ababoin07 wrote:

ArnoHu wrote:

GoK 6.530 plays a perfect game at 10 sec think time vs. White Dove (P2) - 99% vs. 92% accuracy: https://lichess.org/A3aEfB1c#102

When GoK vs stockfisch 16 XD

ababoin07 wrote:

ArnoHu wrote:

GoK 6.530 plays a perfect game at 10 sec think time vs. White Dove (P2) - 99% vs. 92% accuracy: https://lichess.org/A3aEfB1c#102

When GoK vs stockfisch 16 XD

HasiLover_Test wrote:

ababoin07 wrote:

ArnoHu wrote:

GoK 6.530 plays a perfect game at 10 sec think time vs. White Dove (P2) - 99% vs. 92% accuracy: https://lichess.org/A3aEfB1c#102

When GoK vs stockfisch 16 XD

I could run a GoK - Stockfish 16 game, but then maybe Stockfish only has 1 or 2 seconds per move and GoK is competition mode

ArnoHu wrote:

HasiLover_Test wrote:

ababoin07 wrote:

ArnoHu wrote:

GoK 6.530 plays a perfect game at 10 sec think time vs. White Dove (P2) - 99% vs. 92% accuracy: https://lichess.org/A3aEfB1c#102

When GoK vs stockfisch 16 XD

I could run a GoK - Stockfish 16 game, but then maybe Stockfish only has 1 or 2 seconds per move and GoK is competition mode

This was the most recent one I let it play back in February: https://lichess.org/5kqGaKJK (level 7 where GoK still has chances; level 8 would be interesting as well, I have not done that yet)

ababoin07 wrote:

So now there is a functional ‘Undo’ ('U' key)

ArnoHu wrote:

ababoin07 wrote:

So now there is a functional ‘Undo’ ('U' key)

GoK Blitz 2 (black) vs. Destructor on TW, 98% vs. 90% accuracy: https://lichess.org/7h6QtO8Q#86

Destructor had good accuracy, except for the one blunder. It became quite slow during endgame. I thought it would get faster with fewer pieces.

ababoin07 wrote:

I will reprogram Destructor from A to Z, too many bugs, so if you have ideas for generating ultra-fast pseudo-legal movements, I am a taker!

birdracerthree wrote:

https://lichess.org/study/FYcWqN5E/gHxC0UjC - GoK selected Bxg4 (it still thinks it is losing), WD plays Nxg4 (king tropism can't tell the difference between the two)

https://lichess.org/study/FYcWqN5E/aBHTfs6m - GoK doesn't reach the depth to see the correct move, but it sees its partial effect when it selected …fxe6 as black

https://lichess.org/study/FYcWqN5E/rjdN55I7 - No engine finds the correct move

https://lichess.org/study/FYcWqN5E/UF11KrPx - GoK sees this at the expected depth of 8. WD reaches ply and does not see the solution.

https://lichess.org/study/FYcWqN5E/p7Wrp4XT - GoK plays Kh4, but its evaluation doesn't change.

https://lichess.org/study/FYcWqN5E/XV0qmrTs - GoK and Element find the solution, but WD does not. Element is faster in this position than WD (that should never happen, esp. in endgame where a TTable is extremely useful).

Bonus : 8/3k4/2p1Rp2/p1pP1P2/2P4P/2P1K3/8/r7 b - - 0 41 - GoK finds the correct move during depth 13

https://lichess.org/study/FYcWqN5E/1jj0Hz6b - White Dove sees the checkmate in roughly 5 seconds. GoK does not see it on competition mode. I guess the aggressive extensions come in handy here.

oiseauracertrois wrote:

ababoin07 wrote:

I will reprogram Destructor from A to Z, too many bugs, so if you have ideas for generating ultra-fast pseudo-legal movements, I am a taker!

Your Transposition Table implementation is probably incorrect. I suggest removing it and adjusting the simpler things in the code. Element reached 6+8 with “only” alpha-beta pruning and solid move ordering in the search (MVV-LVA+PSTs). Make sure everything still works when you make a change.

Is your engine minimax or negamax? Please make sure everything works before you implement anything new. I like to run self-play games with my engine (this inspired White Dove to do the same). As you can see, Element is not foolproof. Here is the most recent self-play game - https://lichess.org/study/7hMXEJDB/gzd0h43G#59

ArnoHu wrote:

birdracerthree wrote:

https://lichess.org/study/FYcWqN5E/gHxC0UjC - GoK selected Bxg4 (it still thinks it is losing), WD plays Nxg4 (king tropism can't tell the difference between the two)

https://lichess.org/study/FYcWqN5E/aBHTfs6m - GoK doesn't reach the depth to see the correct move, but it sees its partial effect when it selected …fxe6 as black

https://lichess.org/study/FYcWqN5E/rjdN55I7 - No engine finds the correct move

https://lichess.org/study/FYcWqN5E/UF11KrPx - GoK sees this at the expected depth of 8. WD reaches ply and does not see the solution.

https://lichess.org/study/FYcWqN5E/p7Wrp4XT - GoK plays Kh4, but its evaluation doesn't change.

https://lichess.org/study/FYcWqN5E/XV0qmrTs - GoK and Element find the solution, but WD does not. Element is faster in this position than WD (that should never happen, esp. in endgame where a TTable is extremely useful).

Bonus : 8/3k4/2p1Rp2/p1pP1P2/2P4P/2P1K3/8/r7 b - - 0 41 - GoK finds the correct move during depth 13

https://lichess.org/study/FYcWqN5E/1jj0Hz6b - White Dove sees the checkmate in roughly 5 seconds. GoK does not see it on competition mode. I guess the aggressive extensions come in handy here.

The last board is cool, great how fast White Dove finds it.

GoK did not support discovered checks within quiescence search so far, that is why it took longer. I created a quick prototype, and it finds it in less than a second.

One thing to note, GoK continues searching for quicker checkmates if checkmate is found during quiescence search, so you must check logfile for when the move is first found with checkmate evaluation.

birdracerthree wrote:

Do you think Element has something that prevents TTables from working? I found this pseudocode, but it didn't work when I tried to put it in Element. Does anyone have any ideas? Why is the line

“if α ≥ β then
return ttEntry.value”
in the code?

ababoin07 wrote:

2 hours lost fighting with the generation of movements, balance: I had forgotten a block ‘delete all of’

ababoin07 wrote:

oiseauracertrois wrote:

ababoin07 wrote:

I will reprogram Destructor from A to Z, too many bugs, so if you have ideas for generating ultra-fast pseudo-legal movements, I am a taker!

Your Transposition Table implementation is probably incorrect. I suggest removing it and adjusting the simpler things in the code. Element reached 6+8 with “only” alpha-beta pruning and solid move ordering in the search (MVV-LVA+PSTs). Make sure everything still works when you make a change.

Is your engine minimax or negamax? Please make sure everything works before you implement anything new. I like to run self-play games with my engine (this inspired White Dove to do the same). As you can see, Element is not foolproof. Here is the most recent self-play game - https://lichess.org/study/7hMXEJDB/gzd0h43G#59

I am reprogramming entirely destructor; indeed, I have plenty of ideas to optimize everything and fix the bugs but they are too complicated to implement on an already existing project: so I am reprogramming the engine….