I have a classification problem. I am trying to classify strings using
I normalize the data before training.
So far I get the best results with addition, but I cannot understand why. The difference is quite significant (2 percentage points in f-measure and accuracy).
P(abc) = p(a) * p(b) * p(c)
P(abc) = p(a) + p(b) + p(c)
P(abc) = log(p(a)) + log(p(b)) + log(p(c))
In all of the examples,
P / len(abc).
Multiplication seems to be the theoretically correct way to go, but the results are contradicting.
Can anyone venture an explanation? thanks in advance