It may make sense not to compute a single estimate for the root node of the tree, but generalize this in a way that is similar to the SSG idea: At some point during the search, every node of the node frontier F is labeled by 1 through |F|. Every node in F represents from now on a "subroot". An individual estimation is computed for every subroot.
Quote by @pierre-lebodic:
I don't know how this would be applicable for every possible method. For sampling methods (whatever probabilities we use), we would have nothing to compute an estimate with if we're at a leaf.
I should elaborate on this and show how this influences the methods.
It may make sense not to compute a single estimate for the root node of the tree, but generalize this in a way that is similar to the SSG idea: At some point during the search, every node of the node frontier F is labeled by 1 through |F|. Every node in F represents from now on a "subroot". An individual estimation is computed for every subroot.
Quote by @pierre-lebodic:
I should elaborate on this and show how this influences the methods.