Upgrade to Chess.com Premium!

Houdini 1.5a ignores mates in one


  • 11 months ago · Quote · #21

    Irontiger

    LoveYouSoMuch wrote:
    jaaas wrote:

    Could it be, that the engine has somehow hardcoded into it, that a K+B vs K+N situation is already drawn (this would explain why all moves from the given positions are evaluated as 0.00, and no attempts to consult a tablebase are made either)?

    i am somewhat sure (not too sure) that this is the case.

    Probably this. Houdini sees that White has only a minor piece "and hence cannot win", and Black has only one minor piece "and hence cannot win", deduces the game is drawn, and plays the first move he finds.

  • 11 months ago · Quote · #22

    jaaas

    This has already been largely cleared up upon - in post #16, TBentley pointed out a thread on another forum where this oddity had also been discovered and brought to the attention of Robert Houdart, who confirmed that Houdini 1.5a indeed had these assumptions hardcoded. Here's the link again:

    http://talkchess.com/forum/viewtopic.php?t=37637&postdays=0&postorder=asc&topic_view=flat&start=0

  • 11 months ago · Quote · #24

    Fugazy_Crapov

    jaaas wrote:
    FEDTEL wrote:

    what chess GUI are you using?

    The GUI shouldn't really matter, but it's all the same for me in both Arena 3.0 (with Gaviota tablebases enabled) and Tarrasch (no tablebases functionality here yet).

    Just an FYI, you can configure Tarrasch to use Gaviota tablebases with Houdini 1.5 by creating a custom parameter.  The extra parameter is "Gaviotatbpath" (without the quotes) and the value is the path, e.g., c:\gaviota. 

    Oddly, Houdini can't solve this position with or without tablebases!

  • 11 months ago · Quote · #25

    jaaas

    Thanks, I wasn't aware of that. I used that parameter and a Houdini kibitzer immediately saw a mate in 127 in the following position.

  • 11 months ago · Quote · #26

    JJZ03

    What The...

  • 11 months ago · Quote · #28

    Fugazy_Crapov

    jaaas wrote:

    Thanks, I wasn't aware of that. I used that parameter and a Houdini kibitzer immediately saw a mate in 127 in the following position.

     

    That's insane!  Looks like it runs into the 50 move rule, though. 

  • 11 months ago · Quote · #29

    LoveYouSoMuch

    no, pawn moves.

  • 11 months ago · Quote · #30

    jaaas

    As a matter of fact, this particular case just barely (by one[!] move, but still) does run into the 50-move rule. There are neither pawn moves nor any captures made in moves 6 through 55, which means that right after move 55 the 50-move rule would normally have kicked in.

    Here the whole sequence (notice moves 5, 55, and 56):

    Forced wins provided by Nalimov or Gaviota tablebases are not constrained by limits such as the 50-move rule (although one could create tablebases by means of a modified algorithm that would conform to that, or similar, rule).

     

    Thanks to 6-man tablebases, the following position has been made known to be a win for White in no less (with optimal defense on the part of Black) and no more (with optimal play on White's part) than 262 moves:

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

    I don't have 6-man tablebases available locally and haven't looked through the winning sequence, thus I can't tell at the moment whether the 50-move rule might prevent an actual win from happening in this particular case.

    However, firstly, the 50-move rule does not adjudicate (at least it should not - the implementation on various chess servers may vary though) a game by itself - a draw, similarly to occurences of threefold repetition, needs to be claimed by one of the participants if it is observed that the conditions for an application of the rule have been met.

    Secondly, such ridiculously long winning sequences are more of a theoretical significance, and most likely would never have been found/proven without the assistance of tablebases in the first place, thus constraints geared towards practical play, such as a move limit rule, may be as well be abstracted from when contemplating such winning lines.

  • 11 months ago · Quote · #31

    FirebrandX

    jaaas wrote:
    I don't have 6-man tablebases available locally and haven't looked through the winning sequence, thus I can't tell at the moment whether the 50-move rule might prevent an actual win from happening in this particular case.

    http://www.shredderchess.com/online-chess/online-databases/endgame-database.html

  • 11 months ago · Quote · #32

    jaaas

    I know that resource (the backend is probably the same as for this page), but it can be a bit more tedious for such extremely long lines having to click through, and you don't get to see the complete list of moves either. In Arena, you can just click the "Demo" button, and the engine "plays" against itself, in such cases essentially just following the winning path provided by the tablebase; more importantly, you end up having the complete move list which may be scrolled through or saved for other purposes.

    Anyway, for the above position I clicked throught the first 51 moves and no piece was captured, thus this winning line abstracts from the 50-move rule as well.

  • 10 months ago · Quote · #33

    tanjerr

    WHAT THE


Back to Top

Post your reply: