python-tensorflowHow do I save a trained model using Python and TensorFlow?

Saving a trained model using Python and TensorFlow is a simple process. The following steps outline the process:

1. Create a Saver object, specifying the variables you want to save:

saver = tf.compat.v1.train.Saver(var_list=tf.global_variables())

2. Call the save() method of the Saver object, specifying the path where you want to save the model:

saver.save(sess, './my_model.ckpt')

3. To restore the model from a saved checkpoint, use the restore() method of the Saver object:

saver.restore(sess, './my_model.ckpt')

4. To save the model as a SavedModel, use the tf.saved_model.save() method:

tf.saved_model.save(sess, './my_model')

5. To restore the model from a SavedModel, use the tf.saved_model.load() method:

tf.saved_model.load(sess, './my_model')

6. To save the model as a frozen graph, use the tf.graph_util.convert_variables_to_constants() method:

tf.graph_util.convert_variables_to_constants(
    sess, sess.graph_def, ['output_node_name'])

7. To restore the model from a frozen graph, use the tf.import_graph_def() method:

graph_def = tf.GraphDef()
with tf.gfile.GFile('frozen_model.pb', 'rb') as f:
    graph_def.ParseFromString(f.read())

tf.import_graph_def(graph_def)

Helpful links

• Saving and Restoring Models
• Freezing a Model

Edit this code on GitHub