'HH' Study Database, Edition 4

[guyhaw]

Harold van der Heijden is one of the leading lights in the Study Community and has worked tirelessly on its behalf as EG editor, composer, judge and database compiler.

His 'HHdbIV' database has no rival and now accounts for over 76,000 of the surmised 90,000 studies thought to have been published q.v. http://www.hhdbiv.nl/reviews where you can find a video of the product-launch, more information and an easy ordering process.

Of possible interest to this community is the fact that Eiko Bleicher and I worked with CQL, pgn2fen, good-ole Excel and Nalimov's EGTs to check all the sub-7-man positions in the mainlines of these studies. Obviously, if White is challened to win, all mainline positions must be won for White - and if White is challenged to draw, all mainline positions should be drawn.

Eiko and I found over 3,000 studies where this was not the case. Naturally, study composers are not infallible and the database represents the product of their labours. Many of these 'cooks' were known, and some errors merely highlighted mis-transcriptions of the original studies at some poinit.

Eiko and I also identified all sub-7-man zugs in the mainline positions. We found 7,690 Type A (draw/win, lose/draw or lose/win) zugs, and 15,192 Type B(win/win_faster, lose/lose_slower) zugs.

The corpus of studies will in turn challenge, delight, surprise and entertain you and I recommend it - and ours must be one of the largest-scale uses of the EGTs to date.


Oh - there is a larger-scalse use of the EGTs: we have analysed all the sub-7-man positions in the games of Chessbase's BIG CHESSBASE 2010.

g

[Kirill Kryukov]

Hi Guy,

Interesting, thanks for posting!

Does the "'HH' Study Database, Edition 4" already contain the fixes to those over 3000 cooked studies? (or removal of those that can't be fixed). If not, then is there some other way we can access your data? May be you published the cooked studies somewhere along with your analysis, or may be you sent the analysis to Mr. van der Heijden for the next version of the database?

Oh - there is a larger-scalse use of the EGTs: we have analysed all the sub-7-man positions in the games of Chessbase's BIG CHESSBASE 2010.

Same here - are you planning to share any of the analysis?

Best,
Kirill

[guyhaw]

Apologies - I perhaps did not make myself clear.

After Eiko B ran all sub-7-man mainline positions in HHdbIII past his copy of the Nlalimov EGTs, I analysed the findings and sent all the information about EGT-faulted studies and wrong-value-positions to Harold van der Heijden.

He incorporated almost-all the informatioin into HHdbIV ... so HHdbIV is a much better representation of the 'ground truth' of these studies than HHdbIII. Whenever you find 'cook GH/EB' or 'Haworth=G Bleicher=E' in the .pgn text of HHdbIV, that's connected with the analysis we did of HHdbIII.

The studies are 'good value' even if it turns out that the stipulation was not actually achievable - the reasons why not are instructive.


HH does not in fact give himself credit for translating our information, provided as position-evaluations, into over 5,300 'cooks' justified in move-terms.

For example, if there's a btm draw positioin in a study and the given move by Black does not take advantage of this, all the draw-moves we implicitly pointed to are given as cooks by GH/EB although in fact it was HH who revisited the Nalimov EGTs and extracting the drawing moves.


Currently, I'm not doing any more work on the Chessbase BIG DATABASE 2010 as I'm still focused on HHdbIV ... and Eiko is now producing the full data for me on Type B zugs. Re Type A and B zugs, qv http://centaur.reading.ac.uk/4518/ .

The volume of information in connection with the games is much greater than for the studies - e.g. there are 54 spreadsheets in play rather than two as spreadsheets don't handle more than 1m lines of data.

However, I've taken a look at the frequency of games finishing in a position which does not have the theoretical value of the advertised outcome. K Regan points out that DGT software has recently had a lot to do with this - the pgn interpreting an intervention by the arbiter to move the Kings to the centre as a move by the side-to-move! I guess I'll have to check the advertised outcome against the theoretical value of the penultimate position

We've aleady had a look at:
- maximal depths,
- the number of positions with DTM depths over 50, 100 and 150,
- the number of games that enter sub-7-man space with theoretical value 'x' and finish with theoretical vallue 'y'
- the number of Type A and B zugs occurring over the board

One amusement is that there is exactly one game that features castling in a s7m-endgame. In fact, the winner mates by castling on the last move. Kudos to the youngster who did this: they are likely to be the only person in the world to have done so in a 'Chessbase' game. He rather underlines the fact that those small endgames that feature non-null castling rights should be computed: would have saved me a lot of time manually looking at those studies which do feature castling (and always as an essential part of the successful strategy).

g

[Kirill Kryukov]

Thanks, it's more clear now. Also thanks for the link to "6-man Chess and Zugzwangs" piece, I enjoyed it.

[jshriver]

Interesting read so far. I was amazed to see that even 5-1 was taken into consideration. Since nalimov doesn't support that what dataset was used and is it available with source code probing code?

There was a thread here last year about 5-1 so I'll have to dig it up.
I'm curious though but what do you mean by "study"? 80k study in my mind means 80k independent topics of research (even if related).

-Josh

[guyhaw]

If JS is referring to my paer on '6-man Chess and Zugzwangs', the 5-1 zugs were supplied by Marc B after a search of EGTs created by him using Yakov Konoval's code.

'6-man' should strictly have been 'Sub-7-man' but no-one's perfect. There are sets of KNPPK and KNPPPK zugs, and and a challenge is to string them together in a Knight's tour. Writing the paper turned into a journey of discover, aided by assistance from HvdH, MB, Noam Elkies and Ken Regan.

A 'Chess Endgame Study', q.v. Google or Wikipedia, is a chess composition which challenges White either to draw or win, and whose solution is technically correct and artistically attractive. The most famous Chess Study is probably the Barbier-Saavedra study, often just known as 'The Saavedra'. You can see it here at http://www.hhdbiv.nl/multimedia under 'Movies' as it starts the video of Harold's launch event.

g