EMR & EMP

[Michael Sherwin]

Early Move Reductions & Early Move Pruning done if the confidence level falls too low. For the confidence level i'm just accumulating the history table values and dividing them by the ply. Seems good for an extra 3 to 4 ply and some elo gain. Deep in the tree pruning seems to be safe.

[Karlo Bala]

Early Move Reductions & Early Move Pruning done if the confidence level falls too low. For the confidence level i'm just accumulating the history table values and dividing them by the ply. Seems good for an extra 3 to 4 ply and some elo gain. Deep in the tree pruning seems to be safe.

What is the difference between EMR, EMP and History pruning?

[Michael Sherwin]

Early Move Reductions & Early Move Pruning done if the confidence level falls too low. For the confidence level i'm just accumulating the history table values and dividing them by the ply. Seems good for an extra 3 to 4 ply and some elo gain. Deep in the tree pruning seems to be safe.

What is the difference between EMR, EMP and History pruning?

It uses the accumulated history total for all moves by a side for the entire line and divides that by the ply. I call that a confidence value for the entire line. Therefore if the confidence value for the line drops below a certain value based on the ply the move is either reduced or pruned. I'm trying different values right now. So at really shallow ply the confidence has to be less than 1% but deeper in the tree anything from 8% to 16% is what is testing best right now. But I also have other controls like !inCheck and the position eval, etc.

[konsolas]

Early Move Reductions & Early Move Pruning done if the confidence level falls too low. For the confidence level i'm just accumulating the history table values and dividing them by the ply. Seems good for an extra 3 to 4 ply and some elo gain. Deep in the tree pruning seems to be safe.

What is the difference between EMR, EMP and History pruning?

It uses the accumulated history total for all moves by a side for the entire line and divides that by the ply. I call that a confidence value for the entire line. Therefore if the confidence value for the line drops below a certain value based on the ply the move is either reduced or pruned. I'm trying different values right now. So at really shallow ply the confidence has to be less than 1% but deeper in the tree anything from 8% to 16% is what is testing best right now. But I also have other controls like !inCheck and the position eval, etc.

Why is it called Early Move Pruning? Do you just prune early moves or the entire subtree?

[Michael Sherwin]

Early Move Reductions & Early Move Pruning done if the confidence level falls too low. For the confidence level i'm just accumulating the history table values and dividing them by the ply. Seems good for an extra 3 to 4 ply and some elo gain. Deep in the tree pruning seems to be safe.

What is the difference between EMR, EMP and History pruning?

It uses the accumulated history total for all moves by a side for the entire line and divides that by the ply. I call that a confidence value for the entire line. Therefore if the confidence value for the line drops below a certain value based on the ply the move is either reduced or pruned. I'm trying different values right now. So at really shallow ply the confidence has to be less than 1% but deeper in the tree anything from 8% to 16% is what is testing best right now. But I also have other controls like !inCheck and the position eval, etc.

Why is it called Early Move Pruning? Do you just prune early moves or the entire subtree?

If remaining depth is small and confidence in the line is also small I am experimenting with pruning that move or calling qSearch if ahead in the eval. I'm using this on moves before normal LMR and LMP. The idea that I'm trying to promote is if confidence falls too low for a side then a deeper search is not needed as undoubtedly there are better alternatives higher up in the tree. It does not have to be based on history counters. History counters were just simple to implement.