Fair point. It silently ignored unimplemented functionally because moses2
originally implemented only a small subset.
But that subset is getting larger so hopefully there won't be too many
functionality that will cause a throw.
I'll go with your suggestion
Sent while bumping into things
On 3
Hello Nat,
for NMT ensembles, you just average the probability distribution of
different models at each time step before selecting the next hypothesis
(or hypotheses in beam search). If you're familiar with Moses, this is
similar to what happens when we combine different feature functions in
