See https://github.com/official-stockfish/S ... -653510021 and its successor.syzygy wrote: ↑Sun Jul 05, 2020 12:20 pm ...
WIth a TB draw this is more complicated since a draw score does not guarantee a draw. But if the root position is a TB draw, then Stockfish knows that and could again make use of that knowledge quite easily. (Perhaps this is being planned and has simply not yet been done in the current master.)
WDL
Moderators: hgm, Rebel, chrisw
-
- Posts: 6442
- Joined: Tue Jan 09, 2007 12:31 am
- Location: PA USA
- Full name: Louis Zulli
Re: WDL
-
- Posts: 5563
- Joined: Tue Feb 28, 2012 11:56 pm
Re: WDL
I do agree that feature creep is bad, but that is why the feature should not have been added to begin with. Now it just makes SF look dumb in certain situations where that could be avoided.zullil wrote: ↑Sun Jul 05, 2020 12:35 pmSee https://github.com/official-stockfish/S ... -653510021 and its successor.syzygy wrote: ↑Sun Jul 05, 2020 12:20 pm ...
WIth a TB draw this is more complicated since a draw score does not guarantee a draw. But if the root position is a TB draw, then Stockfish knows that and could again make use of that knowledge quite easily. (Perhaps this is being planned and has simply not yet been done in the current master.)
-
- Posts: 6442
- Joined: Tue Jan 09, 2007 12:31 am
- Location: PA USA
- Full name: Louis Zulli
Re: WDL
Agreed.syzygy wrote: ↑Sun Jul 05, 2020 3:01 pmI do agree that feature creep is bad, but that is why the feature should not have been added to begin with. Now it just makes SF look dumb in certain situations where that could be avoided.zullil wrote: ↑Sun Jul 05, 2020 12:35 pmSee https://github.com/official-stockfish/S ... -653510021 and its successor.syzygy wrote: ↑Sun Jul 05, 2020 12:20 pm ...
WIth a TB draw this is more complicated since a draw score does not guarantee a draw. But if the root position is a TB draw, then Stockfish knows that and could again make use of that knowledge quite easily. (Perhaps this is being planned and has simply not yet been done in the current master.)
-
- Posts: 5563
- Joined: Tue Feb 28, 2012 11:56 pm
Re: WDL
So I did not add it to Cfishzullil wrote: ↑Sun Jul 05, 2020 3:46 pmAgreed.syzygy wrote: ↑Sun Jul 05, 2020 3:01 pmI do agree that feature creep is bad, but that is why the feature should not have been added to begin with. Now it just makes SF look dumb in certain situations where that could be avoided.zullil wrote: ↑Sun Jul 05, 2020 12:35 pmSee https://github.com/official-stockfish/S ... -653510021 and its successor.syzygy wrote: ↑Sun Jul 05, 2020 12:20 pm ...
WIth a TB draw this is more complicated since a draw score does not guarantee a draw. But if the root position is a TB draw, then Stockfish knows that and could again make use of that knowledge quite easily. (Perhaps this is being planned and has simply not yet been done in the current master.)
-
- Posts: 4889
- Joined: Thu Mar 09, 2006 6:34 am
- Location: Pen Argyl, Pennsylvania
Re: WDL
I cannot disagree with your comments - I like WDL, but that is probably just me - but not that I believe it would ever be 100 percent accurate even most of the time except for known cases - mates and known draws.syzygy wrote: ↑Sun Jul 05, 2020 12:20 pmAnd why would anyone look at WDL in the first place.
But if a prediction of the probabilities of win, draw and loss as the outcome of the current game is shown, then it is a bit peculiar not to show accurate numbers when those numbers are known.
SF uses a conversion function to map (score, game_ply) to a winning percentage for white (and then applies the same function to (-score, game_ply) to find the winning percentage for black = losing percentage for white). It should not be difficult to let this function map a TB win or better to 100 percent. (Perhaps it does so, I did not check.)
WIth a TB draw this is more complicated since a draw score does not guarantee a draw. But if the root position is a TB draw, then Stockfish knows that and could again make use of that knowledge quite easily. (Perhaps this is being planned and has simply not yet been done in the current master.)
-
- Posts: 4556
- Joined: Tue Jul 03, 2007 4:30 am
Re: WDL
We've had these discussions before. I only sound angry in your head. Are you incapable of reading my messages with your internal voice in a calmed and cheerful voice?
Can I help in any way? Would adding smilies at the end of my sentences help or something? Like:
Because, I never do anything that makes me angry, if these discussions angered me I'd have stopped having them long time ago. Just having fun.Ovyron wrote: ↑Sun Jul 05, 2020 6:40 amLeela models WDL on the fly . Stockfish applies some arbitrary formula that takes score and move number to bring out some user misleading abomination .
They're not doing anything similar whatsoever, Stockfish's WDL just appears the same at the end .
Stop pretending they're the same because the only thing they have in common is the name .
-
- Posts: 6442
- Joined: Tue Jan 09, 2007 12:31 am
- Location: PA USA
- Full name: Louis Zulli
Re: WDL
This thread is about engines providing WDL scores when evaluating chess positions. You seem to be asking about Syzygy endgame tables.Nordlandia wrote: ↑Sat Jul 11, 2020 11:25 am I think I have downloaded all the WDL files. But I have to ask to be certain.
1495 files total at 8.24 TB . Is that correct ?
1495 seems wrong. Shouldn't it be 145 + 365 + 1001 = 1511?
Maybe start a new thread so folks can confirm the number.