Every time I go back to work on the journal paper, I’m struck more and more about how much the reviewers seem to miss the point. Perhaps it’s the writing; more likely they were just looking for criticism and made some incorrect conjectures by doing so, but we have reviews that ask us to explain how our methodology relates to kernel learning (it doesn’t… it’s a completely separate field), what happens if there’s a trifurcation in the trees (we already stated that we treat it as two bifurcations), to clarify what we mean when we say using a breadth-first approach is more robust (we already state it pretty clearly: “Using breadth-first labeling, a missed branch will only cause changes in the encoding at or below the level at which the branch is missed, whereas a depth-first approach could potentially change the labeling at all levels of the tree”), and finally, “It is not very clear how these precision percentage values were obtained, especially when k becomes 2 to 5. Please clarify.”, even though we state many times throughout the paper that these are cross-validated k-nearest neighbor classification accuracies.

One of the reviewers even asks us to cite a paper we’ve never referenced and don’t use any of the techniques from.