Cheng 4.40

[mar]

Any estimates of improvement against 4.39 ?

I'm hoping for ~30 elo, could be less or more (especially depending on TC), I typically get much less from self-play (less than a half), so I leave this to independent testers.
In hyperbullet selfplay I got about +77 against 4.39 (10k 10+0.1s games), but it will be less against other engines, I'm pretty sure of that.

[mar]

Not easy to explain:
A delay process with pawn endgames.
Or maybe in better english ... a sneaky / creeping process with pawn-structurs in endgames.

Often I am thinking after Cheng switched rooks, bishops or knights the pawn-positions after ... must be lose.
Often I am thinking why Cheng don't try to bring rooks on better positions.
2-line, 7-line on the board for rooks, the rooks are not aggressively enough in endgames.

I'm aware that my passer evaluation (among other things) is not great, perhaps that's what you mean, Frank?

[Frank Quisinsky]

Hi Martin,

I am not a programmer, looking more in statistics, generated from big databases or searching games, engines forced a faster way to win (Andscacs, Booot, SlowChess, Hakkapelitta, Spark, Fizbo, Wasp with his very aggressive pawns do), Aristarch or ETChess in the past can do, Glaurung, Gothmog, AnMon, Phalanx, Gromit, Junior and often Hiarcs can do. Schooner can do with very aggressive knight, bishophs (engine try to hold the minor-pieces, very own way Schooner like to do). Mid-game and king safty with many pieces on board is most important for myself because I am using engines most for opening analyzes. A great combination with many pieces on board ... that's what I like most.

I am today with around 2050 Elo not a very strong player, but often I am thinking I am playing with 2150 Elo, often I am thinking 1900 and on bad days not more as 1600. I can see with statistics and the knowledge I have with a lot of holes only.

But yes, after all I have the opinion Cheng lost Elo in endgames and the reason is the pawn structure, often build after a bad trade pieces in late mid-game and often I am thinking endgames with rooks can be better.

But this isn't my main point (end-games).
What I like to see ... Cheng is great and for myself in the group of mid-game attacker!
Same for Cheese ... late place in FCP Qualify-Tourney-2021.
But with Wasp, Cheng I have the most fun here with Cheese ... again Elo isn't important, not for myself!
Topple plays also very nice games.

Best
Frank

[Archimedes]

I didn't include an Android build this time, because of what Aart wrote about the new security changes in Android - assuming it only affects running engines from SSDs (for now), I probably should've added an Android compile - but since I don't own an Android device at the moment (plus I was lazy to set up NDK), I decided to pass on Adroid this time. Someone else can build Android binaries for sure.

Android NDK uses the clang compiler. And with it, you have to change very often parts the source code for proper compiling (but this can be done very often in a minimal invasive way). For Android compiles, i simply use ndk-build from Android NDK.

Cheng 4.40:
https://app.box.com/s/235inq8ab3rlu0objnzeaeiunknlaz4r

[Dann Corbit]

Whether or not Cheng ever becomes top 10, I am a big fan of the engine.
I think it has beautiful code.
It is clear.
It is easy to read.
It reminds me of Bas Hamstra's Tao, which was similar in being very well put together.

I am a fan.

[Wolfgang]

Any estimates of improvement against 4.39 ?

I'm hoping for ~30 elo, could be less or more (especially depending on TC), I typically get much less from self-play (less than a half), so I leave this to independent testers.
In hyperbullet selfplay I got about +77 against 4.39 (10k 10+0.1s games), but it will be less against other engines, I'm pretty sure of that.

Looks like ~50 to 70 ELO @ our 40/4 list
https://cegt.forumieren.com/t1404-testi ... ivisto-4-0

Cheng 4.40 = ~2859
Cheng 4.39 = 2793

[mar]

Android NDK uses the clang compiler. And with it, you have to change very often parts the source code for proper compiling (but this can be done very often in a minimal invasive way). For Android compiles, i simply use ndk-build from Android NDK.

Cheng 4.40:
https://app.box.com/s/235inq8ab3rlu0objnzeaeiunknlaz4r

Thanks, I thought NDK uses gcc, but I guess it doesn't matter - I compile with clang for Mac, so should work just fine.

[mar]

Dann, we both know that's not quite true The code is unnecessarily verbose and certainly not "beautiful", but at least it doesn't crash when ucinewgame is missing or when illegal fen is passed to the engine, because there's simply no reason to

Wolfgang: that's probably too good to be true, but at least it shows some real improvement, which is good

[Dann Corbit]

Dann, we both know that's not quite true The code is unnecessarily verbose and certainly not "beautiful", but at least it doesn't crash when ucinewgame is missing or when illegal fen is passed to the engine, because there's simply no reason to

Wolfgang: that's probably too good to be true, but at least it shows some real improvement, which is good

There is more than one kind of beauty. There is the beautiful, terse elegance of Olithink, and the expressive beauty of Cheng.
It is like the question:
Which is more beautiful, the SR71 Blackbird, an arrangement of cherry blossoms, or the voice of Luther Vandross?
The answer is that they are all equally beautiful, but in very different ways.

There is ugly code too. I have read code so ugly it made me genuinely angry.

[Angle]

Cheng 4.40 is playng very well in my current tourney:
http://talkchess.com/forum3/viewtopic.php?f=6&t=76231
https://www.youtube.com/watch?v=aN9hAzWRwPI