After 200 games the two versions finished with about equal score after a winning streak from v0.31 in the end.
I have a 'gut feeling' the v0.32 is be better, I can not see how it would not be. However at this point I am not looking for differences that take thousands of games to determine.
Mediocre is not nearly good enough to start chasing 1-2 elo points here and there, I want atleast 50-70 points (about 60% won games) before I am satisfied.
I think I will leave futility pruning and the evaluation as it is for a while and start looking at how extensions are handled. Currently Mediocre has only a very crude check extension, so I will take a look at extensions for mate threat (if null move returns a mate value the branch is searched more carefully), recapture, single-response and something regarding passed pawns.
I will need a new way of handling the extensions and probably make use of fractional plies for some of them. And while doing this I will take care of the repeated work Mediocre currently is doing for determining if a position is in check (it is done in the check extension, late move reduction and futility pruning, this should only be done once).
Off I go. :)
3 comments:
Yep there's also more work to do on a chess engine! :-) Do not underestimate evaluation though, you're more likely to get 70 elo from improving the evaluation than from adding extensions!
Evaluation is important, and there are certainly many elo-points to be gained there.
I have however noticed that Mediocre is superior to most engines searching approximately to the same depths as it. This should indicate that the current evaluation is fairly good.
I have also observed that most engines that search 1-3 ply deeper in the middle game, actually is out-searched by mediocre in the end game. I wonder why?
I have also noticed that most games lost by mediocre is lost in the midle game in positions where it is out-searched and fail to see some long combinations. I think getting a few more plys in the middle game would increase mediocre's strength _a_lot_. This is of course easier said than done :)
I have also observed that most engines that search 1-3 ply deeper in the middle game, actually is out-searched by mediocre in the end game. I wonder why?
Maybe they extend too much! :-)
Post a Comment