Once the Tail Robust Teacher has been trained we prepare a dataset that encodes the knowledge of the tail robust teacher. To create the dataset do following steps:
-
Run
create_aug_dataset.ipynb
to create a dataset that has adds each label as a data point and it's relevant label as itself. -
Run `create_teacher_distilled_dataset.py to create a dataset that uses the high-scoring data point label pairs (as per the Siamese Teacher) to augment the ground truth.
Using the new trn_X_Y train any extreme classification method.