Question: unexpecting results using MPR

[Svobikl]

Hi,
first of all than you for sharing a code, I was testing your code using Microsoft Research Paraphrase corpus that was tested in the original article as well. But I have a really poor result - accuracy only about 67% that is equal to the major voting strategy, as the majority class is in about 66% cases.

Have you tried your code on this dataset?