Scratch Chess Engine - Game of Kings

birdracerthree

ArnoHu wrote:
birdracerthree wrote:
S_P_A_R_T_Test wrote:
birdracerthree wrote:
birdracerthree wrote:
ArnoHu wrote:
Have you tried lately? The over-aggressiveness was caused by futility pruning which was still in there from the classic implementation, I fixed that before I ran the 10-games-without-mistakes series (https://lichess.org/study/Z5N8Io1T) some days ago. NNUE by far does not play as aggressive any more. In my tests it has not lost a single game since. There was one draw, also my fault because I had screwed 3-fold-rep detection.
Fine, I’ll test it myself

All jokes aside, is it just me or are people speaking in a ever-so-slightly aggressive tone?
Tested : https://lichess.org/study/iU1kkNli/SdwCBtaF#66
Guess who scored another brilliant move…
Edit : Game 2 : https://lichess.org/study/iU1kkNli/yVJdqnId#43

White Dove finds the winning move in 0.4 seconds! And due to it's, um, slightly aggressive piece values and king safety evals, scores it as -14 lol. (Even SF says it's -6, but WD sees less, but just has a way more sensitive function lol.
That's not due to material scaling?

https://lichess.org/study/iU1kkNli/SdwCBtaF#66
No mistake according to Stockfish?

I have labeled Nf4 as a brilliant move.

ArnoHu

birdracerthree wrote:
ArnoHu wrote:
birdracerthree wrote:
S_P_A_R_T_Test wrote:
birdracerthree wrote:
birdracerthree wrote:
ArnoHu wrote:
Have you tried lately? The over-aggressiveness was caused by futility pruning which was still in there from the classic implementation, I fixed that before I ran the 10-games-without-mistakes series (https://lichess.org/study/Z5N8Io1T) some days ago. NNUE by far does not play as aggressive any more. In my tests it has not lost a single game since. There was one draw, also my fault because I had screwed 3-fold-rep detection.
Fine, I’ll test it myself

All jokes aside, is it just me or are people speaking in a ever-so-slightly aggressive tone?
Tested : https://lichess.org/study/iU1kkNli/SdwCBtaF#66
Guess who scored another brilliant move…
Edit : Game 2 : https://lichess.org/study/iU1kkNli/yVJdqnId#43

White Dove finds the winning move in 0.4 seconds! And due to it's, um, slightly aggressive piece values and king safety evals, scores it as -14 lol. (Even SF says it's -6, but WD sees less, but just has a way more sensitive function lol.
That's not due to material scaling?

https://lichess.org/study/iU1kkNli/SdwCBtaF#66
No mistake according to Stockfish?
I have labeled Nf4 as a brilliant move.

Oh, I did not see. I thought you still refer to over-aggressiveness.

HasiLover_Test

S_P_A_R_T_Test wrote:
White Dove v7.25 Has Been Released!

This update mostly addresses some of the issues in the games against GoK, but also comes with a new (and I think really cool) UI!

(I know I said yesterday that I wouldn't be doing many updates, but I guess there may be some exceptions lol.)

Wow, the UI is very nice and its performance is very good.

birdracerthree

HasiLover_Test wrote:
S_P_A_R_T_Test wrote:
White Dove v7.25 Has Been Released!

This update mostly addresses some of the issues in the games against GoK, but also comes with a new (and I think really cool) UI!

(I know I said yesterday that I wouldn't be doing many updates, but I guess there may be some exceptions lol.)
Wow, the UI is very nice and its performance is very good.

Raise your hand if you have the worst UI out of everyone *raises hand*.

Element’s UI has never been a priority because it doesn’t effect playing strength. Maybe it will be a nice change of pace for me.

ArnoHu

HasiLover_Test wrote:
S_P_A_R_T_Test wrote:
White Dove v7.25 Has Been Released!

This update mostly addresses some of the issues in the games against GoK, but also comes with a new (and I think really cool) UI!

(I know I said yesterday that I wouldn't be doing many updates, but I guess there may be some exceptions lol.)
Wow, the UI is very nice and its performance is very good.

