Removing Impossible positions

[Alvaro]

Hi,
while examining Martin Fierz's EGTB compressor, I notice he tried to remove 'impossible positions' to improve compression, but later disabled this code.
In the code he commented that impossible positions are positions where there are no unmoves.
Have you tried this trick? Is this really safe? And what % do these impossible positions represent in the entire db?

best regards,
Alvaro

[BertTuyt]

I did not apply the trick.
Maybe even more interesting is the number of illegal positions.
These are positions which are not theoretically possible through game-play.
See below diagram of an illegal position.
But here it is even harder to prove which positions are illegal.
For 20-20 I was able to calculate and determine exactly the number of legal positions.
For 19 - 19 I was unable so far.



Bert

[Alvaro]

Thanks, that's what I meant, illegal positions, I'm not an english native speaker.
If only we could find some rules to detect this maybe there could be some gain in the final compressed size of the database.

Alvaro

[Rein Halbersma]

Hi,
while examining Martin Fierz's EGTB compressor, I notice he tried to remove 'impossible positions' to improve compression, but later disabled this code.
In the code he commented that impossible positions are positions where there are no unmoves.
Have you tried this trick? Is this really safe? And what % do these impossible positions represent in the entire db?

best regards,
Alvaro

When the positions are genuinely impossible, I think it's not dangerous to exclude "impossible" positions where no unmove is possible. E.g. take this one:

black to move

In all draughts variants (checkers, international, etc.), there is no white move that could have produced this position. However, consider this seemingly related one:


black to move

Under the international rules, this could have arisen through a capture sequence:


black to move

After 1... g3-f2 2. g1xa3 we get the previous diagram. So you can't just test for regular "unmove", but also for "uncapture" to conclude that a position is actually illegal. This seems to at least to give a lot of overhead to determine illegality (I am not saying it's incorrect, but it seems at least inefficient).