WDL numbers not updating? about ceres

[whelanh]

FYI: I was running Lc0 net #69626

[rooklift]

Position is r4rk1/pp1bp1b1/2np1pp1/6Pq/2P5/1PN2PP1/P1NQPKBn/R6R w - - 0 17

[rooklift]

Checking for myself, near the start of the search I received the line:

info depth 10 seldepth 2 time 600 nodes 1387 score cp -103 wdl 84 288 628 tbhits 0 nps 2311 multipv 16 pv c3d5 string M= 129

And later I received

info depth 16 seldepth 12 time 26126 nodes 139291 score cp 42 wdl 84 288 628 tbhits 0 nps 5332 multipv 2 pv c3d5 c6e5 d5e7 g8f7 e7d5 f6g5 f2g1 f8h8 a1f1 e5c6 f1f2 string M= 140

So it seems WDL didn't get updated. (Actually not sure if it is ever updating for any position...)

[dje-dev]

Ceres shows the W/D/L of the static position, not the average W/D/L over the subtree. But your observation (thanks!) makes it clear that subtree average W/D/L would be much more interesting useful. The fix is pushed and will be bundled into the next release.

[whelanh]

Thank you very much. Ceres is a very interesting engine!

…

On Tue, Jul 20, 2021, 21:42 dje-dev ***@***.***> wrote: Ceres shows the W/D/L of the static position, not the average W/D/L over the subtree. But your observation (thanks!) makes it clear that subtree average W/D/L would be much more interesting useful. The fix is pushed and will be bundled into the next release. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#51 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC2JOXPQ4UTATV3FXJXHUD3TYYQYLANCNFSM5ATRIO4A> .