why pretrained model is not used for training in self-training stage?

[cultivater]

Hi yumeng, Thanks for your great work! But I have a question in the 'fit' function: https://github.com/yumeng5/WeSTClass/blob/master/model.py#L222

self.classfier is only used to get the 'y_pred_last' in the first iteration, for checking stop conditions, and you only use self.model for training. We know that self.classfier is the model pretrained in the pretraining stage, so why self.classfier is not used during self-training?