How Can I Save and Restore Trained TensorFlow Models?

Saving and Restoring Trained TensorFlow Models

TensorFlow provides seamless capabilities for saving and restoring trained models, allowing you to persist and reuse your models in various scenarios.

Saving the Model

To save a trained model in TensorFlow, you can use the tf.train.Saver class. Here's an example:

import tensorflow as tf

# Prepare placeholders and variables
w1 = tf.placeholder(tf.float32, name="w1")
w2 = tf.placeholder(tf.float32, name="w2")
b1 = tf.Variable(2.0, name="bias")
feed_dict = {w1: 4, w2: 8}

# Define an operation to be restored
w3 = tf.add(w1, w2)
w4 = tf.multiply(w3, b1, name="op_to_restore")
sess = tf.Session()
sess.run(tf.global_variables_initializer())

# Create a saver object
saver = tf.train.Saver()

# Run the operation and save the graph
print(sess.run(w4, feed_dict))
saver.save(sess, 'my_test_model', global_step=1000)

Restoring the Model

To restore a previously saved model, you can use the following process:

import tensorflow as tf

sess = tf.Session()

# Load the meta graph and restore weights
saver = tf.train.import_meta_graph('my_test_model-1000.meta')
saver.restore(sess, tf.train.latest_checkpoint('./'))

# Access saved variables directly
print(sess.run('bias:0'))  # Prints 2 (the bias value)

# Access and create feed-dict for new input data
graph = tf.get_default_graph()
w1 = graph.get_tensor_by_name("w1:0")
w2 = graph.get_tensor_by_name("w2:0")
feed_dict = {w1: 13.0, w2: 17.0}

# Access the desired operation
op_to_restore = graph.get_tensor_by_name("op_to_restore:0")

print(sess.run(op_to_restore, feed_dict))  # Prints 60 ((w1 + w2) * b1)

For additional scenarios and use cases, refer to the resources provided in the provided answers, which delve deeper into saving and restoring TensorFlow models.

The above is the detailed content of How Can I Save and Restore Trained TensorFlow Models?. For more information, please follow other related articles on the PHP Chinese website!