Yes, looks cool, nice every engine has its own “visual” character.

birdracerthree

ArnoHu wrote:
birdracerthree wrote:
ArnoHu wrote:
birdracerthree wrote:
S_P_A_R_T_Test wrote:
birdracerthree wrote:
birdracerthree wrote:
ArnoHu wrote:
Have you tried lately? The over-aggressiveness was caused by futility pruning which was still in there from the classic implementation, I fixed that before I ran the 10-games-without-mistakes series (https://lichess.org/study/Z5N8Io1T) some days ago. NNUE by far does not play as aggressive any more. In my tests it has not lost a single game since. There was one draw, also my fault because I had screwed 3-fold-rep detection.
Fine, I’ll test it myself

All jokes aside, is it just me or are people speaking in a ever-so-slightly aggressive tone?
Tested : https://lichess.org/study/iU1kkNli/SdwCBtaF#66
Guess who scored another brilliant move…
Edit : Game 2 : https://lichess.org/study/iU1kkNli/yVJdqnId#43

*snip*
That's not due to material scaling?

https://lichess.org/study/iU1kkNli/SdwCBtaF#66
No mistake according to Stockfish?
I have labeled Nf4 as a brilliant move.

Oh, I did not see. I thought you still refer to over-aggressiveness.

No, I ran a game to test GoK NNUE and your point was proven.

I didn’t realize GoK NNUE was significantly better than it was previously. The last time I ran a test game with Element, I was playing GoK 7.010.

I will put this out now, NNUE stands for “Efficiently Updatable Neural Network”. That is all I have to say about that.

Last edited by birdracerthree (July 19, 2024 04:22:59)

ArnoHu

I decided to give NNUE buttons a flat style:

Last edited by ArnoHu (July 19, 2024 04:38:58)

ArnoHu

ArnoHu wrote:
HasiLover_Test wrote:
S_P_A_R_T_Test wrote:
White Dove v7.25 Has Been Released!

This update mostly addresses some of the issues in the games against GoK, but also comes with a new (and I think really cool) UI!

(I know I said yesterday that I wouldn't be doing many updates, but I guess there may be some exceptions lol.)
Wow, the UI is very nice and its performance is very good.

Yes, looks cool, nice every engine has its own “visual” character.

10sec test game against GoK NNUE (white) - did LMR changes limit WD's checkmate detection? It ran into mate-in-3:
https://lichess.org/WfU780Gr#37

ArnoHu

birdracerthree wrote:
ArnoHu wrote:
birdracerthree wrote:
ArnoHu wrote:
birdracerthree wrote:
S_P_A_R_T_Test wrote:
birdracerthree wrote:
birdracerthree wrote:
ArnoHu wrote:
Have you tried lately? The over-aggressiveness was caused by futility pruning which was still in there from the classic implementation, I fixed that before I ran the 10-games-without-mistakes series (https://lichess.org/study/Z5N8Io1T) some days ago. NNUE by far does not play as aggressive any more. In my tests it has not lost a single game since. There was one draw, also my fault because I had screwed 3-fold-rep detection.
Fine, I’ll test it myself

All jokes aside, is it just me or are people speaking in a ever-so-slightly aggressive tone?
Tested : https://lichess.org/study/iU1kkNli/SdwCBtaF#66
Guess who scored another brilliant move…
Edit : Game 2 : https://lichess.org/study/iU1kkNli/yVJdqnId#43

*snip*
That's not due to material scaling?

https://lichess.org/study/iU1kkNli/SdwCBtaF#66
No mistake according to Stockfish?
I have labeled Nf4 as a brilliant move.

Oh, I did not see. I thought you still refer to over-aggressiveness.
No, I ran a game to test GoK NNUE and your point was proven.

I didn’t realize GoK NNUE was significantly better than it was previously. The last time I ran a test game with Element, I was playing GoK 7.010.

I will put this out now, NNUE stands for “Efficiently Updatable Neural Network”. That is all I have to say about that.

Yeah, I hope I made it as efficient as possible on Scratch.

Still, on slow systems… it played (and won) at ~150 NPS (S3) against Shallow Blue in the Depth 4 tournament.

