mmt Vs. Ovyron (G4 D5 BG2)

[Ras]

Yes, I've seen SSDs and NVMe (when I tested it for a bit) actually get stressed when SF was hitting tablebases.

It might make sense to mount that partition with -noatime under Linux. Back in the day, I ruined a USB stick with 6 piece Nalimov tables within hours because I had forgotten noatime so that every read access was also a (pointless!) write access.

[Zenmastur]

What I don't want is a book that has these massively long computer lines that have never been played and never will be played.

On the other hand, many of these lines are very forcing on white so maybe I'll end up with a bunch of very long computer lines even if that's not the intent.

[Zenmastur]

Yes, I've seen SSDs and NVMe (when I tested it for a bit) actually get stressed when SF was hitting tablebases.

It might make sense to mount that partition with -noatime under Linux. Back in the day, I ruined a USB stick with 6 piece Nalimov tables within hours because I had forgotten noatime so that every read access was also a (pointless!) write access.

I'm using windows pro. I've checked the drive I originally bought for TB's. It has the exact number of writes that it took to put the TB's on it. So basically no wear at all. The newer drive has the same writes on it and it's highly over-provisioned. It's write life is 1,600 TB IIRC. I suppose I could wear it out, but that's not likely to happen anytime soon.

[jp]

Stockfish-dev has failed to demonstrate a solution.

info depth 91 seldepth 102 multipv 1 score cp 359 nodes 6332175484277

Stockfish-dev seems to be doing a bit better here

info depth 84 seldepth 105 multipv 1 score cp 547 nodes 12623729511469

It then kept on failing high up to +8.96 when I stopped the search.

Another thing I'm not sure of: does SF really know anything when (in an easier position) its eval shoots to +56, but its PV hasn't demonstrated a solution? The end of the PV is close to converting (e.g. within 10 plies), but along the PV it's bounced between closer and further from converting, i.e. not even roughly monotonic progress. Maybe SF is just fooling itself?

[Zenmastur]

Anyone ever do this before or have any suggestions?

I've done it. I have such a backsolved tree for all of chess (which was used to build my chessmap.)

My suggestion is to resist the temptation of quantity over quality, all those 15ply games will provide a lot of noise, and variety isn't a good thing (the only reason you want variety is that there's some mainline you don't want to miss, so playing more lines would catch it; if you'd somehow miss it there's something else to fix before starting this.)

What has worked best is doing it the other way around, you build trunks of analysis lines of the highest quality moves that you can find, then you search for alternatives to those moves at decreasing depth. Starting with low depth will only make you realize later how much time you wasted in some lines (say, having 2000 lines for a poor move and 100 for the best move when it should be the other way around.)

Maybe there is a better way.

The problem with playing games is that you end up analyzing the same positions over and over again which wastes a bunch of computer time. The upside is it's easy to totally automate.

If I break all the opening lines in to FENs and remove duplicates then I could use mass analysis of the FENs without duplicating effort. To get better analysis of the more important lines I could break the FENs into batches based on frequency of occurrence, giving more time to those positions that occur more often. But, this has to be done manually. This is likely too much manual work, but wastes less computer time. I would like to associate the analysis with the games they belong to, but am unsure how this could be automated. Another down side is that other than the ending evaluation there is no method to determine the lines win/loss percentages.

I'm still undecided on exactly how I'm going to do this. Maybe I can do it both ways. Do the fen analysis first then build a book from them and then play games based on the book. Is there a way to limit the length of the book lines used in Cutechess or Littleblitzer so you don't end up with games that follow super deep book lines? Or, maybe some other program that runs engine matches can do this.

Regards,

Zenmastur

[mmt]

1. g4 d5 2. Bg2 Bxg4 3. c4 c6 4. Qb3 e6 5. Qxb7 Nd7 6. Nc3 Ne7 7. cxd5 exd5 8. d4 Rb8 9. Qa6 Rb6 10. Qd3 Ng6 11. h3 Be6 12. Nf3 Bd6 13. h4 h5 14. b3 Nf6 15. Bg5 O-O 16. e3 Re8 17. Kf1 Bg4 18. Ne1 Bb4 19. Na4 Rb8 20. Nc2 Be7 21. f3 Be6 22. Nc5 Bc8 23. Kf2 Nd7 24. Ne6 Qa5 25. Bxe7 Rxe7 26. b4 Qb6 27. Ng5 Ba6 28. Qa3 Rbe8 29. Bf1 Bxf1 30. Raxf1 Qc7 31. Qd3 a5 32. a3 axb4 33. axb4 Qd6 34. Rfg1 Nb6 35. Qf5 Nf8 36. Re1 Nc4 37. Nh3 Ra7 38. Qxh5 Ra2 39. Re2 Qe7 40. e4 dxe4 41. fxe4 Qd7 42. Ng5 f6 43. Nf3 Nd6 44. Nfe1 Rxe4 45. Qf3 Qe6 46. Rxe4 Nxe4+ 47. Kg1 Rb2 48. Rh2 Rb3 49. Nd3 f5 50. Qf1

[d]5nk1/6p1/2p1q3/5p2/1P1Pn2P/1r1N4/2N4R/5QK1 b - - 1 1

[Ovyron]

If I break all the opening lines in to FENs and remove duplicates then I could use mass analysis of the FENs without duplicating effort.

