data bias

[coopslarhette]

• num samples with class 0: 1261
• num samples with class 1: 1653
• num samples with class 2: ~32.2k

~91.7% of samples have a class of 2, aligns with classification report score of 0.92 for class 2.

need to augment/balance data samples size for each class to remove bias

[coopslarhette]

https://github.com/makcedward/nlpaug/blob/master/example/textual_augmenter.ipynb (see block 10)

probably want something like this