Disappointing LMR Results

[lauriet]

>>>LMR works even with randomly sorted move list.

Can you explain this ?
Surely it would not be very effective if for instance you reversed you ordering so that the best moves came last in the list ?

[Karlo Bala]

>>>LMR works even with randomly sorted move list.

Can you explain this ?
Surely it would not be very effective if for instance you reversed you ordering so that the best moves came last in the list ?

Two excellent explanations:
http://www.talkchess.com/forum/viewtopi ... 04&t=35955
http://www.talkchess.com/forum/viewtopi ... 16&t=35955

[theturk1234]

Ok, so you're saying that before anything else, try a zero-window search. Then if it fails high widen it. If that fails high, then try LMR?
It sounds like a lot of searching......

David

[jdart]

You are trying to resolve the fail-high with as few extra nodes as possible. Sometimes the zero-width full-depth search will do it. It is worth a try. Another refinement is to do the two-stage research only if the reduction was above a certain amount.

--Jon

[theturk1234]

Ok, that sounds reasonable. If I get a fail high after the null window search, couldn't I just go straight into PVS because this could be a PV node?

[Karlo Bala]

Ok, that sounds reasonable. If I get a fail high after the null window search, couldn't I just go straight into PVS because this could be a PV node?

Code: Select all

if (nodetype == PV)
{
   if (movenum == 1)
      v = search(-beta,-alpha,depth-1,...);
   else  
   {
      v = search(-(alpha+1),-alpha,depth-1-reduction,...);
      if (v > alpha)
      {
          if (reduction > 0)
             v = search(-(alpha+1),-alpha,depth-1,...);
          if (v > alpha) 
             v = search(-beta,-alpha,depth-1,...);
      }
   }
}
else // here always stands alpha == beta-1
{
   if (movenum == 1 || reduction == 0)
      v = search(-beta,-alpha,depth-1,...);
   else  
   {
      v = search(-beta,-alpha,depth-1-reduction,...);
      if (v > alpha) // v >= beta, same thing, alpha == beta-1
          v = search(-beta,-alpha,depth-1,...);
   }
}