Brute Force?

[Viz]

SimpleSearch is more or less it, pure alpha-beta with SF NNUE.

Is the code available for SimpleSearch? I can't find it anywhere....

https://github.com/vondele/Stockfish/tr ... verysimple

[Brunetti]

Check out Murka 1.2 by Igor Korshunov (WildCat). Full brute force. Negamax without QS:
https://transfiles.ru/25z2c

Hi,
yes, I just checked it, the search is really basic and it plays at a maximum depth of 6 or so. I can easily beat it myself

Alex

[JVMerlino]

SimpleSearch is more or less it, pure alpha-beta with SF NNUE.

Is the code available for SimpleSearch? I can't find it anywhere....

https://github.com/vondele/Stockfish/tr ... verysimple

Thank you!

[Jjaw]

I would suggest using SOS 5.01 UCI. It has engine parameters for Brute force evaluation only and another that uses material evaluation only. It also has a setting to turn off null move. It is an Arena partner engine.

[Rowen]

If an engine uses only brute force, how would this impact on its playing style, would it 'just' be a 'dumb ' tactical calculating machine?

[hgm]

If it would not have QS, and the static evaluation would not take account of hanging pieces, it would be worse than dumb. 'Insane' would probably cover it better. Its main aim would be to keep the opponent busy with (possibly moderately sacrificial) exchanges up to the horizon, and then trade a fat piece in the conviction that it gained it.

Of course it is perfectly possible to make a tactically aware static evaluation, which can play at high level. LC0 can play nearly GM level without any search.

[Werewolf]

If it would not have QS, and the static evaluation would not take account of hanging pieces, it would be worse than dumb. 'Insane' would probably cover it better. Its main aim would be to keep the opponent busy with (possibly moderately sacrificial) exchanges up to the horizon, and then trade a fat piece in the conviction that it gained it.

Of course it is perfectly possible to make a tactically aware static evaluation, which can play at high level. LC0 can play nearly GM level without any search.

And yet...for every extra ply searched it would be slightly less "insane".

[Bo Persson]

If it would not have QS, and the static evaluation would not take account of hanging pieces, it would be worse than dumb. 'Insane' would probably cover it better. Its main aim would be to keep the opponent busy with (possibly moderately sacrificial) exchanges up to the horizon, and then trade a fat piece in the conviction that it gained it.

Of course it is perfectly possible to make a tactically aware static evaluation, which can play at high level. LC0 can play nearly GM level without any search.

And yet...for every extra ply searched it would be slightly less "insane".

But it is not like removing the QS will give you time for extra plies.

[hgm]

And yet...for every extra ply searched it would be slightly less "insane".

That remains to be proven. There is no guarantee that fixed-depth minimax converges with increasing depth. In fact there have been papers claiming the opposit. For a sufficiently erratic evaluation it was shown that extra plies did not help at all.

I suppose that if you let the depth really tend to infinity, so that all lines that are not ending in checkmate must terminate by 50-move or repetition (i.e. search the entire game tree), your result stays fixed on the game-theoretical value. But what you can do in practice with fixed depth will remain very far below the 100 ply you need even to reach the first 50-move draw.

Of course searching deeper will make it see checkmates earlier; mate scores are not affected by whether you merely do static evaluation in the leaf nodes or QS. So even with a completely random evaluation a deeper searching engine will tend to convincingly beat one that searches less deep. So to judge the quality of play in positions that are not forced mates you would have to compensate for this effect. E.g. by adjudicating the games sufficiently far before the checkmate occurred.

[Werewolf]

And yet...for every extra ply searched it would be slightly less "insane".

That remains to be proven. There is no guarantee that fixed-depth minimax converges with increasing depth. In fact there have been papers claiming the opposit. For a sufficiently erratic evaluation it was shown that extra plies did not help at all.

I suppose that if you let the depth really tend to infinity, so that all lines that are not ending in checkmate must terminate by 50-move or repetition (i.e. search the entire game tree), your result stays fixed on the game-theoretical value. But what you can do in practice with fixed depth will remain very far below the 100 ply you need even to reach the first 50-move draw.

Of course searching deeper will make it see checkmates earlier; mate scores are not affected by whether you merely do static evaluation in the leaf nodes or QS. So even with a completely random evaluation a deeper searching engine will tend to convincingly beat one that searches less deep. So to judge the quality of play in positions that are not forced mates you would have to compensate for this effect. E.g. by adjudicating the games sufficiently far before the checkmate occurred.

Can I check we're debating the same thing?
I'm talking about having an EF. It may not be great, but it's there. Say material values plus a few other things. Nothing special, but something.

I'm then claiming that two identical engines - pure brute force in terms of search with nothing else, no pruning or QS - but one doing (say) a 3 ply search and the other doing (say) a 9 ply search would typically result in a victory to the 9 ply one. Do you agree with that point?

If so, I would argue that while it still plays some terrible chess, it won't be as bad as the 3 ply search.