Banksia GUI released

[zenpawn]

First, thank you for adding Chess960 support for Winboard engines and for fixing the issue with using a book. I can confirm they are both working.

Question: After a lichess game finishes, the log tab is cleared and the game on the screen is an old one played offline. How does one find the PGN and log for the previously played lichess games?

In the General tab of Settings dialog, there is an item "Online game log". It is designed for logging PGN games for online games. It works for Local Area Network games but... not for Lichess ones

I will fix it for the next release. Thanks for the report.

That will be great. And likewise for the logs showing the communications? It's via those that I'm able to check, for example, whether the "level" command was sent (which it currently is not, causing Winboard engines to lose on time in lichess). Thanks.

[Vinvin]

BTW, BSG always uses the red line for white and blue for black players. Their names should be displayed on clock boxes which quite closed to the graph.

I see nothing to remind this. Please write it somewhere or let chose the graph color in the settings.

BSG has implemented some standards about scores. The way to view scores as your suggestions is called "Score in white view". Users can set it up in Settings dialog as the following image:

Thanks ! But I see the score is not inverted in the "Moves" window.

One more remark about the logarithmic scale of the graph, it's a bit too strong (+1 is halfway to +10).
Lichess has implemented a formula to compute the vertical value on the graph, it uses the winning probability. See here : https://www.chessprogramming.org/Pawn_A ... lationship

Also, I saw a strange behavior of the "Moves" window, when I restart BSG, the police is too big when I click "+" then "-" on the police size (return same size) the size is OK. Note I use a 4K display with a scale of 200% in the Windows Display Preferences.

Here is a screenshot of the tuned BSG I use now
https://www.dropbox.com/s/y2vtry09hu7gd ... 1.png?dl=0

Note that I had some crashes when I toy with windows inside the interface. I think I had some other strange behaviors but I have to investigate more.

Thanks for your work !

Vincent

[Jonathan003]

[/quote]


For a quick test, I set the depth to 8 only, then convert from gm2600.bin (347 KB) to g2600.pgn (114 KB):




[/quote]


If I convert the book this way the conversion is done in a split of a second. The pgn file is only 112 kb and contains only 844 games. So it is impossible all lines are there.

[Jonathan003]

[/quote]

Almost all lines can be found on the g2600.pgn (except the last two lines - they are shorter than 8 depth.



Can you help to check manually like the above steps? Thanks again.
[/quote]

Why do you think almost all lines are there? I find many more examples of lines not being there.
Here are some more examples:

[zenpawn]

It appears the "level" command is still not being sent to Winboard engines to tell them the time control.

I am struggling to reproduce the problem. Can you send me an engine log? Thanks.
Note that "level" will be sent only when the timer is the standard type.

What would be a non-standard type? I challenged with 180s + 2s, i.e., 3+2 from Banksia's lichess bots window. The log looked like:

Code: Select all

RookieMonster < xboard
RookieMonster < protover 2
RookieMonster > done=0
RookieMonster < force
RookieMonster < d2d4
RookieMonster < g8f6
hm::turkjs 1822 > move d2d4
RookieMonster < force
RookieMonster < b1c3
RookieMonster < d7d5
hm::turkjs 1822 > move b1c3
RookieMonster < force
RookieMonster < f2f3
RookieMonster < e7e6
hm::turkjs 1822 > move f2f3
RookieMonster < force
RookieMonster < e2e4
RookieMonster < c7c5
hm::turkjs 1822 > move e2e4
RookieMonster < force
RookieMonster < f1b5
RookieMonster < time 18507
RookieMonster < otim 18763
RookieMonster < go
hm::turkjs 1822 > move f1b5
RookieMonster > RookieMonster 1.7.5
RookieMonster > rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
RookieMonster > turn = white
RookieMonster > computer playing as white
RookieMonster > >
RookieMonster > feature myname="RookieMonster 1.7.5"
RookieMonster < accepted myname
RookieMonster > feature debug=1
RookieMonster < accepted debug
RookieMonster > feature reuse=0
RookieMonster < accepted reuse
RookieMonster > feature setboard=1
RookieMonster < accepted setboard
RookieMonster > feature variants="normal,fischerandom,3check,kingofthehill"
RookieMonster < accepted variants
RookieMonster > done=1

followed by the thinking output from RM until it lost on time. Note: It was considering the right position after the book (ignore the FEN as that was from the startup), so that's working now.

Hope that helps.

[phhnguyen]

I see nothing to remind this. Please write it somewhere or let chose the graph color in the settings.

Thanks for the suggestion. It will be implemented for the next release.

BSG has implemented some standards about scores. The way to view scores as your suggestions is called "Score in white view". Users can set it up in Settings dialog as the following image:

Thanks ! But I see the score is not inverted in the "Moves" window.

It was kept purposely (to show original scores). BSG will show inverted scores in the coming release

One more remark about the logarithmic scale of the graph, it's a bit too strong (+1 is halfway to +10).
Lichess has implemented a formula to compute the vertical value on the graph, it uses the winning probability. See here : https://www.chessprogramming.org/Pawn_A ... lationship

It is one of my headache issues since graphs of logarithmic scale look so bad and not smooth!

However, I don't understand the method from that page, which formula/function to calculate y from a score. Can you or someone explain more?

BTW, I have tried the Sigmoid function but the graph looks worse than the logarithmic scale.

