I wonder if an alternate precision_recall function could be passed a minimum similarity threshold instead of a number of candidates. It seems like that could result in better scores, since if a tree name only has record names that aren't very similar, including the less-similar names would hurt precision.
I wonder if an alternate precision_recall function could be passed a minimum similarity threshold instead of a number of candidates. It seems like that could result in better scores, since if a tree name only has record names that aren't very similar, including the less-similar names would hurt precision.