Hands-On Artificial Intelligence for Beginners

Saving model checkpoints

In Tensorflow, a checkpoint is a binary file that contains the model weights and gradients that were calculated during training. Should you want to load up a model for further training, or access the model at a certain point during training, we can save and restore checkpoints with all of the training information that we need. To save a checkpoint, we can use a saver utility that is provided to us in native TensorFlow: 

save_model = os.path.join(job_dir, 'saved_mnist.ckpt')saver = tf.train.Saver()

Then, during the training cycle, we can periodically save checkpoints by calling the saver: 

saver.save(sess, save_model)

Note that you can choose to save the model checkpoints in whatever directory that you wish. A TensorFlow ...