Again, rep-draws (and score aging)

[Fritz Grau]

1. Following the hash to reconstruct the intended PV seems to be a good heuristic to reduce the problem of repetitions after hash access; but could not the crucial hash entries already have been replaced by new entries in the meantime? Btw, I didn't understand the term "all-side".

2. I would be interested to know whether your aim is to produce a theoretically sound engine (even if it is less efficient than the approach that simply ignores the repetition-draw pollution problem), or if you want to improve your engine's gameplay. In the latter case, my guess (based on very little experience) is that
- it pays off to count a first repetition as a draw (so rep-draws are found in half the search depth)
- one can safely ignore the rep-draw problem: Unless the side which the engine believes to profit from the draw changes, there are no problems at all, and in the rare case that it does, it takes only a few plies for the erratic old hash entries to become irrelevant due to insufficient depth.

[Chan Rasjid]

Hello Fritz,

My interest in repetition pollution is simply to see if hashing can be made theoretically clean and yet not too complicated.

... (ignoring the rep-draw pollution problem is like a kind of pruning which has very little draw-backs)...

I'll think about this later. It may be related to this that Fruit and Crafty are near GM levels despite (completely ?) ignoring hash pollution. I'm not sure about Rybka as Vasik did pay some attention to repetition hash pollution in past postings.

Best Regards,
Rasjid

[Chan Rasjid]

Hello Muller,

...This would make it possible to extract the continuation from the hash table after a hash hit. ...

Dr Hyatt may laugh on reading this! Maybe there is not a single working chess program that stops after a hash hit!

On a more serious note, I remember some early brief posts by Fabien about not doing hash cutoff when on PV nodes. This at least will catch any repetition in the most likely PV.

Best Regards,
Rasjid