Colossus 2008a available

[mclane]

bug in time management ??

[Graham Banks]

bug in time management ??

According to some. Have you struck that problem Thorsten? Perhaps I've just been lucky.

[mclane]

i have not seen it yet.

[fern]

What I have noted is that Colossus goes not so deep in the search as other versions. It get around 13 plys in middle game, which is low for any average engine these days.
Maybe Martin is testing a code more complex OR he is tunning the thing to do a wider search to avoid pitfalls.
Fern

[swami]

I haven't tested it yet, but I seem to recall posts from testers like Patrick, CEGT people etc reporting the bug to the author.

[MartinBryant]

Unfortunately it is true.

Patrick reported it here...
http://www.talkchess.com/forum/viewtopi ... 62&t=23606

My explanation in that thread...

"It's not the speed of the game as such but the lack of any Fischer time increment that's causing the problem. Unfortunately I always test with time increments so didn't notice the problem when there isn't any.

The driver behind using the time more aggressively was that I was sick of seeing Colossus lose still having maybe 30%+ of its clock time remaining.
Hence I biased it to use much more time early (and hence play better early). It then leaves less time for the endgame, but at least you've survived that far! With even a small time increment it isn't so much of a problem, but with no time increment... Ooops!

One more for the fix list "

I have confirmed myself that the problem manifests itself worse with a time control of "all moves in fixed time" with no increment.
Less so with repeating controls as at least the engine only has to make a certain number of moves before each control.
And less so again if you have a nice Fischer increment to hide the problem altogether

But hey, without these setbacks, we'd all be producing engines of Rybka standard ...
...or maybe not!

[mclane]

i see.that could explain why i do not have those "problems" because i tested in 40 moves in 10 minutes on the athlon machine.

i will take a closer look on the other machines, i think on the core2duo i use an increment time control blitz 25' + 10".... maybe there it could happen ....

ok. thank you for the information. will keep an eye on it and collect game and data and send it to you.