Seer

[mar]

to find mate in 3 in KQk has nothing to do with 50 move rule (and not eval either)
if your engine fails to see it, then you should fix your search

[JohnWoe]

Nothing to be fixed. My search is perfect.

I was talking in general

[connor_mcmonigle]

+256 is a mate score and Seer see's the mate at (69. b8=Q {+256.00/18 1s}). With previous versions, I neglected to implement mate distance pruning and mate distance eval adjustments as I never observed any issues with delivering mate . Clearly, that's really stupid. Additionally, I used floating point values for alpha and beta in previous versions which made this difficult.

I'll release a new version later today which fixes this behavior.

[Sylwy]

+256 is a mate score and Seer see's the mate at (69. b8=Q {+256.00/18 1s}). With previous versions, I neglected to implement mate distance pruning and mate distance eval adjustments as I never observed any issues with delivering mate . Clearly, that's really stupid. Additionally, I used floating point values for alpha and beta in previous versions which made this difficult.

I'll release a new version later today which fixes this behavior.

Seer 1.2 plays now on Fritz 17 GUI (GUI linked to Syzygy bases). Absolutely any problem in adjudication (a GUI one, of course) !

[connor_mcmonigle]

Thanks for the new version with +200 !
But I think there is still room for improvements

best wishes

Werner
[Event "40 Züge in 8 min"]
[Site "GreKo, DESKTOP-K5TPRSK"]
[Date "2020.11.18"]
[Round "1.1"]
[White "Seer 1.2NN x64 1CPU"]
[Black "FabChess 1.16 x64 1CPU"]
[ECO "C41"]
[Result "1/2-1/2"]

...

no mate with K+Q - K
50 move rule!!

This behavior should be fixed in v1.2.1. Seer should now play the shortest path to mate. This release also resolves a minor race condition that could have mangled the best move output, though likely never would have actually happened in practice.

https://github.com/connormcmonigle/seer ... tag/v1.2.1

[Daniel Anulliero]

Nothing to be fixed. My search is perfect.

I was talking in general

I guess even the sf or komodo or ethereal search is not completely perfect, so, congratulations, you've found the gral

[gerhard]

... sorry but non of the compiles v 1.21 is working on my cpu AMD Ryzen

[Sylwy]

This behavior should be fixed in v1.2.1. Seer should now play the shortest path to mate. This release also resolves a minor race condition that could have mangled the best move output, though likely never would have actually happened in practice.

https://github.com/connormcmonigle/seer ... tag/v1.2.1

Thank you ! Already in test. Will be a double test. Seer 1.2 & Seer 1.2.1 vs. Deep Sjeng c't 2010. I'm curious to see the differences in game and result.

[Guenther]

... sorry but non of the compiles v 1.21 is working on my cpu AMD Ryzen

Hard to imagine, with all the compilations available.
What do you mean with not working?
You know you must have the right weights file for playing with it? (current save.bin and give the path to it by uci options)

If you start the generic one from commandline what happens? Just click the exe and type uci.
(znver1 should be also working for you, it is for older Ryzen)

Code: Select all

uci
id name Seer 1.2.1
id author Connor McMonigle
option name Clear Hash type button
option name Threads type spin default 1 min 1 max 512
option name Hash type spin default 128 min 1 max 65536
option name Weights type string default ../train/model/save.bin
uciok

[connor_mcmonigle]

... sorry but non of the compiles v 1.21 is working on my cpu AMD Ryzen

You'll need to download the Weights (save.bin) file from the release page. Then, before the go command is sent to the engine, you must set "Weights" UCI parameter to a path to the Weights file. Hopefully this resolves your issue.

For an older (1000 or 2000 series) AMD Ryzen system, seer_znver1.exe compile should work. For a 3000 series part, the seer_znver2.exe compile is recommended.