Are EgivenF and CoherenceProb substituted by EgivenFCoherence in featuer extraction?

[Huiying Li]

Was curious why there is no p(e|f) in the extracted grammar file. Have this guess after read features.py. Anyone confirm?

Thanks!
Huiying

[Chris Dyer]

There should be a p(e | f), but not a p(f | e)- this is because the
way the corpus is indexed, it is hard to compute the marginal counts
of each e, but doing so for f is easy.

> --
> You received this message because you are subscribed to the Google Groups
> "cdec users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to cdec-users+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Huiying Li]

Thanks Chris.
Yeah there is a p(e|f), but the feature actually being used (imported to scorer) is named EgivenFCoherence, where the denominator is sample count (f) instead of count (f).
So I was a bit confused.