Runs fine but not tested thoroughly, you never know with all the pgn quirks.
BTW, you now can use commented games as well.
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Wed Aug 16, 2023 3:27 pm
It quits without any warning or message, creating an empty blunder.pgn file and a temp file which is identical to the original (1st 4 games from twic1497), but in the root directory of the program.
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Wed Aug 16, 2023 3:56 pm
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Wed Aug 16, 2023 4:05 pm
Like before, it's not a full install, only the new executable.
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Wed Aug 16, 2023 4:48 pm
I know, I re-downloaded the 7z just in case, substituted the exe and run.
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Wed Aug 16, 2023 6:17 pm
My bad, gamesplit.exe is missing, copy it in the root.
I will make a new full install later.
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Wed Aug 16, 2023 9:43 pm
Seems to work, I'll test it more thoroughly this weekend, see if it's faster than Aquarium's 2-step blunder check. I'm concerned that the splitting and move-by-move output will add overhead.
You said the limit is 10K games, right?
matejst likes this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Thu Aug 17, 2023 1:03 am
10K is the limit of gamesplit, I will put a limit warning on it in the final version when the pgn exceeds 10,000.
Other than that, will add the game number (game 10 of 200 or so) in progress so you can see how many you still have to go. If things go too fast on the screen just press escape and it will pause.
Makes me wonder what the purpose is, are you really going to analyze 15M games? For creating an opening book?
If there are any wishes left then lemme know.
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Thu Aug 17, 2023 6:08 am
10K is gamesplit's file limit, you can have as many games as you want inside each file.
Yes, I will filter the whole DB trough the tool. The ultimate purpose will be to get a book, but not as an immediate result. This is "just" another step in the long road.
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Thu Aug 17, 2023 9:35 am
I am using SF bench function to analyze. This means epds. So one single pgn game is converted to epd and then feed to SF. Hence splitgame one game only.
What you can do, or already figured out yourself is to install SOMU in 4-8 different folders feeding each with 10,000 positions for a tremendous speedup.
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Thu Aug 17, 2023 3:50 pm
I was thinking more like 16 folders. That's how the speed comparison will be run.
Admin likes this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Thu Aug 17, 2023 4:35 pm
Even so, it will keep you busy for the rest of the year
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Thu Aug 17, 2023 5:31 pm
It's aimed mainly at the winter period. Give the cost of electricity, I can't justify using the computers en masse (all six of them) if it's not going to also contribute to the living room's heating.
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Thu Aug 17, 2023 6:45 pm
Well, it's an interesting project, perhaps you could look for people willing to help. From my end I can offer you 50 or 100Gb web space for sharing data.
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Mon Aug 21, 2023 9:14 am
1. check pgn for > 10.000 games, if so abort, we don't want information loss.
2. Added game number in the display as progress indicator, example game 33 of 10000.
3. Program exits in a normal way with a message when the analysis has finished.
4. Now that I understand what the tool is meant for, removed the firing up of the analyzed pgn display when the when analysis has finished.
-----
Download - 1. somu.7z 2. somu.exe And copy somu.exe in the somu folder.
Reason, Github does not accept file uploads > 25Mb
The progress indicator is a welcome change. It's a good thing then, that the price for electricity remained above 0.11 €/kWh this weekend, dissuading me from testing. I would've likely re-tested the new version anyway.
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Mon Aug 21, 2023 3:01 pm
Admin wrote:
5 of the 6 games were solved by the book.
For human play, 50cp may be acceptable. Fidgeting with that value, same as with analysis depth, will be key in obtaining the desired result.
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Tue Aug 22, 2023 11:45 am
I have been wondering if games like these are useful, in fact (I think) they are deceptive. With a margin of 50 cp none will be truncated, meanwhile the score for white has increased to > 1.5 pawn. You don't want to have such lines in your book as playable.
Instead of checking the last move for 50 cp a suggestion would be to check the last 2 moves with 50 cp (thus 169-99=70) and truncate the game as 1. d4 * IOW, don't trust this game.
Your view?
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Tue Aug 22, 2023 10:24 pm
End positions will be played on at a later stage, so there's no need to see if the eval gets above a certain threshold. That said, it'd be useful to stop analyzing past a given unbalance. If you can add a the option to truncate the game once a given value has been reached in cp, great.
matejst and Nezhman like this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Wed Aug 23, 2023 9:44 am
You don't have to use them if you use the default values, the tool will behave as usual in that case.
1. You will be prompted to type a maximum score in cp, if you type 100 games will be truncated whenever a score of 100 or -100 is detected. If you press enter the default value (999) is used.
2. You will be prompted to type a maximum move number, if you type 20 all games are truncated at move 20 if not earlier. If you press enter the default value (999) is used.
Executable to download is SUMO-NEW.EXE
Windows oddity, SOMU.EXE now also here is labelled as a virus, renaming it to somu-new.exe is okay for Windows. What a mess these false alarms have become. OTOH, better 100 times false than one time overseen.
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: EAS - for Stefan Wed Aug 23, 2023 5:22 pm
IMPORTANT?
I noticed even SF16 still uses HCE eval in its bench function, using the latest versions on https://abrok.eu/stockfish/ only NNUE eval is used.
Code:
Stockfish 16 by the Stockfish developers (see AUTHORS file) info string classical evaluation enabled info depth 1 seldepth 1 multipv 1 score cp 199 nodes 132 nps 132000 hashfull 0 tbhits 0 time 1 pv e5g6 f7g6
versus
Code:
Stockfish dev-20230822-440feecb by the Stockfish developers (see AUTHORS file) info string NNUE evaluation using nn-c38c3d8d3920.nnue info depth 1 seldepth 1 multipv 1 score cp -93 nodes 56 nps 56000 hashfull 0 tbhits 0 time 1 pv b8d7
Ozymandias likes this post
Ozymandias
Posts : 622 Join date : 2020-11-23
Subject: Re: EAS - for Stefan Wed Aug 23, 2023 5:28 pm
Home 