Last edited by ArnoHu (July 19, 2024 05:17:37)

-089-

birdracerthree wrote:
-089- wrote:
birdracerthree wrote:
birdracerthree wrote:
*snip* as you like to say

Currently testing e4 e5 gameplay now.
Update : SB runs into Fried Liver (although only GoK plays it and it doesn’t do it correctly), plays dxc6 in Scotch and Four Knights Scotch and gets to around +1.
you're right, I didn't catch your joke at all lol it was 5 am and I had just woken up before I wrote that post. was really tired, I'm not a morning person haha

so as far as I understand what you're saying, it survived the fried liver but lost some material, right? I only had time for 2 test games myself (insanely slow device, 2-3k Nps max), 4+0 faced a fried liver against the 1500 chesscom bot. it not only held, but went on to gain a massive material advantage before unfortunately running into a threefold trap. could you maybe give me the PGNs or lichess links? if you have them that is, no worries if you don't
I forgot to respond, so I will now
1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Ng5 d5 5. exd5 Nxd5 6. Nxf7 Kxf7 7. Qf3+ Ke7 8. Bxd5 Nd4 9. Qf7+ Kd6 10. Nc3 Kc5 11. b4+ Kxb4 12. Rb1+ Ka5 13. Bxb7 Nxc2+ 14. Kd1 Bxb7 15. Qc4 Ne3+ 16. fxe3 Bf3+ 17. gxf3 Qd7 18. Rb5+ Qxb5 19. Qxb5#
With SF 11 assistance as usual

thanks, I see. in that one test game I had it didn't come to 6. Nxf7, is that a standard move? I will deal with that as soon as I get back

ababoin07

I am disgusted, there is a bug in the generation of Destructor 2 moves, for the ‘en passant’, if the king is in check, and the ‘en passant’ cancels the check, it is a legal move, yet this move is not considered legal, this is the only known case where there is a generation bug. If you see another bug of generation , please let me know. I do not think that Destructor 2 will have a library of openings, it has a very weird style of play that is not compatible with many openings, moreover, hee will be very aggressive, which is I think better (I would inquire)

Last edited by ababoin07 (July 19, 2024 07:07:26)

ArnoHu

I just implemented piece drag & drop (GoK NNUE Dev Version: https://scratch.mit.edu/projects/1033757275/ ), and it was not trivial due to GoK's legacy user input handling. I welcome tests, esp. some fuzz testing e.g. by wildly dragging / clicking pieces and clicking squares, but also complete games. Please let me know how it works. Thank you!

GoK originally was pen-stamp-pased, and every mouse click was handled by the stage, there were no click events on sprites, and I had to keep using these message interfaces for backward compatibility.

birdracerthree

-089- wrote:
birdracerthree wrote:
-089- wrote:
birdracerthree wrote:
birdracerthree wrote:
*snip* as you like to say

Currently testing e4 e5 gameplay now.
Update : SB runs into Fried Liver (although only GoK plays it and it doesn’t do it correctly), plays dxc6 in Scotch and Four Knights Scotch and gets to around +1.
you're right, I didn't catch your joke at all lol it was 5 am and I had just woken up before I wrote that post. was really tired, I'm not a morning person haha
*snip*
I forgot to respond, so I will now
1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Ng5 d5 5. exd5 Nxd5 6. Nxf7 Kxf7 7. Qf3+ Ke7 8. Bxd5 Nd4 9. Qf7+ Kd6 10. Nc3 Kc5 11. b4+ Kxb4 12. Rb1+ Ka5 13. Bxb7 Nxc2+ 14. Kd1 Bxb7 15. Qc4 Ne3+ 16. fxe3 Bf3+ 17. gxf3 Qd7 18. Rb5+ Qxb5 19. Qxb5#
With SF 11 assistance as usual
thanks, I see. in that one test game I had it didn't come to 6. Nxf7, is that a standard move? I will deal with that as soon as I get back

This is considered to be a “basic” opening trap that “everyone” knows. I’m pretty sure most players at 1000+ USCF are familiar with this. There is also 6. d4!, but that is significantly less common. Of course, “everyone” is in quotation marks. Nxd5 is still the most popular move in the lichess database.

Do you have a chess rating of any kind?

For a while, White Dove fell for it. WD happens to select Bc5 right now.

Last edited by birdracerthree (July 19, 2024 12:28:43)

ScratchChessChampion

Scratch Chess Federation August Update

Here is the list of SCF rated engines who need to be retested against the Komodo bot to adjust the ratings accurately. If anyone else would like to be included, please let me know. Thanks!

