Question/possible bug in casing

[bunny-therapist]

I don't know enough to say what is right or wrong here, but this crate is here doing

cand.casing /= 1.0 + cand.tf.ln_1p();

but the same computation in the LIAAD implementation of yake reads:

self.WCase = max(self.tf_a, self.tf_n) / (1. + math.log(self.tf))

(WCase is what we call "casing", tf_n appears to be what we call tf_u, even though the way it is computed differs.)

In other words, we have 1+ln(1+tf) in the denominator but LIAAD has 1+ln(tf)
We differ by a +1 in the logarithm since we are using ln_1p instead of ln.

Is this intentional? Is this correct?

[bunny-therapist]

Yeah, this is definitely a bug. Fixing it brings a lot of examples closer to LIAAD/yake.

@xamgore