First post. Below an evaluation of a DGT Pi game with a new chess engine named Drofa vs. my desktop CM9000 program. Unfortunately, I am unable to evaluate Pi engine Drofa, because the Pi displayed ....."abort" at Drofa move 13. Could have been an interesting Poisened Pawn Sicilian....except Drofa 12. .. Qc5?? pretty much ended Blacks chances. Maybe Drofa just gave up?
Posts : 17 Join date : 2020-11-22 Age : 68 Location : Bermondsey, London
Subject: Re: first post Mon Nov 23, 2020 12:04 pm
Hiya,
I’ll have a look at Drofa more closely, thanks for looking at it .....
Al.
Piman likes this post
Piman
Posts : 114 Join date : 2020-11-23
Subject: Re: first post Mon Nov 23, 2020 3:20 pm
Sc@lly wrote:
Hiya,
I’ll have a look at Drofa more closely, thanks for looking at it .....
Al.
Hi Sc@lly,
Nice to see you here.
Finally got the Pi to update (a wireless router at my setup had been changed to a switch, so Pi is now further from the main wireless router...maybe that was it). In any case, Drofa seems to have a problem...check it out and let me know.
Maybe it was because I was always calling it Dorfa which I corrected to Drofa after I typed my post
Sc@lly likes this post
Piman
Posts : 114 Join date : 2020-11-23
Subject: Re: first post Tue Nov 24, 2020 11:53 pm
OpenCritter
vs CM9000. Took a look at OpenCritter 1.1.38 (Brain) on the Pi vs CM9000 and ........it was a very long.......draw (I called it at move 100). A Tarrasch French and CM has a slight lead all along. At 42. Qxd4, CM will win an exchange, but Black has an extra pawn. White does win the pawn back, but OpenCritter pushes an "f" pawn down the board and eventually forces a R+B vs 2B draw. New program OpenCritter appears to be Fully Functional on the Pi. A good addition to the Pi, mid-level in the Weiss, Teki, Tunguska range, maybe around 2700, which is .......quite strong.
A look at the Arasan update on Pi vs CM9000. Below, Arasan gets a win. But, Arasan did not look good early....I was actually concerned about the Pi not functioning properly (Uber engine Arasan on Pi has always handily beaten the ~2700 Elo (on PC) CM9000). But here, for nearly 40 moves CM9000 calculated the better game according to SF analysis....CM9000 should have drawn this game. In fact, the game was close to a repetition draw around move 50, when Arasan opted out with 52. .. Ke7. And then, Arasan gets a break: CM can't see far enough into the future with 59. Rc2? Seems like a reasonable move.....but it actually loses a badly placed Bishop at move 64. After 77. Rxf1?? it's Game Over, as the Black "e" pawn must now be stopped with the Rook (Although SF says 71. g5? is already The End). Took 100 moves for mate....very out of character for Arasan compared to virtually all other games against CM9000. However, Arasan is FF to mate on the Pi....absolutely consistent in this respect.
Posts : 17 Join date : 2020-11-22 Age : 68 Location : Bermondsey, London
Subject: Re: first post Fri Nov 27, 2020 12:42 pm
Hi Charlie,
Thanks for continuing to test my Self Compiled RPi Engines. I had a quick look at Drofa thinking it may be the “4 core” issue you sometimes have, but it only uses 1 core. However OpenCritter uses 4 cores at default, so I’m surprised that worked on your set-up.
When I get some spare time I will add 4 levels to chose the numbers of cores with OpenCritter and test Drofa myself before deciding what to do with it. I have my fingers in too many pies at the moment.
Cheers,
Al.
