My personal feeling is the Houdini author must have studied Stockfish's source code very carefully.
I can't find anything in the code that says otherwise. It certainly looks like the Houdini source code. It would've taken even the best talent months (or years?) of hard work to develop it, so it doesn't look like a prank. It was not a clone of Stockfish in the sense that the codebase was quite different, certainly not a 10 minutes copy-and-paste. I took a quick look into it, it looked like a valid strong chess engine. The alleged source code might have been published online. The main reason it looks different is because of the Dutch translation and the fact that Stockfish has evolved significantly since Houdart last touched the Stockfish codebase. In other words, Houdini did not notably build upon Stockfish in any meaningful way.
(Source: nextchessmove's elo data + CCRL data) Later, when Houdini 6 was released, this was also true. However, the code came from TurboPack and was translated into C++, so it's not clear that that was original work, either.Īdjusting for the time of release, the version of Stockfish on GitHub at the time of Houdini 5's release was more or less equal in strength to Houdini 5. For example, the licensing system was not from Stockfish. In fact, in one file, evaluatie.cpp, you can spot the naming conventions used by the Hex-Rays decompiler.
There were changes, but Mark Lefler, developer of Komodo, noted that a number ofaa changes were simply adding features reverse engineered from Komodo. The leaked source code was compiled by Ed Schroder and found to have a 1:1 correspondence with Houdini in a vast number of test positions, while others, like Simon Guenther, have found that the leaked code behaves identically in the start position even up to high depths. The alleged performance improvements over Stockfish go away after removing this trick. Exploiting modular arithmetic tricks and archived Houdini games, this exact code was shown to be present in the release version of Houdini. To hide its origins, code was included that multiplied the node count by 8/7. A number of Stockfish developers, including former maintainers, have noted as such. The Stockfish folks, along with Mark Crowther of TWIC, have made the chess world an immeasurably better place over the years for their work, and at no cost to us.The search, evaluation, etc., are almost identical to Stockfish, with the notable exception of being rewritten in Dutch. So, kudos to Stockfish and Joost VandeVondele in particular for doing this. A position the old version gave as +4.2 was around +2.6 on the new one, and another position that was around +2.5 went down to +1.5-1.6, which also squared with my sense of the position. Regarding “common sense”, I just compared the new version with a very recent one, and the difference was dramatic. First, the evals are more in keeping with common sense (which is what we as humans need) second, they will be consistent.
Fishtest stockfish update#
To keep this value steady, it will be needed to update the win_rate_model()įrom time to time, based on fishtest data. With this patch, a 100cp advantage will have a fixed interpretation, The eval needed to have 50% win probability at fishtest LTC (in cp and internal Value): The current evaluation changing quite a bit from release to release, for example, No longer related to the classical parameter PawnValueEg (=208). The reason to introduce this normalization is that our evaluation is, since NNUE, "100 centipawns" for a position if the engine has a 50% probability to winįrom this position in selfplay at fishtest LTC time control. The win_rate_model() such that Stockfish outputs an advantage of To the UCI centipawn result used in output. Normalizes the internal value as reported by evaluate or search I’ve repeatedly found that advantages that older engines might have thought were around +1.5 have more than doubled, as if an extra pawn and a bit of extra comfort translates to an extra piece or more.Įvery so often I check on the Stockfish development page to try out the latest version of their engine today, I did so and found this:
Fishtest stockfish plus#
After it joined the neural net revolution, the evaluations changed, massively: now advantages that seemed clear or on the cusp between a clear and a decisive plus shot up massively. If White had an extra pawn and everything else was pretty normal, one would expect to see an evaluation around +1, give or take one or two tenths of a point. Before Stockfish took a page out of the Alpha Zero playbook and went with the neural net approach, its evaluations were in keeping with human understanding.
0 kommentar(er)
