My take on history prunning, late move reductions.

[Michael Sherwin]

deleted

[Uri Blass]

I wonder how do you get the following conclusion:

"I decided to remove LMR from Fruit and play a match between the two versions. There was virtually no differance IMHO in playing strength between them."

As far as I know fruit earns much from late move reductions.

I am also not sure if I understand your formula

(depth * 4 / iterationDepth + 5)

Do you mean that depth is the remaining depth and iterationDepth is usually bigger.

Does it mean that in typical case when you search iteration depth 10 and the remaining depth is 2 you stop to search if you already searched
5.8 moves and when the remaining depth is 6 you stop to search if you already searched 7.4 moves

In other words you are more aggresive in pruning when the remaining depth is small.

Does it mean that you practically do selective search because you never search more than 9 moves with score of less than alpha?

Uri

[Michael Sherwin]

deleted

[Uri Blass]

In responce to Uri,

*I wonder how do you get the following conclusion:
*
*"I decided to remove LMR from Fruit and play a match between the two versions. There was virtually no differance IMHO in playing strength between them."
*
*As far as I know fruit earns much from late move reductions.

I conducted a 50 game Nunn2 match between Fruit 2.1 w/LMR that I compiled and one that I compiled without.LMR. I do not remember the time controls or the exact result but the lead changed several times and the result was almost even.

I think that 50 games are not enough to get conclusions and it is also possible that late move reduction help fruit against other opponents more than it help against itself.

I think that it is interesting to know how many plies different programs get from late move reduction.

Today in most cases it only does movei less than twice faster in getting the same depth when I run it for 2 minutes on one position.

Uri