Piman likes this post
Piman
Posts : 114 Join date : 2020-11-23
Subject: Re: first post Sat Nov 28, 2020 3:28 am
Cinnamon 2.3.4
A look at Cinnamon update...this one 2.3.4. I had an overclocked (25MHz) Excalibur GM out on the table, so I figured that would be a good opponent. A wild game, with big eval swings.....but then that old Cinnamon bugaboo: Cinnamon 2.3.4 in Brain mode locked up the Pi at move 11:
The previous game, a Cinnamon update Pi lock up, has been seen consistently since Cinn update V2.1. Therefore, for a sanity check, I ran the "OLD" version known as Cinnamon IGWT (apparently, a pious version). Below vs. Excalibur GM 27 MHz, I forced the previous Old Indian opening and IGWT quickly varied with 5. .. Nf2? (Cinn 2.3.4 gxf4 is stronger). So, IGWT builds up another nice advantage, but then goes pawn grabbing with 17-18 Qa6, QxP? GM27, with only the Q developed actually takes a small eval lead according to SF. By move 24, GM has a +2 eval lead, but appropriately blunders with 24. Qg5+? (develop a piece with Neg2). Then after GM 29. Kf1?? it's Game Over....massive trades will result in a B vs N endgame, but Black has a winning "a" pawn. IGWT finally catches on and starts the pawn push and wins the Bishop at move 45. By move 54, it's an easy Q for Black and mate in 10.....and something bizarre 🤮🤮🤮🤮 happens. Cinnamon IGWT freezes the Pi. A first....I have never seen the original Cinnamon program freeze the Pi. I am clueless, IGWT has always worked flawlessly to mate:
[Event "30 sec/move"] [Black " Cinnamon IGWT 2019 Brain"] [Date "Nov 28, 2020"] [Result "*"] [Round "IGWT also locks up the pi!!!"] [Site "Exotic"] [White "GM27 MHz"]
A little more than a year ago, Cinnamon IGWT beat a RS2250 in 49 moves. I replayed that opening, forcing the Nimzowitsch against RS2250. Not easy, as takebacks and move times differences from the original, had IGWT calculating new moves. At 9. .. Bb6, I let this game go its own way (originally 9. .. Ba7). Long story short, Cinnamon IGWT was FF this game, no Pi lock up problems, finished all the way to a mate...60.h7 Rf1# 0-1. I did notice one thing unusual : When IGWT gets close to mate, the program takes MORE than 30 seconds for some moves in the mate sequence. But....the Pi does not allow time averaging or calculating longer than the selected move time; the Pi is very rigid: select a move time and the engines, virtually all of them, must move exactly at that move time (or in Brain mode, before ....but NEVER after). Maybe there is some little quirky timing issue in the IGWT mating sequence, that can lock up the pi...but the previous Cinnamon game is the first time I've seen IGWT lock up the Pi. The good news is that Pi lock up seems to occur when the game is already a clear win for Cinnamon IGWT.
Today's game is below, SF eval says RS2250 23. Qb3? is Game Over....quick win by Cinnamon.
A look at Combusken (Normal) update on Pi vs CM9000....and a very evenly calculated repetition draw. Engines generally eschew threepeat draws, but not in this game. An English and the game shows that Combusken is probably calculating at about 2700 on the Pi, whereas that number is over 3000 on CCRL.
Combusken update is FF on the Pi....like Arasan, Combusken updates are very reliable and fully functional every single update. Excellent Pi chess engine.
CT800 (Normal) vs CM9000....maybe a bit overmatched. CT versions have improved over 200 points on CCRL (to 2400+), so maybe this version will have a chance vs CM9000....but it will be tough.
Ow! That hurt. The good news is that CT800 update is FF to mate on the Pi. The bad news is that CT had a rough time against CM9000: moves like 15. Nc2 (Nc4), 17. Bh3? and 20. e4?? (this just loses a pawn) equal Game Over. Very strong game by CM9000 on PC; took full advantage of weak moves. CT800 is not a powerhouse on the Pi vis a vis the uber engines, however it is solid "human" level program; Fully Functional with selectable Elo levels...I give it .
