection & is not duplicated in the UPITT collection. My conclusion is
that the ChessLab collection is missing games that are available in
other collections.
---
III) Kasparov - Anand
For some time now, I've wanted to add to my own site an overview of all
games played between Kasparov and Anand. I've already looked up all
games played between the two players since their 1995 match & decided to
check this against the ChessLab database.
The 'Latest games' database returned 61 games. The 'Historical archive'
returned 'Sorry, no games found...'. The following table gives the
breakdown by result, e.g. the first line shows that Kasparov has won
nine times with the Black pieces...
Anand, V Kasparov, G 0-1 9
Anand, V Kasparov, G 1-0 5
Anand, V Kasparov, G =-= 11
Kasparov, G Anand, V 0-1 3
Kasparov, G Anand, V 1-0 15
Kasparov, G Anand, V =-= 18
...while the result over all 61 games is +24-8=29 for Kasparov. How
reliable is this calculation?
Looking at the results for the 1995 PCA title match, I saw that games 1
& 7 are duplicated, and that game 5 is missing. The duplicate game 1 has
a version with Kasparov as White, although Anand had White in the odd
numbered games.
When I find the time to look at this more carefully, I'll post the
analysis. All I can say for now is that Kasparov is an overwhelming
favorite in any new match, which is really no surprise!
---
IV) Olafsson - Fischer
I often encounter discrepancies between PGN files and the printed
literature. A recent example is the game Olafsson - Fischer from round
11 of the 1958 Portoroz Interzonal. The PGN game score I have on file
does not match the score given in Fine's 'The World's Great Chess
Games'. The discrepancy starts where the PGN move 37...R8e4 does not
match Fine's 37...R(K1)-K4, which is 37...R8e5 in algebraic notation.
This type of discrepancy normally disappears after a move or two when
the game scores converge to the same line. In these cases it is
impossible to tell which move is correct & it usually makes little
difference to the result.
What made this case different was that the PGN score & Fine's score fail
to converge. The endings of both scores are completely different until
Fischer resigns on his 44th move. When I crosschecked with Wade &
O'Connell's 'Bobby Fischer's Chess Games', I found a third variation.
Wade's score matches Fine's until it diverges on the 40th move. It
converges on the next move, not to Wade's score, but to the PGN score.
If Wade's score is correct, then Olafsson's 40th move was a blunder
which would have allowed Fischer to gain the advantage.
How many versions of this game does ChessLab's database have? A search
on 'Olafsson - Fischer' returns four games, of which one is from the
1958 Interzonal, and which matches the PGN score. The move 37...R8e4 is
more logical than 37...R8e5, so it seems that ChessLab has the correct
score.
---
While I was finishing this review & double checking my work, I
frequently received the error message...
'A network error occurred: unable to connect to server (TCP Error:
Broken pipe) The server may be down or unreachable. Try connecting again
later.'
...from the ChessLab server. This usually means that the server is
having problems. Because of the complex database engine required to
drive a service like this, I suppose that Shneyderman operates the
server himself. He must be a very busy person -- writing the software,
operating the server, & loading the new games. This is a tremendous
task.
Let there be no doubt. This database is a great tool & I have already
used it as a tool to prepare a new research. If a game has been
converted to digital format, the chances are good that it has been
loaded on ChessLab's database. But, as with many Web-based resources,
its results are not completely trustworthy and need to be confirmed
against another source. Judging from his help pages, Shneyderman is very
receptive to any additions & corrections which other people might offer.
Maybe I'll send him a collection of my own games -- it's the only way
they'll ever be published!
Bye for now,
Mark Weeks
Sent via Deja.com http://www.deja.com/
Before you buy.
duff...@my-deja.com
ChessLab database and the CD Ultimate Game Collection by Chess For Less
1996, here is what i found for some years:
ChessLab Ultimate
1867: 111 111
1868: 78 78
1869: 68 69
1870: 114 113
1871: 141 141
1872: 69 70
1873: 208 209
1874: 77 77
1875: 48 48
1876: 123 123
1877: 124 124
1878: 198 199
1879: 88 87
1880: 177 182
1881: 174 174
Isn't that a little curious all this coincidence in the number of games
in each of them for the same year.Are they related in any way ?Hum...i
let you draw your own conclusion.
Mark Weeks
Mark Weeks <10004...@compuserve.com> wrote:
> What made this case different was that the PGN score & Fine's score
fail
> to converge. The endings of both scores are completely different until
> Fischer resigns on his 44th move. When I crosschecked with Wade &
> O'Connell's 'Bobby Fischer's Chess Games', I found a third variation.
> Wade's score matches Fine's until it diverges on the 40th move. It
> converges on the next move, not to Wade's score, but to the PGN score.
In case anyone is paying attention, I mangled that last sentence. It
should read, 'It converges on the next move, not to [Fine's] score, but
to the PGN score'.
Here's the game score, annotated using ChessBase. WAD-BFCG & FIN-WGCG
refer to the books by Wade & Fine. - MW
[Event "izt"]
[Site "Portoroz"]
[Date "1958.??.??"]
