natural
natural copied to clipboard
Bayes and LogisticRegression aren't quite as interchangeable as suggested
I've been trying to experiment with the results comparing the 2 classifiers, and the docs suggest one should be a direct replacement for the other (see opening para here). But I suspect they're not quite as interchangeable as the docs suggest. Bayes can return multiple categories, whereas I guess LR can only return true or false? But if you supply true
and false
to Bayes, they get converted to text strings in getClassifications()
, so dealing with the return values becomes complex in code where switching classifiers is a configuration option.
Ideally I would like Bayes to deal with true
and false
as values rather than strings, but I'm not sure how easy it would be to allow Bayes to handle non-uniform types for the classifications. At least it should probably be pointed out on the docs page that they don't necessarily return the same data types.