About process_data.py

[don-tpanic]

I may have missed something but it seems in the process_data.py file you have only created input features as unigrams but the original paper uses both unigram and bigrams. Also I wonder would you be so kind to explain which part of your code reflects the idea which is stated in the paper as combine pivot features and most frequent features since I neither seem to find where the concept of SCL was implemented.

Thank you.