[Round "11"]
[White "Olafsson F"]
[Black "Fischer R"]
[Result "1-0"]
[ECO "D38"]
[PlyCount "87"]
1. c4 Nf6 2. Nc3 e6 3. Nf3 d5 4. d4 Bb4 5. cxd5 exd5 6. Bg5 h6 7. Bh4 c5
8. e3 Nc6 9. Rc1 c4 10. Be2 Be6 11. O-O O-O 12. Nd2 Be7 13. b3 g5 14.
Bg3 Ba3 15. Rc2 Nb4 16. bxc4 Nxc2 17. Qxc2 dxc4 18. Nb5 Bb4 19. Nc7 Bxd2
20. Nxe6 fxe6 21. Bxc4 Qe8 22. Qxd2 Ne4 23. Qd3 Nxg3 24. hxg3 Rf6 25.
Qe4 Rc8 26. Bb3 Qd7 27. Rd1 Re8 28. f4 Qh7 29. Qe5 Qf5 30. g4 Qxe5 31.
dxe5 Rf7 32. f5 Rc7 33. Rd6 Rc5 34. Bxe6+ Kf8 35. Bb3 Rcxe5 36. Rxh6
Rxe3 37. Rg6 R8e4 (37... R8e5 38. Rxg5 Rg3 39. Rg8+ Ke7 40. Kh2 (40. g5
Re2 41. Bd5 Kd6 42. Bf3 Rxa2 43. f6 Ke6 44. Re8+ {WAD-BFCG}) 40... Rc3
41. Be6 Re2 42. g5 Rcc2 43. Bd5 Kd6 44. Bf3 {FIN-WGCG}) 38. Rxg5 Rg3 39.
Rg8+ Ke7 40. g5 Re2 41. Bd5 Kd6 42. Bf3 Rxa2 43. f6 Ke6 44. Re8+ 1-0
Mark Weeks
> there are almost no banner ads
The site now has banner ads for Club Kasparov.
> This gives:-
> - Historical archive W-0.561 B-0.439
> - Latest games W-0.541 B-0.459
> which indicates that White's expected point value is decreasing over
> time. I did some more tests of this interesting statistics function,
> which I'll post on the discussion group as a response to this article.
Here are some more stats showing % won by White, % won by Black, & %
drawn. The searches were on the 'latest games'.
1.e4 'Error during querying database. Please,
restrict your search by adding extra parameters. Please accept our
apologies -- at ChessLab we are in the process of massive database
cleanup and updates. Your database searches will be successful again
within 2-3 hours.'
I had the same error later. It seems that some positions, apparently
those which occur too frequently in the database, cause a timeout.
1.e4 e5 Timeout
1.e4 e5 2.Nf3 41%-29%-29%
1.e4 e5 2.Nc3 41%-30%-28%
1.e4 e5 2.f4 43%-37%-18% (An unnecessary risk?)
1.e4 c5 Timeout
1.e4 c5 2.Nf3 37%-33%-29%
1.e4 c5 2.Nc3 35%-37%-27%
1.e4 c5 2.c3 35%-32%-33%
1.e4 c5 2.d4 cxd4 3.c3 37%-41%-20% (Ditto.)
1.e4 e6 Timeout
1.e4 e6 2.d4 40%-30%-29%
1.e4 Nf6 38%-32%-28%
1.e4 Nf6 2.e5 40%-31%-28%
1.e4 c6 38%-29%-31%
1.e4 c6 2.d4 38%-29%-31%
1.e4 c6 2.Nc3 40%-31%-27%
1.e4 d5 42%-30%-27%
1.e4 a6 52%-29%-18%
---
1.d4 d5 2.c4 40%-24%-34%
1.d4 d5 2.c4 c6 39%-23%-37%
1.d4 d5 2.c4 e6 41%-24%-34%
1.d4 Nf6 2.c4 39%-28%-31%
1.d4 Nf6 2.c4 e6 40%-26%-33%
1.d4 Nf6 2.c4 c5 40%-32%-26%
1.d4 Nf6 2.c4 g6 41%-28%-30%
1.d4 Nf6 2.c4 g6 3.Nc3 37%-32%-30%
The 1.d4 Nf6 2.c4 g6 results are surprising. White's chances drop &
Black's chances increase after 3.Nc3, the move suggested by Chesslab's
'Analyse' function. This makes me suspect that there is something wrong
with the calculation.
1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 41%-29%-29%
1.d4 Nf6 2.c4 g6 3.Nc3 d5 38%-27%-34%
This is also suspicious. This indicates that Black has better chances
playing other than the King's Indian or the Gruenfeld. What else is
there?
1.d4 Nf6 2.c4 g6 3.Nf3 39%-26%-33%
1.d4 f5 41%-28%-29%
---
1.c4 c5 37%-26%-36%
1.c4 e5 37%-31%-30%
---
1.g4 36%-48%-15%
1.f4 36%-40%-22%
1.a3 27%-43%-28%
(NB - Anderssen played 1.a3 in his match against Morphy.)
---
Except for those results from the King's Indian complex, the stats
confirm my intuition. Of course, there's a lot more to developing an
opening repertoire than looking at statistics. If you're not comfortable
with the positions or don't understand the themes, your expected score
will be closer to 0%. - MW