#网络结构
w1 = tf.Variable(tf.truncated_normal([in_units, h1_units], stddev=0.1))
b1 = tf.Variable(tf.zeros([h1_units]))
y = tf.nn.softmax(tf.matmul(w1, x) + b1)
tf.add_to_collection('network-output', y)

x = tf.placeholder(tf.float32, [None, in_units], name='x')
y_ = tf.placeholder(tf.float32, [None, 10], name='y_')
#损失函数与优化函数
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1]))
train_step = tf.train.AdamOptimizer(rate).minimize(cross_entropy)

saver = tf.train.Saver()
with tf.Session() as sess: 
    sess.run(init) 
    saver.save(sess,"save/model.ckpt") 
    train_step.run({x: train_x, y_: train_y})

tf.add_to_collection('network-output', y)

with tf.Session() as sess:
  saver = tf.train.import_meta_graph('./model.ckpt.meta')
  saver.restore(sess, './model.ckpt')# .data文件
  pred = tf.get_collection('network-output')[0]

  graph = tf.get_default_graph()
  x = graph.get_operation_by_name('x').outputs[0]
  y_ = graph.get_operation_by_name('y_').outputs[0]

  y = sess.run(pred, feed_dict={x: test_x, y_: test_y})

x = tf.placeholder(tf.float32, [None, in_units], name='x')
y_ = tf.placeholder(tf.float32, [None, 10], name='y_')

  x = graph.get_operation_by_name('x').outputs[0]
  y_ = graph.get_operation_by_name('y_').outputs[0]

  sess.run(pred, feed_dict={x: test_x, y_: test_y})

run(fetches, feed_dict=None, options=None, run_metadata=None)