BigLion and fine 70

[Dieter Bürßner]

Geschrieben von: / Posted by: Dieter Bürßner at 10 December 2002 01:39:13:
Als Antwort auf: / In reply to: Re: BigLion and fine 70 geschrieben von: / posted by: Sune Fischer at 10 December 2002 00:30:02:

One other idea (when you reach the depth faster). By *disabling* repetition detection, I get more consistent results. It might be a good idea for debugging/testing (because in this position, it should decrease the luck factor - see my argument in the other post).

Ah, I thought of doing something similar though less dramatic, namely not storing repetition draws at all. Just like we have a special "range" for mate scores, it is possible to make a range for repetition draws.

This will unfortunately not help much. There will be many non PV lines, that were arrived at, because somewhere deeper in the search another line was refuted due to repetition. Those positions will be stored in the HT. When you retrieve them, you may have reached the position with another path, and the earlier refutation will just not be valid anymore. The problem seems to be generally unsolvable in an efficient way.
Regards,
Dieter

[Dieter Bürßner]

Geschrieben von: / Posted by: Dieter Bürßner at 10 December 2002 20:24:14:
Als Antwort auf: / In reply to: Re: BigLion and fine 70 geschrieben von: / posted by: Miguel A. Ballicora at 09 December 2002 21:10:34:

For this particular discussion, I think that we have to define "faster".
I have seen that sometimes you can get the solution at smaller depths but it takes much more nodes (and time, of course).

I agree. For some results I cited, I could not give a good time measure. First of all, it is solved that fast, that the overhead of the output routine is significant. Than, for the really not very scientific test with random move ordering, I needed a call to some pseudo random number generator, which again gives a significant overhead.
But I think, in that specific case, it won't matter that much. Random move ordering used more depth, more nodes, ... (all this with material only eval, and when the target is to find a pawn win. Otherwise Kb1 could be found for other reasons earlier, which makes interpretation even more difficult).
GCP cited disagreeing results (worse move ordering yielding in lower depth).
I think, there is a too big luck factor (mainly because of HTs) in this position, to make it easy to understand all the influences properly.
BTW. Out of curiosity, I let it run without HTs. As can be expected from Uri's line, a version with material eval found a pawn win at depth 26 (after 6 hours ...).
Regards,
Dieter