Interesting ideas

[Henk]

I would like to read about chess ideas, alternative approaches that does not work. Maybe it will help to generate other ideas.

Any interesting (bad) ideas ? Show me the trash.

[hgm]

Well, just counting number of attackers and protectors, and trying to estimate how much you were going to gain compared to the current eval from that and the lowest attacker protector of each potential victim instead of doing a QS definitely did not work well in Usurpator II.

Doing a null move in QS, and then extending specific threat evasions when it fails low because of a threat, did not work for micro-Max.

Searching contact check drops in QS in Shokidoki caused a hopeless search explosion.

'Static mobility', where I assigned the contribution of each piece to the mobility score as determined at d=1 (where it did both null move and search its own move, so it would be known for both sides), and then add it to the piece values to be used in QS, did not work at all in Fairy-Max (-100 Elo or so). This could be due to mis-tuning, though; it was never proven that adding normal mobility would be beneficial to Fairy-Max.

[Henk]

1) I once tried using simulated annealing or another optimization algorithm to select the branch in the search tree that would be extended. So each sample point is a path in the tree.


2) Terminate the search earlier if one or two quiet moves are played in a row. For these positions can be evaluated and are quiet.

[stegemma]

I would like to read about chess ideas, alternative approaches that does not work. Maybe it will help to generate other ideas.

Any interesting (bad) ideas ? Show me the trash.

Just search for my posts... they are full of strange and often bad ideas

[Henk]

3) Do null moves for both players. If they show there are no threats then position is quiet. So only evaluate that position and do no search.

[cdani]

Some in a zillion that did not work:

* Penalization for defending pinned piece.

* Bonus for attacking pinned piece (was already present in a penalization for pinned piece).

* In move generation, first two square moves of pawns, then one square moves.

* Singular extension

* Do not extend captures in first alpha_beta after search_root.

* In quiescence try losing captures if are checks.

* When no hash move, try null move threat as a hash move.

* Remove penalization of bad developement in endgame.

* Extend if check at search_root.

[Rebel]

Some in a zillion that did not work:

* Penalization for defending pinned piece.

Confirmed.

* Bonus for attacking pinned piece (was already present in a penalization for pinned piece).

Confirmed.

* Singular extension

Confirmed.

* In quiescence try losing captures if are checks.

Works for me.

Can't post my trash, forum software reports, max text exceeded.

[Henk]

If there are many you can always post the most interesting ones. Maybe there were ideas that were so promising that it would at least make 100 or 200 ELO. But unfortunately it didn't.

[cdani]

Some in a zillion that did not work:

* In quiescence try losing captures if are checks.

Works for me.

Nice! Will try again. Thanks.

Can't post my trash, forum software reports, max text exceeded.

[Dann Corbit]

An alternative for quiescent search:
BMOBUNT
Best Move Only, Back Up N Times

The idea is that as a replacement for quiescent search, you create an alternative that has a branching factor of approximately 1.

Just choose the best move that your normal search would have chosen and search none of the alternatives.

Of course, you might fail low. When that happens, you decrement your N counter, back up, do the search on the next most promising alternative, and continue.

When your N counter hits zero or you reach an acceptable depth, stop.

The reason I suggest this is that the capture moves are rarely what would actually be played. It is another way to examine the horizon.

You have two parameters for the method: How many times you are allowed to back up (N) and the depth that must be achieved to alleviate worry (AWD).