Set UCI_ShowWDL by default to false

[vondele]

UCI_ShowWDL might not be shown by GUIs that don't know the option,
but crash on the WDL output, effectively making it hard for users to
turn it off and run the engine. This sets it by default to false.

fixes #2787

No functional change.

[snicolet]

Then maybe it would be a good idea to rename the option from "UCI_ShowWLD" to "Show WLD stats" (and default to false), so that the users actually see the option in the GUI panels, try it and report on the GUI authors that "your GUI crashes when I click on the option for displaying the WLD stats". This could put more implicit pressure on the GUI authors to respect the UCI protocol and not crash on unknown tokens? Otherwise (if the option is hidden) the GUI authors will sit on the problem and the feature will get less exposure and use in the community.

[vondele]

@snicolet UCI_ShowWLD seems to be what the talkchess thread converged on, and would make this more standard among engines, it seems better to stick to it.

[vdbergh]

BTW there is currently some discussion on TalkChess about this. From the comments, e.g. the example of two bare kings, it seems that it would be more natural to let the wdl probabilities depend on the game phase rather than the ply count. Even though the wdl numbers are meant to be an aggregate, people are of course strongly tempted to apply them to particular positions.

[vondele]

@vdbergh the wdl model is of course just an average representation. The game phase data is available (see previous PR), but isn't quite as smooth.

[LouisZulli]

Perhaps the most salient point I saw on TalkChess is that WDL scores should be exact when Stockfish has access to relevant endgame tables. For example, when the root position is a known draw, Stockfish should give 0 1000 0 for WDL.

[vondele]

just had a look at the thread. As the readme states, WDL numbers are statistics for positions from games on fishtest with the same score and game ply. Any other interpretation will likely be wrong, but sometimes interesting and convenient.

[LouisZulli]

I understand that this is an approximate model based on a specific set of statistical data from fishtest. As such, it strikes me as "fluff", which will simply absorb developers' time in attempts to improve this feature.

[vondele]

Yes, it seems good to have a sufficiently narrow and well-defined scope, right now that is reporting fishtest statistics. As such, feature creep, like doing things differently when being in table base, would not be a good idea.

[vondele]

@Alayan-stk-2 can you, for future submission that ask for options and setting of stockfish, specify that UCI_ShowWDL should be true? Thanks!

[Alayan-stk-2]

@vondele No problem