[Event "30 s/mv"] [Black "CM9000"] [Date "Dec 5, 2020"] [Result "0-1"] [Round "CT800 in a rough loss"] [Site "Performance book"] [White "CT800 V1.41 32 bit Normal"]
An oldie, Fruit Reloaded (2600 Brain) vs CM9000. I was a bit surprised as Fruit had a real tough time with this Sicilian. SF eval deems Fruit calculation 8. .. Qc7? a mistake (0-0 is book). CM attacked extremely well and SF says: "lose the exchange with 21. .. Rxd4 to relieve the pressure". But Fruit says: "21. .. Qb6" which is ?? and Game Over.
Fruit Reloaded is FF on the Pi.....and is probably a lot stronger than its subpar performance in this Sicilian. Another nice job by CM9000 on PC.
No it can't....I gave Fruit reloaded 15th Anniversary a second chance with the White pieces and a better book (Performance book) vs CM9000, but after 14-15, exd5?-cxd5??, SF eval claims White is lost. So, Fruit Reloaded simply is in the group of lower level Pi engines, i.e., humans may have a small chance. Additionally, as previously mentioned in the engineinfo file: moves too fast in Move in X...this appears to still be the case. Fruit simply answers instantaneously, too often (especially, on moves that it needed to think more). Also, I had Pi shutdown difficulties after the previous game, which I didn't mention....but when I had Pi start up problems in this game....I am now convinced that Fruit has some small code issues with full functionality on the Pi, but you eventually get the thing to play a full,game without issues.
[Event "30 s/mv"] [Black "CM9000"] [Date "Dec 7, 2020"] [Result "0-1"] [Round "Fruit, just not that strong"] [Site "Performance book"] [White "Fruit reloaded 15th Anniversary 2600 Brain"]
A look at the recent Arasan update vs CM9000. The exotic book results in a non-optimal (for White) Danish Gambit and the stronger Arasan never got any traction. CM9000 calculated evenly the entire game and by move 30, a Bishop of opposite colors draw was in clear view...I called it at move 56.
Update, Stash 25.0 appears to be FF on the Pi. However, Stash is still not in the CM class, as seen in the below evaluation game: 16. Bxb5?? is Game Over (SF says 16. h3...although SF also says previous 15. Qe1 is better than Nd5). Stash: a good "human level" program on the Pi.
Gave Igel update (Level 20, Brain) a try against CM9000...which has had a very good streak of games against Pi engines. I gave Igel a decent book (GM2600) and the opening was a Closed English for 10 moves. CM hung in there pretty good for 21 moves, but...22. .. Nd7?? was a bad mistake...Game Over. However, Igel must find 23. Ne7+! ....which Igel does. If Igel and CM9000 were humans, CM would have stood up at the board at move 27...........and started clapping. This is one of the finest attacks that I have seen on the Pi...ever. Igel is FF to mate on the Pi......and stronger than Bobby Fischer.
Posts : 115 Join date : 2020-11-26 Location : Acworth, GA (USA)
Subject: Re: first post Fri Dec 11, 2020 9:59 am
I hope you don't mind, thanks for these games.
Sc@lly and Piman like this post
Piman
Posts : 114 Join date : 2020-11-23
Subject: Re: first post Fri Dec 11, 2020 2:38 pm
To AdminX:
You know X, I really like your post. Very professional. In fact, so professional and of such high quality....that I think you would be perfect for conducting future "Pi update" evaluations.
well done.
Whadya think Scal? He's actually quite good.. OpenCritter, Arasan, Combusken, Cinnamon IGWT vs a 2250....all in one post. Excellent!
Sc@lly
Posts : 17 Join date : 2020-11-22 Age : 68 Location : Bermondsey, London
Subject: Re: first post Fri Dec 11, 2020 3:09 pm
Hi Guys,
Yes very nice, just shows what can be done when you know how to use the tools at your disposal.
I had no idea anything like this was possible.
Cheers,
Al.
