What is the difference between Trainer.evaluate() and Trainer.predict()?

So these are the 2 tutorials I am looking at for Image Classification using ViTs:

However, the first one from Huggingface uses trainer.evaluate() to output the metrics, while AI Summer uses trainer.predict(). Is there any substantial difference between the two or are they interchangeable?

FYI, the models I am using are ‘google/vit-base-patch16-224-in21k’, ‘microsoft/cvt-13’, ‘microsoft/resnet-50’, and ‘facebook/convnext-base-224-22k’.

They can both be used, they just have slightly different outputs. evaluate is thought as “Evaluate on the validation dataset on labeled data” while predict is more of a “Predict on new unlabeled data” (but if you have labels, it will compute the metrics).

2 Likes