Peculiarity in computing RG-l

[sajastu]

Hello,

It seems that something is going wrong when I want to compute RG-L. When I pass in a list of hypotheses and references to Pythonrouge, it gives me a RG-L score as the average. While when passing each of these hypotheses and references one-by-one to the package, and taking an average of RG-L at the end, I obtain quite a different score. Not sure what's going on.