I have added a vocabulary and GloVe embedding with all mathematics from the data discarded/ignored.
They have been used to evaluate the specific contribution of the math lexemes to statement classification models. I'm pushing all resources out before submitting a write-up for that experiment.