Piman
Posts : 114 Join date : 2020-11-23
Subject: Re: first post Sat Dec 12, 2020 12:32 am
Igel 2.9-dev-2 CUSTOM
Had to do it: Can Igel win the Danish Gambit against CM9000 (the one that Arasan only draws)? Yup. Same moves as Arasan/CM..until Igel calculates variation 14. Nh4 (although, SF actually prefers the Arasan 14. Bc4 ). The game now changes and small mistakes by CM in the middlegame give Igel a decent lead by move 30. Then, CM 30. .. g6?? (now, one move too late) is Game Over. And "thar she blows": Igel 31. Ne7+. Th...Th...Th....That's all folks.
A look at Rodent4 update vs CM9000. A very close game out of a QGD, drawn until CM 42. .. Rf8? Now, Rodent wins an exchange and has a clear win by move 60. But Rodent4 then moves the K to the Kingside to apply mate....but mate can't be done that way: the Bishop covers h7....NO way you're getting in there. Rod still has a late chance with 71. Rc7!...but 71. Rb7? = Draw. Good game: Nice attack by Rodent and CM capitalizes on a mistake to draw. These two are very close in strength.
Excellent game by Rubi Update (Brain & 4 Cores!) vs CM9000. A "Fun" Nimzo-Larsen Opening and RubiChess has the Black pieces completely tied up by move 30 and the Bishop pair aimed directly at the King...Game Over.
Rubi, with Brain and four cores, is FF on the Pi. Unfortunately, the author, apparently, will never fix the NNUE version "in Brain, use every single second of move time, every single move...even at mate in 1" issue; this makes the engine very user unfriendly.
A look at Stash upgrade (Normal) vs Tasc R30 2.2. Stash is not quite at CM9000 level, so I let Tasc have a shot...shoulda been an easy win. But, it wasn't; the venerable Tasc gave Stash all it could handle in a Pirc/Czech for 50 moves. In fact, SF analysis says that had Tasc found 50. Ne3!, Tasc probably would have won, but 50. Qd3? gave Stash a second life. And then, disaster for Tasc: 52. g3?? Game Over. This breathing room move for White is a loser, as Stash eventually is up a Rook and Tasc graciously Resigns at move 69 .
Stash update is FF on the Pi...a very good "human level" engine on the Pi (Well, let's say a very good Master Level engine on the Pi)....it's way too strong for me.
On CCRL, Supernova has shown an outstanding 500 Elo improvement since V1, so I gave SN a tough one: CM9000. Supernova, absolutely held its own for 30 moves and then blunder 32. Rxb5?? (Bxg4 holds the game). At move 63, with mate in sight Supernova locked up the Pi....not a big deal, it was mate in 3 anyway. Supernova is a very entertaining Pi engine that moves super quick (and often) in Brain mode (maybe too often); but makes for a very fast paced game. I give Supernova and maybe soon, with continuing improvements, Supernova will be at the CM level.
[Event "30 sec/mv"] [Black "CM9000"] [Date "Dec 18, 2020"] [Result "0-1"] [Round "Supernova not at CM level, yet"] [Site "Carlsen2 book"] [White "Supernova 2.2 Brain"]
Weiss: another author who has made terrific progress in strength ~3000 on CCRL. Weiss used to have trouble with CM9000...not any more. Can this Weiss update win with the handicap of the "Fun" Book and Black pieces?
Outstanding game...best I've seen from Weiss....and Weiss update 4 Cores, Normal is FF to mate on the Pi. By move 18, material had reduced to the salient Q+2R+minor on both sides. This was a draw...just a predictable draw. But Weiss simply outmaneuvered CM9000, slowly but surely, in the next 15 moves to have a measurable advantage by move 32....very well done. SF analysis says CM 33. Rd2? and 34. d5?? were Game Over...a -4 eval lead for Weiss. Weiss wins the exchange and in a Kt+3P (CM) vs R+3P endgame...puts on an absolute clinic on how to smother an opponent's position.
Home 
which I corrected to Drofa after I typed my post 
).