Monday, May 28, 2012

Vitruvius Follow Up


      As mentioned in a comment on the previous post, a test was conducted where the engine was given 25 test positions chosen from games in which a typical human positional sacrifice was played. The results can be viewed HERE. The ‘Human Version’ performed remarkably well.
      Its predecessor, DeepSaros3, can be downloaded HERE or HERE.  You can also download opening books. Version 3.0 is inspired by Wilhelm Steinitz and the evaluation functions have been modified to follow the rules dictated by Steinitz in the following publications:
- The Modern Chess Instructor (1889)
- Various numbers of "International Chess Magazine" (1886, 1901)
      So far I have not had a chance to try DeepSaros3 other than in a short 5 minute match against Houdini 1.5 (x64).  The results were surprising and indicate this engine is definitely worth further evaluation. DeepSaros 3 won all four games.
      Some comments made on the Open Chess Forum:


Everything I've seen posted would indicate that it's the weakest of the Ippolits, and buggy in MultiPV mode.

... I don't think that the utility of an engine stands or falls with its strength, once it's "strong enough" (and Vitruvius appears to be as strong or stronger than R4, at least in some testing). If Vitruvius manages to suggest lines which make sense to human players, analyzing or preparing for games with other humans, and which differ from the typical "computer moves" without being blunders, that would be a useful achievement, even if it loses against Houdini.
At some point I stopped using Vitruvious for analysis of correspondence games as it never suggested any idea that was better than what I could get with other engines and its crazy tries were always very easy to refute (contrast with Zappa Mexico II Dissident Aggressor which now and then still suggests plans that blow the socks of Rybka 4.1 or Critter 1.4.)
... I don't know if Vitruvius may be good or not for port-mortem analysis of human games, or OTB preparation (where needs compared to correspondence games analysis are entirely different) I advice people to remain skeptical.

No comments:

Post a Comment