UHO openings contain many too imbalanced positions

[PavelChess]

The purpose of imbalanced opening was to reduce draws percentage. The expected outcome was that successful patch will increase wins/losses ratio.

I have ran handicapped match of 2 copies of recent version. One copy got search limit of 200000 nodes, and other of 100000. Openings book UHO_Lichess_4852_v1.epd.
The match ended by result 416 wins - 122 losses - 462 draws.

So about 25% of positions from first 500 of the book are so imbalanced that even handicap 2:1 couldn't give pair result 1.5-0.5.

Usually good patch can give +0.5 to +2 ELO. This is much less than enough to avoid loss at those positions that 2:1 handicap had no success.

This can lead to wrong or unwanted testing results, for example:

1. Gaining patches test needs much more games
2. Gaining ELO patch can fail because of too low gain
3. Simplification patch can pass because a test shows less ELO loss than real.
4. Simplification patch that fails needs much more games
5. Influence of this at LTC may be greater than at STC

I suggest to clean up imbalanced opening positions that even handicap for example 2:1 cannot avoid losses at them

[robertnurnberg]

Thanks for sharing your data. Just to say that we do have this on our radar and are tracking the sharpness of the book exits on the Chess Cloud Database chessdb.cn, see the repo UHOtrack. So in future we may well filter some of the positions from the current book. For now we do not see an urgent need to change the book, though.

[vdbergh]

@PavelChess If you suspect the book can be improved then you should make a new version and test it according to the procedure described here https://github.com/official-stockfish/fishtest/wiki/Fishtest-faq#how-to-compare-opening-books