Handle skewness

[smadha]

Classes are highly imbalanced

$ cut -d $'\t' -f3 invited_info_train.txt | grep -c 0
218428
$ cut -d $'\t' -f3 invited_info_train.txt | grep -c 1
27324

Few ideas to handle skewness -

1. Take 27324 from class 0 to balance data set
2. Create dictionary of words/character only from class 1. So words not present in class 1 will never be a feature.