TensorFlow基礎チュートリアル:モデルの永続化(モデルの保存と読み取り)

TensorFlowは訓練したモデルを保存することができ、訓練中に切断されただけでなく、前回の訓練過程を継続することができる.移行学習を行い、他人の訓練モデルに基づいて自分のモデルを訓練することもできます.とても便利です.
TensorFlowモデルcheckpointを保存した後、次のファイルを生成します.
|—checkpoint |—model_name.data-00000-of-00001 |—model_name.index |—model_name.meta
model_nameは定義されたモデル名model_name.metaは図ファイルmodel_name.dataはデータファイル

保存モデル

saver = tf.train.Saver()            #  saver  
saver.save(sess, checkpoint_path)   # sess          

リカバリモデル

データのみロード

saver.restore(sess, checkpoint_path)  #           sess

図とデータのロード

meta_path = 'model_name.meta'   #   
model_path = 'model_name'       #    
saver = tf.train.import_meta_graph(meta_path)  #   
with tf.Session() as sess:
    saver.restore(sess, model_path)     #    sess     
    graph = tf.get_default_graph()
    x = graph.get_tensor_by_name('InputData:0')  #     tensor

すべてのプログラムコード(TensorFlowベースチュートリアル:簡単なDNNを構築して手書きデジタル認識を実現)の書き換え
トレーニングコード

# coding: utf-8
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

import tensorflow as tf

learning_rate = 0.001
train_epochs = 10
batch_size = 64
checkpoint_path = 'checkpoint/'

n_input = 784
n_hidden1 = 100
n_hidden2 = 100
n_classes = 10

#name  ，      
x = tf.placeholder(tf.float32, shape=[None, n_input], name='InputData')
y = tf.placeholder(tf.float32, shape=[None, n_classes], name='LabelData')

weights = {'w1': tf.Variable(tf.random_normal([n_input, n_hidden1]), name='W1'),
                  'w2': tf.Variable(tf.random_normal([n_hidden1, n_hidden2]), name='W2'),
                  'w3': tf.Variable(tf.random_normal([n_hidden2, n_classes]), name='W3')}
biases = {'b1': tf.Variable(tf.random_normal([n_hidden1]), name='b1'),
                'b2': tf.Variable(tf.random_normal([n_hidden2]), name='b2'),
                'b3': tf.Variable(tf.random_normal([n_classes]), name='b3')}

def inference(input_x):
    layer_1 = tf.nn.relu(tf.matmul(x, weights['w1']) + biases['b1'])
    layer_2 = tf.nn.relu(tf.matmul(layer_1, weights['w2']) + biases['b2'])
    out_layer = tf.matmul(layer_2, weights['w3']) + biases['b3']
    return out_layer

#         
with tf.name_scope('Inference'):
    logits = inference(x)
with tf.name_scope('Loss'):
    loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=y))
with tf.name_scope('Optimizer'):
    optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
    train_op = optimizer.minimize(loss)
with tf.name_scope('Accuracy'):
    pre_correct = tf.equal(tf.argmax(y, 1), tf.argmax(tf.nn.softmax(logits), 1))
    accuracy = tf.reduce_mean(tf.cast(pre_correct, tf.float32), name='acc')
    print(accuracy)

init = tf.global_variables_initializer()
saver = tf.train.Saver()

with tf.Session() as sess:
    sess.run(init)
    total_batch = int(mnist.train.num_examples / batch_size)

    checkpoint = tf.train.get_checkpoint_state(checkpoint_path)  #  checkpoint  
    if checkpoint and checkpoint.model_checkpoint_path:
        saver.restore(sess, checkpoint_path+'model.ckpt')        #    
        print('continue last train!!')
    else:
        print('restart train!!')
    for epoch in range(train_epochs):
        for batch in range(total_batch):
            batch_x, batch_y = mnist.train.next_batch(batch_size)
            sess.run(train_op, feed_dict={x:batch_x, y:batch_y})   
        if (epoch+1) % 5 == 0:
            loss_, acc = sess.run([loss, accuracy], feed_dict={x:batch_x, y:batch_y})
            print("epoch {},  loss {:.4f}, acc {:.3f}".format(epoch, loss_, acc))
            saver.save(sess, checkpoint_path+'model.ckpt')    #    model.ckpt

    print("optimizer finished!")
    print("     "， checkpoint_path)

モデルリカバリとテストセット精度の計算

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
import tensorflow as tf

meta_path = 'checkpoint/model.ckpt.meta'            #   
model_path = 'checkpoint/model.ckpt'                #    
saver = tf.train.import_meta_graph(meta_path)       #   

with tf.Session() as sess:
    saver.restore(sess, model_path)                 #    
    graph = tf.get_default_graph()
    x = graph.get_tensor_by_name('InputData:0')     #    
    y = graph.get_tensor_by_name('LabelData:0')
    accuracy = graph.get_tensor_by_name('Accuracy/acc:0')

    #         
    test_acc = sess.run(accuracy, feed_dict={x:mnist.test.images, y:mnist.test.labels})
    print('test accuracy', test_acc)

githubソースダウンロードhttps://github.com/gamersover/tensorflow_basic_tutorial/tree/master/model_save_tutorial