1. GoK Classic (Difficult - Competition Mode)
2. GoK Classic (Medium)
3. White Dove (P5) *new
4. White Dove (P4)
5. Destructor 2 *new
6. Chess AI v2.3 *new
7. GoK NNUE (Difficult)

Last edited by ScratchChessChampion (July 20, 2024 00:30:10)

ababoin07

ScratchChessChampion wrote:
Scratch Chess Federation August Update

Here is the list of SCF rated engines who need to be retested against the Komodo bot to adjust the ratings accurately. If anyone else would like to be included, please let me know. Thanks!

1. GoK Classic (Difficult - Competition Mode)
2. GoK Classic (Medium)
3. White Dove (P5) *new
4. White Dove (P4)
5. Element (7+8) *new

Can you test Destructor Alpha?

birdracerthree

ScratchChessChampion wrote:
Scratch Chess Federation August Update

Here is the list of SCF rated engines who need to be retested against the Komodo bot to adjust the ratings accurately. If anyone else would like to be included, please let me know. Thanks!

1. GoK Classic (Difficult - Competition Mode)
2. GoK Classic (Medium)
3. White Dove (P5) *new
4. White Dove (P4)
5. Element (7+8) *new

Element 7+8 is a little awkward; if Element’s IID search takes longer than usual, it decreases the depth by 1. In the middlegame, the 5+8 IID search will usually run into this, so Element 7+8 will usually run 6+8. Should I disable the failsafe so Element always runs 7+8?

We also have the two engines you found

Last edited by birdracerthree (July 19, 2024 13:21:50)

ArnoHu

ScratchChessChampion wrote:
Scratch Chess Federation August Update

Here is the list of SCF rated engines who need to be retested against the Komodo bot to adjust the ratings accurately. If anyone else would like to be included, please let me know. Thanks!

1. GoK Classic (Difficult - Competition Mode)
2. GoK Classic (Medium)
3. White Dove (P5) *new
4. White Dove (P4)
5. Element (7+8) *new

GoK NNUE, please!

-089-

birdracerthree wrote:
-089- wrote:
birdracerthree wrote:
-089- wrote:
birdracerthree wrote:
birdracerthree wrote:
*snip* as you like to say

Currently testing e4 e5 gameplay now.
Update : SB runs into Fried Liver (although only GoK plays it and it doesn’t do it correctly), plays dxc6 in Scotch and Four Knights Scotch and gets to around +1.
you're right, I didn't catch your joke at all lol it was 5 am and I had just woken up before I wrote that post. was really tired, I'm not a morning person haha
*snip*
I forgot to respond, so I will now
1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Ng5 d5 5. exd5 Nxd5 6. Nxf7 Kxf7 7. Qf3+ Ke7 8. Bxd5 Nd4 9. Qf7+ Kd6 10. Nc3 Kc5 11. b4+ Kxb4 12. Rb1+ Ka5 13. Bxb7 Nxc2+ 14. Kd1 Bxb7 15. Qc4 Ne3+ 16. fxe3 Bf3+ 17. gxf3 Qd7 18. Rb5+ Qxb5 19. Qxb5#
With SF 11 assistance as usual
thanks, I see. in that one test game I had it didn't come to 6. Nxf7, is that a standard move? I will deal with that as soon as I get back
This is considered to be a “basic” opening trap that “everyone” knows. I’m pretty sure most players at 1000+ USCF are familiar with this. There is also 6. d4!, but that is significantly less common. Of course, “everyone” is in quotation marks. Nxd5 is still the most popular move in the lichess database.

Do you have a chess rating of any kind?

For a while, White Dove fell for it. WD happens to select Bc5 right now.

no OTB rating if that's what you mean.
1155 rapid on chesscom, probably a little underrated by now, I stopped playing other people a little over a year ago.
it doesn't surprise me that I wouldn't know even the most basic traps, I only ever bothered to learn a single opening and even that was a long time ago. my main focus was on tactics, puzzles are the only thing I still do semi regularly (almost at 2600). I would describe myself as more of a casual player lol

ArnoHu

Please help me beta-test drag&dropping pieces on the next Chess Neural Network Version at https://scratch.mit.edu/projects/1033757275/ Feedback welcome! Thank you.

ababoin07

ArnoHu wrote:
Please help me beta-test drag&dropping pieces on the next Chess Neural Network Version at https://scratch.mit.edu/projects/1033757275/ Feedback welcome! Thank you.

I did not notice any bug but maybe it would be necessary to add a transparent square around each suit to facilitate the drag - drop (the size of a tile of the calendar) (it's difficult to drag on mobile)

Discuss Scratch