Also, I saw a strange behavior of the "Moves" window, when I restart BSG, the police is too big when I click "+" then "-" on the police size (return same size) the size is OK. Note I use a 4K display with a scale of 200% in the Windows Display Preferences.

Here is a screenshot of the tuned BSG I use now
https://www.dropbox.com/s/y2vtry09hu7gd ... 1.png?dl=0

Can you post the screen of BSG when it is "too big"? Thanks

Note that I had some crashes when I toy with windows inside the interface. I think I had some other strange behaviors but I have to investigate more.

Thanks for your work !

Vincent

If you know the steps to reproduce that crash, it will help us much. Thanks again.

[phhnguyen]

It appears the "level" command is still not being sent to Winboard engines to tell them the time control.

I am struggling to reproduce the problem. Can you send me an engine log? Thanks.
Note that "level" will be sent only when the timer is the standard type.

What would be a non-standard type? I challenged with 180s + 2s, i.e., 3+2 from Banksia's lichess bots window. The log looked like:

Code: Select all

RookieMonster < xboard
RookieMonster < protover 2
RookieMonster > done=0
RookieMonster < force
RookieMonster < d2d4
RookieMonster < g8f6
hm::turkjs 1822 > move d2d4
RookieMonster < force
RookieMonster < b1c3
RookieMonster < d7d5
hm::turkjs 1822 > move b1c3
RookieMonster < force
RookieMonster < f2f3
RookieMonster < e7e6
hm::turkjs 1822 > move f2f3
RookieMonster < force
RookieMonster < e2e4
RookieMonster < c7c5
hm::turkjs 1822 > move e2e4
RookieMonster < force
RookieMonster < f1b5
RookieMonster < time 18507
RookieMonster < otim 18763
RookieMonster < go
hm::turkjs 1822 > move f1b5
RookieMonster > RookieMonster 1.7.5
RookieMonster > rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
RookieMonster > turn = white
RookieMonster > computer playing as white
RookieMonster > >
RookieMonster > feature myname="RookieMonster 1.7.5"
RookieMonster < accepted myname
RookieMonster > feature debug=1
RookieMonster < accepted debug
RookieMonster > feature reuse=0
RookieMonster < accepted reuse
RookieMonster > feature setboard=1
RookieMonster < accepted setboard
RookieMonster > feature variants="normal,fischerandom,3check,kingofthehill"
RookieMonster < accepted variants
RookieMonster > done=1

followed by the thinking output from RM until it lost on time. Note: It was considering the right position after the book (ignore the FEN as that was from the startup), so that's working now.

Hope that helps.

Oh, I see. Thanks a lot for the report!

Frankly speaking, I did not test BSG's Lichess bot with WB engines (sorry). The info Lichess sends to bots is quite similar to UCI info thus it is easy to implement for UCI engines but not for Winboard ones, I probably missed something. Working on that.

[Vinvin]

One more remark about the logarithmic scale of the graph, it's a bit too strong (+1 is halfway to +10).
Lichess has implemented a formula to compute the vertical value on the graph, it uses the winning probability. See here : https://www.chessprogramming.org/Pawn_A ... lationship

It is one of my headache issues since graphs of logarithmic scale look so bad and not smooth!

However, I don't understand the method from that page, which formula/function to calculate y from a score. Can you or someone explain more?

Sure !
Enter the Eval in the given formula : 1/(1+10^(-Eval/4)) , it will return a value between 0 and 1.
"0" means black is completely winning (bottom of the graph), "1" means white is completely winning (top of the graph), "0.5" means the game is even (middle of the graph).
Some examples :
+1 : 1/(1+10^(-1/4)) = 0.64 (64%)
+5 : 1/(1+10^(-5/4)) = 0.947
-3.5 : 1/(1+10^(3.5/4)) = 0.11766

[phhnguyen]

One more remark about the logarithmic scale of the graph, it's a bit too strong (+1 is halfway to +10).
Lichess has implemented a formula to compute the vertical value on the graph, it uses the winning probability. See here : https://www.chessprogramming.org/Pawn_A ... lationship

It is one of my headache issues since graphs of logarithmic scale look so bad and not smooth!

However, I don't understand the method from that page, which formula/function to calculate y from a score. Can you or someone explain more?

Sure !
Enter the Eval in the given formula : 1/(1+10^(-Eval/4)) , it will return a value between 0 and 1.
"0" means black is completely winning (bottom of the graph), "1" means white is completely winning (top of the graph), "0.5" means the game is even (middle of the graph).
Some examples :
+1 : 1/(1+10^(-1/4)) = 0.64 (64%)
+5 : 1/(1+10^(-5/4)) = 0.947
-3.5 : 1/(1+10^(3.5/4)) = 0.11766

Thanks a lot! It works

[phhnguyen]

For a quick test, I set the depth to 8 only, then convert from gm2600.bin (347 KB) to g2600.pgn (114 KB):

If I convert the book this way the conversion is done in a split of a second. The pgn file is only 112 kb and contains only 844 games. So it is impossible all lines are there.

For ply = 1, there are 9 opening lines (you see the tree at level 1 have 10 branches as the following image, BSG has ignored one since it has no child)
For ply = 2, there are 49 opening lines
For ply = 3, there are 114 opening lines
For ply = 4, there are 227 opening lines
For ply = 5, there are 346 opening lines
For ply = 6, there are 506 opening lines
For ply = 7, there are 661 opening lines
For ply = 8, there are 844 opening lines

All numbers are reasonable, some are checked manually.