EBF definition problem in chess wiki

[kbhearn]

But you're not looking for the 'actual' branching factor...

T(n+1)/T(n) is what is trying to be predicted... and the sqrt formula from your one limited dataset seems to be closest to doing so across the full depth of the options discussed for this one specific position, with this one specific engine.

T(n)^(1/n) is reasonable for some portion of the graph (n >= 13 in this case). Is this specific to this position, or is it typical that crossing a threshold depth makes your 'real' factor useful in practice.

Really a lot more data points from many typical positions would be needed to show that one formula outperformed the others from a practical standpoint.

[Daniel Shawul]

But you're not looking for the 'actual' branching factor...

I am calculating EBF according to text book's definition. So that is what I am trying to predict. The ratio you are using is very much local (admitted by yourself). So there is no argument here. Lets please not extend this thread more than it needs to be. Your number of posts have already doubled

[bob]

I use EBF to decide if I should start a new iteration with an increased ply or not. I'm afraid if I use the classic EBF formula I will start an iteration which I won't be able to finish if the EBF is too small, and that I won't start on an iteration when the EBF is high even if I probably will have enough time to finish it. But you are saying in the latter case that I should allocate more time when the EBF is high because it is an important position?

Regards
Jacob

That is what Hsu talked about in Deep Blue search. I have not yet tried that myself, but have had it on my to-do list for years...