If you're going to analyze FENs you only need to give scores to leaf nodes and backsolve. Any position that isn't a leaf node will get a score coming from its mainline (the best moves you've found from both sides - so, say, you never put the engine to analyze 1.g4 d5 2.e3, because that position's score is given by the strongest future position that can be reached.)

As I mentioned in the other thread, this is done extremely fast with the Chess Openings Wizard software, you just import all your PGNs into the book, analyze all the leaf nodes by exporting them as EPDs, and end up with a fully backsolved tree, where every position will have a score coming from the strongest leaf node that can be reached, so you not only don't need to analyze early positions multiple times, you don't need to analyze them at all.

Another down side is that other than the ending evaluation there is no method to determine the lines win/loss percentages.

I've never found win/lose percentages useful, a position can be 80% white wins, 10% black wins, 10% draws, but after futher inspection it turns out those 10% can always be played, and there's no white defense, so it should hold some -9999.99 eval. It's just that those winning black moves are really hard to find and if you don't white has easy wins, but it's clear the 80-10-10 metric wasn't providing anything useful.

[Ovyron]

Qf1

I'm not sure if 50...Qf6 is better, I'm at the point where I'd use less time on this game by playing whatever instead of making sure what I'm playing is best (efficiency over optimality.)

1. g4 d5 2. Bg2 Bxg4 3. c4 c6 4. Qb3 e6 5. Qxb7 Nd7 6. Nc3 Ne7 7. cxd5 exd5 8. d4 Rb8 9. Qa6 Rb6 10. Qd3 Ng6 11. h3 Be6 12. Nf3 Bd6 13. h4 h5 14. b3 Nf6 15. Bg5 O-O 16. e3 Re8 17. Kf1 Bg4 18. Ne1 Bb4 19. Na4 Rb8 20. Nc2 Be7 21. f3 Be6 22. Nc5 Bc8 23. Kf2 Nd7 24. Ne6 Qa5 25. Bxe7 Rxe7 26. b4 Qb6 27. Ng5 Ba6 28. Qa3 Rbe8 29. Bf1 Bxf1 30. Raxf1 Qc7 31. Qd3 a5 32. a3 axb4 33. axb4 Qd6 34. Rfg1 Nb6 35. Qf5 Nf8 36. Re1 Nc4 37. Nh3 Ra7 38. Qxh5 Ra2 39. Re2 Qe7 40. e4 dxe4 41. fxe4 Qd7 42. Ng5 f6 43. Nf3 Nd6 44. Nfe1 Rxe4 45. Qf3 Qe6 46. Rxe4 Nxe4+ 47. Kg1 Rb2 48. Rh2 Rb3 49. Nd3 f5 50. Qf1 Ng3

[d]5nk1/6p1/2p1q3/5p2/1P1P3P/1r1N2n1/2N4R/5QK1 w - -

[Zenmastur]

Another down side is that other than the ending evaluation there is no method to determine the lines win/loss percentages.

I've never found win/lose percentages useful, a position can be 80% white wins, 10% black wins, 10% draws, but after futher inspection it turns out those 10% can always be played, and there's no white defense, so it should hold some -9999.99 eval. It's just that those winning black moves are really hard to find and if you don't white has easy wins, but it's clear the 80-10-10 metric wasn't providing anything useful.

Win/loss percentages based on human games are useless. They are more useful if they are based on reasonable time control engine matches.

[Zenmastur]

If I break all the opening lines in to FENs and remove duplicates then I could use mass analysis of the FENs without duplicating effort.

If you're going to analyze FENs you only need to give scores to leaf nodes and backsolve. Any position that isn't a leaf node will get a score coming from its mainline (the best moves you've found from both sides - so, say, you never put the engine to analyze 1.g4 d5 2.e3, because that position's score is given by the strongest future position that can be reached.)

As I mentioned in the other thread, this is done extremely fast with the Chess Openings Wizard software, you just import all your PGNs into the book, analyze all the leaf nodes by exporting them as EPDs, and end up with a fully backsolved tree, where every position will have a score coming from the strongest leaf node that can be reached, so you not only don't need to analyze early positions multiple times, you don't need to analyze them at all.

Do you own this software? How buggy is it? What does it cost?