Stefan Pohl Computer Chess

private website for chessengine-tests


The Unbalanced Human Openings XXL Project

 

The goal is the development of a much bigger UHO openings-sets for the Stockfish Framework testings. In the Framework the draw-rates (using normal, balanced openings) raised up to around 85% at STC and around 94% on LTC, because Stockfish gets so strong and when testing one patch, the 2 Stockfishes playing vs. each other are so close in strength. So, the Framework needs unbalanced openings. Because one test in the Framework can have 100000 or more games, it is clear, that my (already released) UHO V3 openings-sets are too small – the biggest set has 25000 openings. A UHO openings-set for the Framework needs at least 100000, better 150000 different opening-lines. So, I started the development of some UHO XXL opening-files.

Goal of the project: Development of 7 UHO XXL files, each 8 moves (16 plies) deep, with increasing eval-interval (eval of the endposition by KomodoDragon 1.0): [+0.80;+1.09], [+0.90;+1.19], [+1.00;+1.29], [+1.10;+1.39], [+1.20;+1.49], [+1.30;+1.59], [+1.40;+1.69].

At the moment, the first eval-interval of [+0.80;+1.09] should work best in the Framework, but if Stockfish gains more strength and the hardware gets faster in the future, the draw-rates will raise again. Then a openings-set with a higher eval-interval can be chosen, to lower the draw-rates again into a valid range of 45%-65%.

 

Because the development of such big files is a huge effort and will take at least 3 months of development time, I build this small website, where you can see the progress of my work. Used tools: pgn-extract, pgn-scanner and Fritz13GUI.

 

Start of the development was 2021/07/12

Latest update of the progress of my work was 2021/07/16

 

Step 1: Build a database out of all games since 1945 of the Megabase 2021 (around 8 million human games) and 2 million human games (2019-2020) out of the LiChess Elite-Database (filtered all (standard) games from lichess to only keep games by players rated 2400+ against players rated 2200+, excluding bullet games.). COMPLETED

 

Step 2: Delete all games out of this 10 million human games database, which were starting from a FEN (Chess960). Delete all comments out of the games. Delete all games with less than 15 moves.  COMPLETED

 

Step 3: Delete all played moves beyond 8 moves / 16 plies. Delete all games, where not both queens are still on the board in the endposition. COMPLETED

 

Step 4: Remove all games with an endposition, that already is in the file (remove all doubles). Result: 2.7 million opening-lines with different endpositions, all 8 moves deep. COMPLETED

 

Step 5: Delete all Tags, replace them with 7 empty standard Tags. Add the ECO-Code Tag. Setting all game results to 1/2-1/2. COMPLETED

 

Step 6: Split the 2.7 million lines file into 5 smaller files for evaluating the endpositions. Each of these smaller files will take around 16 days for evaluating the endpositions (2.5 seconds per position on a 12core AMD Ryzen CPU). After finishing each file, the PC will be rebooted. COMPLETED

 

Step 7: Evaluation of the 5 files of the splitted 2.7 million opening-lines file, called File A-E (using KomodoDragon engine, 3.5 seconds per position on a 12core AMD Ryzen CPU, using the pgnscanner-tool (5 instances, 4 Threads each pgnscanner)) - should take around 3 weeks.

File A: IN PROGRESS

File B: IN PROGRESS

File C: IN PROGRESS

File D: IN PROGRESS

File E: IN PROGRESS

 

Step 8: Merge the 5 analyzed files A-E back together in one file: The UHO XXL raw-data file. NOT DONE

 

Step 9: Check the raw-data file for errors, doubles again. Reformat the Annotator-Tag to „depth xx, eval +xxx“. NOT DONE

 

Step 10: Build (filter) the 7 UHO XXL-files with increasing eval-intervals out of the raw-data file. Build EPD-files out of the 7 UHO XXL-files. NOT DONE

 

Step 11: Build opening-books (FritzGUI, ShredderGUI, Arena, polyglot) out of the 7 UHO XXL files. NOT DONE

 

Step 12: Do several testruns close to SF-Framework conditions (singlethread, 10sec+100ms / 60sec+600ms) using the 7 UHO XX-files and the noob_3mvs-openings-file (which is in use in the SF Framework at the moment) for comparsion. Verify, if the [+0.80;+1.09] UHO XXL file is the best choice right now for the Framework, as expected, or if another eval-interval is better. NOT DONE

 

Step 13: Write a short ReadMe-file for the release-package. NOT DONE.

 

*** RELEASE of the UHO XXL openings, including raw-data file and the testing results ***