Do we need to check self.trainer.sanity_checking?

[adamjstewart]

Quick question. The docs for several LightningModule instance methods mention checking self.trainer.{training|validating|testing|predicting} to apply different logic for different stages. We're applying data augmentation for each stage in on_after_batch_transfer like so:

def on_after_batch_transfer(self, batch, dataloader_idx):                                                                                                                           
    if self.trainer:                                                                 
        if self.trainer.training:                                                    
            aug = self.train_aug                                      
        elif self.trainer.validating:                                                
            aug = self.val_aug                                  
        elif self.trainer.testing:                                                   
            aug = self.test_aug                                          
        elif self.trainer.predicting:                                                
            aug = self.predict_aug                                    
        batch = aug(batch)                                                           
    return batch              

(augmentation attributes are defined in __init__)

Do we also need to check self.trainer.sanity_checking to make sure the augmentations are applied during sanity checks? Or does self.traininer.validating get set to True when sanity checking?

@ashnair1