In TensorFlow 1.x graph mode, tf.train.Saver can save the variables in the current session as a checkpoint and restore those variables in a new session. This mechanism is useful when training is interrupted, when training needs to be split into stages, or when you need to save trained model parameters and continue debugging later.
The example below uses the MNIST dataset to train a two-layer fully connected network: the first session trains for 3 epochs and saves the model; the second session reinitializes the graph variables, restores the weights from the checkpoint, and continues training for another 7 epochs.
Note: this code uses the TensorFlow 1.x API, including tf.Session(), tf.placeholder(), and tensorflow.examples.tutorials.mnist.input_data. If you run it in a TensorFlow 2.x environment, use compatibility mode, such as disabling eager execution and calling the corresponding interfaces through tf.compat.v1; the exact availability depends on your locally installed version.
'''
Save and Restore a model using TensorFlow.
This example is using the MNIST database of handwritten digits
(http://yann.lecun.com/exdb/mnist/)
Author: Aymeric Damien
Project: https://github.com/aymericdamien/TensorFlow-Examples/
'''
from __future__ import print_function
# Import MNIST data
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
import tensorflow as tf
# Parameters
learning_rate = 0.001
batch_size = 100
display_step = 1
model_path = "/tmp/model.ckpt"
# Network Parameters
n_hidden_1 = 256 # 1st layer number of features
n_hidden_2 = 256 # 2nd layer number of features
n_input = 784 # MNIST data input (img shape: 28*28)
n_classes = 10 # MNIST total classes (0-9 digits)
# tf Graph input
x = tf.placeholder("float", [None, n_input])
y = tf.placeholder("float", [None, n_classes])
# Create model
def multilayer_perceptron(x, weights, biases):
# Hidden layer with RELU activation
layer_1 = tf.add(tf.matmul(x, weights['h1']), biases['b1'])
layer_1 = tf.nn.relu(layer_1)
# Hidden layer with RELU activation
layer_2 = tf.add(tf.matmul(layer_1, weights['h2']), biases['b2'])
layer_2 = tf.nn.relu(layer_2)
# Output layer with linear activation
out_layer = tf.matmul(layer_2, weights['out']) + biases['out']
return out_layer
# Store layers weight & bias
weights = {
'h1': tf.Variable(tf.random_normal([n_input, n_hidden_1])),
'h2': tf.Variable(tf.random_normal([n_hidden_1, n_hidden_2])),
'out': tf.Variable(tf.random_normal([n_hidden_2, n_classes]))
}
biases = {
'b1': tf.Variable(tf.random_normal([n_hidden_1])),
'b2': tf.Variable(tf.random_normal([n_hidden_2])),
'out': tf.Variable(tf.random_normal([n_classes]))
}
# Construct model
pred = multilayer_perceptron(x, weights, biases)
# Define loss and optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
# Initializing the variables
init = tf.global_variables_initializer()
# 'Saver' op to save and restore all the variables
saver = tf.train.Saver()
# Running first session
print("Starting 1st session...")
with tf.Session() as sess:
# Initialize variables
sess.run(init)
# Training cycle
for epoch in range(3):
avg_cost = 0.
total_batch = int(mnist.train.num_examples/batch_size)
# Loop over all batches
for i in range(total_batch):
batch_x, batch_y = mnist.train.next_batch(batch_size)
# Run optimization op (backprop) and cost op (to get loss value)
_, c = sess.run([optimizer, cost], feed_dict={x: batch_x,
y: batch_y})
# Compute average loss
avg_cost += c / total_batch
# Display logs per epoch step
if epoch % display_step == 0:
print("Epoch:", '%04d' % (epoch+1), "cost=",
"{:.9f}".format(avg_cost))
print("First Optimization Finished!")
# Test model
correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
# Calculate accuracy
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print("Accuracy:", accuracy.eval({x: mnist.test.images, y: mnist.test.labels}))
# Save model weights to disk
save_path = saver.save(sess, model_path)
print("Model saved in file: %s" % save_path)
# Running a new session
print("Starting 2nd session...")
with tf.Session() as sess:
# Initialize variables
sess.run(init)
# Restore model weights from previously saved model
saver.restore(sess, model_path)
print("Model restored from file: %s" % save_path)
# Resume training
for epoch in range(7):
avg_cost = 0.
total_batch = int(mnist.train.num_examples / batch_size)
# Loop over all batches
for i in range(total_batch):
batch_x, batch_y = mnist.train.next_batch(batch_size)
# Run optimization op (backprop) and cost op (to get loss value)
_, c = sess.run([optimizer, cost], feed_dict={x: batch_x,
y: batch_y})
# Compute average loss
avg_cost += c / total_batch
# Display logs per epoch step
if epoch % display_step == 0:
print("Epoch:", '%04d' % (epoch + 1), "cost=",
"{:.9f}".format(avg_cost))
print("Second Optimization Finished!")
# Test model
correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
# Calculate accuracy
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print("Accuracy:", accuracy.eval(
{x: mnist.test.images, y: mnist.test.labels}))
The core workflow can be summarized in three steps:
- After the graph has been built, create
saver = tf.train.Saver(). - In the training session, call
saver.save(sess, model_path)to save the variables. - In a new session, first build the same graph, then call
saver.restore(sess, model_path)to restore the variables.
Saver saves the values of variables in the graph, not the Python training loop itself. Therefore, if you need to restore training to an exact step, you should also save additional training metadata such as the current epoch, global step, and learning-rate schedule state. A common approach is to define global_step = tf.Variable(0, trainable=False) and pass global_step=global_step into the optimizer's minimize(), so the checkpoint will also include the training step count.
After running the code, a group of checkpoint files usually appears near /tmp/model.ckpt. When restoring, you do not need to read these files manually; just point model_path to the path used when saving.
