what the differences between tf.train.Saver().restore() and tf.saved_model.loader

Question

I'd like to know the differences between tf.train.Saver().restore() and tf.saved_model.loader() .

As far as I know, tf.train.Saver().restore() restores the previously saved variables from the checkpoint file; and tf.saved_model.loader() loads the graph def from the pb file.

But I have no idea about when I should choose restore() or loader() ?

Answer 1

During training, restoring from checkpoints, etc, you want to use the saver. You only want to use the saved model if you're loading your exported model for inference.