Saving and Restoring models in TensorFlow

You can save and restore the models and the variables in TensorFlow by one of the following two methods:

• A saver object created from the tf.train.Saver class
• A SavedModel format based object created from the tf.saved_model_builder.SavedModelBuilder class

Let us see both the methods in action.

Note

You can follow along with the code in the Jupyter notebook ch-11a_Saving_and_Restoring_TF_Models.

Saving and restoring all graph variables with the saver class

We proceed as follows:

1. To use the saver class, first an object of this class is created:

saver = tf.train.Saver()

2. The simplest way to save all the variables in a graph is to call the save() method with the following two parameters: the session object and the path to the file on the disk where the variables will be saved:

with tf.Session() as tfs:
   ...
   saver.save(tfs,"saved-models/model.ckpt")

3.  To restore the variables, the restore() method is called:

with tf.Session() as tfs:
    saver.restore(tfs,"saved...