Releases: tryingsomestuff/Minic
Release 3.41
Executables here : https://github.com/tryingsomestuff/Minic-Dist/tree/a8b829c7a9e361365edd1af99965b85c895932ed/Minic3/3.41
same as 3.40 with no search param exposed to UCI.
For Graham ;-)
Release 3.40
Release 3.40
Executable available here : https://github.com/tryingsomestuff/Minic-Dist/tree/a668d299b8f63c3ba4848d6ec74cc17325da1912/Minic3/3.40
Release 3.39
This release includes a new net named "nyctophobic narwhal", some good search patches and some minor bug fixes.
Overall it plays near 30Elo better than last 3.32 release in self test at short TC.
Release 3.38
For TCEC next event (PawnHash max value fix)
Release 3.37
For TCEC next event
Release 3.33
This version is not for testing, it was release for TCEC Cup11 (S24)
Release 3.32
This release fixes the major FRC issue seen in 3.31 and also comes with some more Elo.
As usual binaries are here : https://github.com/tryingsomestuff/Minic-Dist/tree/883d4aca3253bae95db73678beb778e77c9f9545/Minic3/3.32
Release 3.31
Not much gain in terms of Elo but a lot of small bug fixes.
Binaries are here : https://github.com/tryingsomestuff/Minic-Dist/tree/856f5532b1ca90734994e984331c87cf6d614b4b/Minic3/3.31
Release 3.30
This is an official minor release.
Some search patches, fixes, and some speed up but I don't expect more than +10Elo.
Executables here : https://github.com/tryingsomestuff/Minic-Dist/tree/master/Minic3
Rank Name Elo +/- Games Score Draw
1 rofChade3 74 10 2433 60.4% 48.2%
2 seer 48 10 2432 56.9% 49.5%
3 stockfish.10 22 10 2434 53.2% 44.4%
4 minic_3.30 4 9 2433 50.6% 55.0%
5 minic_3.27 -15 10 2433 47.8% 50.9%
6 stockfish.9 -38 10 2432 44.6% 46.1%
7 komodo-13.02 -96 11 2433 36.5% 42.7%
Release 3.27
This version shall be quite stronger than 3.22 or 3.24, maybe +60 to +70Elo at STC.
The net is using the same architecture as previous one, with a common input layer but various inner layers depending on number of pieces on the board (here with only 2 buckets). Only last bucket was trained with lambda!=1, the first one is always only trained with lambda=1. Data used here were a mix of former data (used for "nylon nonchalance") with additionnal 2B FRC fixed nodes using previous net.
This version also comes with some good search patches and minor fixes. I'm also preparing for some experiment : persistent hash table, game history knowledge (boom / moob), ...
Executable here as usual : https://github.com/tryingsomestuff/Minic-Dist/tree/master/Minic3